WO2021232213A1

WO2021232213A1 - Voiceprint recognition apparatus, voiceprint registration apparatus and cross-device voiceprint recognition method

Info

Publication number: WO2021232213A1
Application number: PCT/CN2020/090930
Authority: WO
Inventors: 高振东; 吴晶; 陈晓
Original assignee: 华为技术有限公司
Priority date: 2020-05-19
Filing date: 2020-05-19
Publication date: 2021-11-25
Also published as: CN114026637A

Abstract

A voiceprint recognition apparatus, a voiceprint registration apparatus and a cross-device voiceprint recognition method. The method is applied to a system, and the system comprises a voiceprint recognition apparatus and a voiceprint registration apparatus. The method comprises: a voiceprint registration apparatus eliminating noise, which is related to the apparatus, in a first voice to obtain registered voiceprint information, wherein the registered voiceprint information provides a basis for a recognition device to perform voiceprint recognition; a voiceprint recognition apparatus eliminating noise, which is related to the apparatus, in a second voice to obtain target voiceprint information; and during voice recognition, the voiceprint recognition apparatus matching the target voiceprint information with the registered voiceprint information to obtain a matching result, wherein the matching result is used for indicating the identity of a user, and the registered voiceprint information comprises a voiceprint obtained after the influence of the apparatus on a biological voiceprint is eliminated, and can be shared with other devices for use. A user can enable a voiceprint recognition function to be realized on the voiceprint recognition apparatus without repeating voiceprint registration, thereby omitting a complicated voiceprint registration process, and realizing cross-device voiceprint recognition.

Description

Voiceprint recognition and registration device, and cross-device voiceprint recognition method

Technical field

This application relates to the technical field of voiceprint recognition, and in particular to a voiceprint recognition and registration device, and a cross-device voiceprint recognition method.

Background technique

Voiceprint recognition is a type of biological recognition technology, which is a recognition technology that uses a computer to determine the identity of a speaker by converting a voice signal into a digital signal. Currently, when the device uses the voiceprint recognition function, it is necessary to register the user's voiceprint in advance. With the continuous development of smart device technology and the increase of voiceprint devices, registering the voiceprint has become a complicated process.

In the current technology, it is possible to establish a voiceprint mapping model between different devices, extract voiceprints from the voice recorded by the first device, perform voiceprint registration, and then extract voiceprint features from the voice commands recorded by the second device. The established voiceprint mapping model maps the voiceprint features to the voiceprint registered by the first device, so that there is no need to register the voiceprint on the second device.

In the above solution, the saved voiceprint mapping model is related to the device, and pairwise mapping of all devices is required. There are too many mapping relationships. For example, when there are Q devices in an environment, a pairwise voiceprint mapping relationship model is established. The number of is Q×(Q-1); if there is a change in the channel of an individual device, the corresponding voiceprint mapping model group is all wrong, and the recognition accuracy is difficult to guarantee.

Summary of the invention

The embodiment of the application provides a voiceprint recognition device in a recognition device, a voiceprint registration device in a registration device, and a method for cross-device voiceprint recognition. The method is applied to a cross-device voiceprint recognition system, and the system includes Recognition device and registration device, where the registration device registers the user's voice; the registration device eliminates the noise related to the registration device in the recorded voice to obtain the user's biological voiceprint, and records the biological voiceprint in the user's voiceprint identification Below is the registered voiceprint information, which can be understood as that the biological voiceprint of the user has been "recorded", and the recorded biological voiceprint of the user is called the user's registered voiceprint information; the registered voiceprint information is the voice Pattern recognition provides the basis; the recognition device performs voiceprint recognition on the user’s voice. After the recognition device receives the voice input by the user, it eliminates the noise related to the recognition device in the voice to obtain the user’s biological voiceprint (called target voiceprint information) ), the voiceprint recognition function is realized through the registered voiceprint information shared by the registered device, that is, the target voiceprint information is matched with the registered voiceprint information, and the user identity is indicated through the matching result.

In the first aspect, an embodiment of the present application provides a voiceprint recognition device in a recognition device. The recognition device may be a terminal, and the voiceprint recognition device may be a processor, a chip, or a chip system in the terminal; or the voice The pattern recognition device is a terminal, which is a component of the recognition device; the device includes a processor, a voice input and output device and a transceiver connected to the processor; the voice input and output device is used to receive the voice input by the user; the processor is used to Eliminate the noise related to the recognition device in the received voice to obtain target voiceprint information, that is, the target voiceprint information is the biological voiceprint after eliminating the influence of the recognition device on the voice, and the user’s biological voiceprint is equivalent to obedient The voiceprint of the speaker that the person hears face-to-face; then, the user's registered voiceprint information is obtained through the transceiver, and the registered voiceprint information includes at least one biological voiceprint template obtained by eliminating noise related to the registered device; this processing The device is also used to match the target voiceprint information with the registered voiceprint information to obtain a matching result, and the matching result is used to indicate the identity of the user; After the noise, the biometric voiceprint information of the user in the voice is extracted. Since the influence of the registered device on the biometric voiceprint is eliminated, the registered voiceprint information can be shared with other recognition devices. The user only needs to use one device ( Registering the voiceprint on the registered device) can realize the voiceprint recognition function on multiple devices. The recognition device does not require voiceprint registration, eliminating the complicated voiceprint registration process, and realizing cross-device voiceprint recognition, which can greatly The user experience is improved, and the registered voiceprint information and the target voiceprint information are biological voiceprints after separating the device-related noise, eliminating the influence of the device itself on the voiceprint. No matter what type of terminal the identification device is, Both have a high accuracy rate.

In a possible implementation, the device further includes a transceiver connected to the processor; the transceiver is used to receive the registered voiceprint information and the voiceprint identification of the registered voiceprint information sent by the storage device or the registration device, and the voiceprint identification is used for Instruct the user; in this embodiment, the registered voiceprint information can be stored through a storage device to provide a basis for sharing the registered voiceprint information with other devices. When the user needs to use a recognition device for voiceprint recognition, there is no need to The voiceprint registration is performed on the recognition device, but the registered voiceprint information in the storage device or the registration device can be shared.

In a possible implementation, the transceiver is also used to send the user's voiceprint identification to the storage device or registration device to request the user's registered voiceprint information; the user's biological voiceprint identification corresponds to the registered voiceprint information Relationship, the user’s voiceprint identification is sent to the storage device or the registration device, and the storage device or the registration device sends the registered voiceprint information corresponding to the voiceprint identification to the voiceprint recognition device.

In a possible implementation manner, the processor is specifically configured to input the voice into the voiceprint extraction model, and eliminate the noise related to the recognition device in the voice through the voiceprint extraction model to obtain target voiceprint information; in the embodiment of the present application, By means of machine learning, the voiceprint extraction model is pre-trained, and the biological voiceprint of the user in the speech is directly extracted through the trained voiceprint extraction model, which has good robustness.

In a possible implementation, the voiceprint extraction model is obtained by learning the corpus collected by multiple devices; the multiple devices may be different types of terminals, and the different types of terminals include, but are not limited to, smart homes Terminals, lighting systems, smart speakers, robots, in-vehicle terminals, user equipment, mobile phones, tablets, personal computers, virtual reality terminal equipment, augmented reality terminal equipment, etc.; in the embodiment of this application, the voiceprint extraction model is multiple The voiceprint information output by the voiceprint extraction model is obtained by learning the corpus collected by different devices. The voiceprint information output by the voiceprint extraction model can eliminate the influence of different devices on the voice to obtain the user's biological voiceprint.

In a possible implementation manner, the processor is also used to perform frequency response compensation on the voice through a filter, and the frequency response compensation is used to eliminate noise related to the recognition device in the voice to obtain target voiceprint information; embodiments of the present application In the process, the voice signal is processed through the filter, the prominent signal is attenuated, and the weaker signal is enhanced, thereby compensating the frequency response, so as to achieve the purpose of removing equipment-related noise. The filtering method is directly The noise signal is filtered, and the output signal is the user's biological voiceprint signal, which is simple and fast to implement.

In a possible implementation manner, the processor is also used to obtain user data associated with the user identity; perform operations corresponding to the user data; in the embodiment of the present application, the processor performs operations corresponding to the historical information, but not Users are required to perform habitual operations every time to improve user experience.

In the second aspect, the embodiments of the present application provide a voiceprint registration device in a registration device. The registration device may be a terminal, and the voiceprint registration device may be a processor, a chip, or a chip system in the terminal; or a voiceprint registration device. The registration device is a terminal, which is a component of the recognition device; the device includes a processor and a voice input and output device connected to the processor; wherein the voice input and output device is used to receive the voice input by the user; the processor is used to eliminate the voice input The noise related to the registered device is used to obtain the user's registered voiceprint information. The registered voiceprint information includes the user's biological voiceprint; the registered voiceprint information is used to provide a basis for other devices to recognize the voiceprint, and the registered voiceprint information can be Registration voiceprint information shared by multiple recognition devices; in this embodiment, the registration device extracts the received user’s voice, eliminates device-related noise in the voice, and obtains the user’s "clean" biological voiceprint (the biological voice The pattern is the closest to the voiceprint of a person when speaking face-to-face). Because the device’s influence on voice is eliminated, the extracted biological voiceprint information can be used as registered voiceprint information shared by multiple devices to achieve cross-device voiceprint recognition, which can greatly Enhance the user experience.

In a possible implementation, the device further includes a transceiver connected to the processor; the transceiver is used to send the registered voiceprint information and the voiceprint identification corresponding to the registered voiceprint information to the storage device or the identification device, and the voiceprint identification is used for Instruct the user; in this embodiment of the application, the transceiver sends the registered voiceprint information and the voiceprint identification corresponding to the registered voiceprint information to the storage device or identification device, and the storage device or identification device is used as the storage of the registered voiceprint information and voiceprint identification The center provides a basis for the sharing of registered voiceprint information.

In a possible implementation, the processor is specifically configured to input the voice into the voiceprint extraction model, and eliminate the noise related to the registered device in the voice through the voiceprint extraction model to obtain registered voiceprint information; in the embodiment of the present application, By means of machine learning, the voiceprint extraction model is pre-trained, and the biological voiceprint of the user in the speech is directly extracted through the trained voiceprint extraction model, which has good robustness.

In a possible implementation manner, the voiceprint extraction model is obtained by learning the corpus collected by multiple devices; in the embodiment of the present application, the voiceprint information output by the voiceprint extraction model can eliminate the voice effects of different devices. Influence to get the user’s biological voiceprint.

In a possible implementation manner, the processor is also used to perform frequency response compensation for the voice through a filter, and the frequency response compensation is used to eliminate noise related to the recognition device in the voice to obtain registered voiceprint information; in this embodiment of the application , The speech signal is processed through the filter, the highlighted signal is attenuated, and the weaker signal is enhanced, thereby compensating the frequency response, so as to achieve the purpose of removing equipment-related noise. After the filter filtering method, it is directly filtered The noise signal, and the output signal is the user's biological voiceprint signal, which is simple and fast to implement.

In the third aspect, the embodiments of the present application provide a cross-device voiceprint recognition method, which is applied to a recognition device and may include the following steps: the recognition device receives the voice input by the user; and then eliminates the noise related to the recognition device in the voice. Obtain the user's biological voiceprint, the user's biological voiceprint is called target voiceprint information; obtain the user's registered voiceprint information, the registered voiceprint information includes at least one biological voiceprint template obtained by eliminating noise related to the registered device ; Match the target voiceprint information with the registered voiceprint information to obtain a matching result, which is used to indicate the identity of the user; in this embodiment of the application, the registered voiceprint information eliminates the noise related to the registered device itself in the voice, The extracted biological voiceprint information of the user himself in the voice, because the registered device’s influence on the biological voiceprint is eliminated, the registered voiceprint information can be shared with other recognition devices. The user only needs to register the voiceprint on the registered device. The voiceprint recognition function can be realized on multiple devices. It is not necessary to register the voiceprint on each recognition device. This saves the complicated voiceprint registration process and realizes cross-device voiceprint recognition, which can greatly improve the user experience. , And the registered voiceprint information and target voiceprint information are biological voiceprints after separating the device-related noise, which eliminates the influence of the device itself on the voiceprint. The recognition device has a higher performance regardless of the type of terminal. The accuracy rate.

In an optional implementation manner, obtaining the registered voiceprint information of the user may include: receiving the registered voiceprint information and the voiceprint identifier corresponding to the registered voiceprint information sent by the storage device or the registered device, where the voiceprint identifier is used to indicate the user.

In an optional implementation manner, before receiving the registered voiceprint information and the voiceprint identifier corresponding to the registered voiceprint information sent by the storage device or the registered device, the method further includes: sending the voiceprint identifier to the storage device or the registered device to request the user's Register voiceprint information.

In an optional implementation manner, removing noise related to the recognition device in the voice to obtain target voiceprint information may include inputting the voice into a voiceprint extraction model, and removing the noise related to the recognition device in the voice through the voiceprint extraction model Obtain target voiceprint information; in the embodiment of this application, the voiceprint extraction model is pre-trained through machine learning, and the biological voiceprint of the user in the voice is directly extracted through the trained voiceprint extraction model, which has better results Robustness.

In an optional implementation manner, the voiceprint extraction model is obtained by learning the corpus collected by multiple devices; in this embodiment of the present application, the voiceprint information output by the voiceprint extraction model can eliminate the voice effects of different devices themselves. Influence to get the user’s biological voiceprint.

In an optional implementation manner, after matching the target voiceprint information with the registered voiceprint information to obtain the matching result, the method further includes: obtaining user data associated with the user identity, and the user data may be historical data of the user's operation. Perform operations corresponding to user data; in the embodiment of the present application, the identification device obtains user data of the user and performs operations corresponding to the historical information, eliminating the need for the user to perform habitual operations each time, thereby improving user experience.

In a fourth aspect, an embodiment of the present application provides a cross-device voiceprint recognition method, which is applied to a registered device, and may include: receiving a voice input by a user; eliminating noise related to the registered device in the voice to obtain registered voiceprint information; The registered voiceprint information includes the user’s biological voiceprint; in this embodiment, the registration device extracts the received user’s voice, eliminates device-related noise in the voice, and obtains the user’s "clean" biological voiceprint (the biological voice The fingerprint is the closest to the voiceprint of a person when speaking face-to-face). Because the device’s impact on speech is reduced, the extracted biological voiceprint information can be used as registered voiceprint information shared by multiple devices to achieve cross-device voiceprint recognition, which can greatly Enhance the user experience.

In an optional implementation, the registration device sends the registered voiceprint information and the voiceprint identification corresponding to the registered voiceprint information to the storage device or the recognition device; the registration device sends the registered voiceprint information and the registered voiceprint to the storage device or the recognition device The voiceprint identification corresponding to the information uses the storage device or recognition device as the storage center for registered voiceprint information and voiceprint identification, providing a basis for the sharing of registered voiceprint information.

In an optional implementation manner, eliminating the noise related to the registered device in the voice to obtain the biological voiceprint information may include: inputting the voice into the voiceprint extraction model, and eliminating the noise related to the registered device in the voice through the voiceprint extraction model In order to obtain registered voiceprint information; in the embodiment of this application, the voiceprint extraction model is pre-trained through machine learning, and the biological voiceprint of the user in the voice is directly extracted through the trained voiceprint extraction model. Robustness.

In an optional implementation manner, eliminating the noise related to the registered device in the voice to obtain the biological voiceprint information may include: performing frequency response compensation on the voice through a filter, and the frequency response compensation is used to eliminate the voice related to the recognition device. To obtain the registered voiceprint information; in the embodiment of the application, the voice signal is processed through a filter to attenuate the highlighted signal and strengthen the weaker signal, thereby compensating for the frequency response, so as to achieve the elimination of equipment related The purpose of the noise, through the filter method, is to directly filter the noise signal, so that the output signal is the user's biological voiceprint signal, which is simple and fast to implement.

In the fifth aspect, the embodiments of the present application provide a voiceprint recognition device, which has the function of realizing the function performed by the recognition device of the third aspect mentioned above; the function can be realized by hardware, or the corresponding software can be executed by hardware Implementation; the hardware or software includes one or more modules corresponding to the above-mentioned functions, the device includes: a voice input and output module for receiving the voice input by the user; a processing module for eliminating the voice received by the voice input and output module Recognize the noise related to the device to obtain target voiceprint information, the target voiceprint information is the user’s biological voiceprint; the transceiver module is used to obtain the user’s registered voiceprint information, which includes the elimination of noise related to the registered device The obtained at least one biological voiceprint template; the processing module is also used to match the target voiceprint information obtained by the processing module with the registered voiceprint information obtained by the transceiver module 1303 to obtain a matching result, and the matching result is used to indicate the user's identity.

In the sixth aspect, the embodiments of the present application provide a voiceprint registration device. The voiceprint recognition device can realize the function performed by the registration device in the fourth aspect; the function can be realized by hardware, or the corresponding software can be executed by hardware. Implementation; the hardware or software includes one or more modules corresponding to the above-mentioned functions, the device includes: a voice input and output module for receiving the voice input by the user; a processing module for eliminating the voice received by the voice input and output module The noise associated with the registered device is used to obtain the user's registered voiceprint information.

In a seventh aspect, an embodiment of the present application provides a cross-device voiceprint recognition system, which includes a registration device and a recognition device; the registration device receives the first voice input by the user, and eliminates the recognition device-related information in the first voice. Noise is used to obtain registered voiceprint information. The registered voiceprint information includes the user’s biological voiceprint. The registered voiceprint information provides a basis for the recognition device to perform voiceprint recognition; the recognition device receives the second voice input by the user; and eliminates the second voice. The noise associated with the recognition device is used to obtain the target voiceprint information, which is the user’s biological voiceprint; the recognition device matches the target voiceprint information with the registered voiceprint information to obtain a matching result, and the matching result is used to instruct the user identity. In the embodiment of this application, the registered voiceprint information refers to the extracted biological voiceprint information of the user in the voice after eliminating the noise related to the registered device itself. Since the influence of the registered device on the biological voiceprint is eliminated, Share the registered voiceprint information with other voiceprint recognition devices. The user only needs to register the voiceprint on one device (registered device), and the voiceprint recognition function can be implemented on multiple devices, without the need for each device Voiceprint registration is performed, eliminating the complicated voiceprint registration process, and realizing cross-device voiceprint recognition, which can greatly improve the user experience, and the registered voiceprint information and target voiceprint information are both after eliminating device-related noise The voiceprint information eliminates the influence of the device on the voiceprint. No matter which voiceprint device the user uses for voiceprint recognition, it has a high accuracy rate.

In an optional implementation manner, the system further includes a storage device; the storage device receives and stores the registered voiceprint information and the voiceprint identifier corresponding to the registered voiceprint information sent by the registration device, the voiceprint identifier is used to indicate the user; The device receives the registered voiceprint information and the voiceprint information corresponding to the registered voiceprint information sent by the storage device. In this embodiment of the application, the storage device stores the registered voiceprint information to provide a basis for sharing the registered voiceprint information with other devices. When the user needs to use other devices (also called voiceprint recognition devices) for voiceprint recognition When you do not need to register the voiceprint on the voiceprint recognition device, you can share the registered voiceprint information that has been registered on the registered device.

In an eighth aspect, this embodiment provides a chip including a processor and a memory. The memory is used to store a program or instruction. When the program or instruction is executed by the processor, the identification device can execute the method of any one of the foregoing third aspect. Alternatively, the registration device is caused to execute the method of any one of the foregoing fourth aspects.

In the ninth aspect, this embodiment provides a computer-readable medium for storing a computer program or instruction. When the computer program or instruction is executed, the computer executes any of the methods of the third aspect, or the computer executes the fourth Any one of the methods.

Description of the drawings

FIG. 1 is a schematic diagram of an embodiment of a communication system in an embodiment of this application;

2 is a schematic flowchart of an embodiment of cross-device voiceprint recognition in an embodiment of this application;

FIG. 3 is a schematic diagram of a process flow of an embodiment of voiceprint registration of a registered device in an embodiment of the application;

4 is a schematic diagram of an embodiment of storing registered voiceprint information in an embodiment of the application;

5 is a schematic diagram of another embodiment of storing registered voiceprint information in an embodiment of the application;

6 is a schematic diagram of another embodiment of storing registered voiceprint information in an embodiment of the application;

FIG. 7 is a schematic diagram of a process flow of an embodiment of voiceprint recognition in an embodiment of this application;

FIG. 8 is a schematic diagram of an application scenario of cross-device voiceprint recognition in an embodiment of this application;

FIG. 9 is a schematic diagram of another application scenario of cross-device voiceprint recognition in an embodiment of this application;

FIG. 10 is a schematic diagram of training a voiceprint extraction model in an embodiment of the application;

FIG. 11A is a schematic diagram of a curve without frequency response compensation in an embodiment of the application;

FIG. 11B is a schematic diagram of a curve after frequency response compensation in an embodiment of the application;

FIG. 12 is a schematic diagram of generating at least one voiceprint information template in an embodiment of this application;

FIG. 13 is a schematic diagram of an embodiment of a device in an embodiment of the application;

FIG. 14 is a schematic diagram of another embodiment of a device in an embodiment of this application.

Detailed ways

The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application. The terms "first", "second", "third", "fourth", etc. (if any) in the description and claims of this application and the above-mentioned drawings are used to distinguish similar objects, without having to use To describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances so that the embodiments described herein can be implemented in a sequence other than the content illustrated or described herein. In addition, the terms "including" and "having" and any variations of them are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those clearly listed. Those steps or units may include other steps or units that are not clearly listed or are inherent to these processes, methods, products, or equipment.

The embodiment of the present application provides a cross-device voiceprint recognition method. The cross-device voiceprint recognition method refers to the completion of voiceprint registration on one device, and the registered voiceprint can be used on multiple unregistered devices. Information to complete the corresponding task. Multiple devices may include the same type of equipment or different types of equipment.

This method is applied to a communication system. For an example of the communication system architecture, please refer to FIG. 1. The communication system includes a server 101 and a plurality of terminals 102. The server 101 may be a server, a server cluster, or a cloud server; the terminal 102 may be a terminal in a smart home, including but not limited to smart home appliances (such as smart screens). , Televisions, smart washing machines, smart air conditioners, etc.), lighting systems, smart speakers, robots, etc.; the terminal 102 can also be a vehicle-mounted terminal, user equipment (UE), mobile phone, tablet computer (pad), Personal computers, virtual reality (VR) terminal devices, augmented reality (AR) terminal devices; as an example and not a limitation, in this application, the terminal may also be a wearable device. Wearable devices can also be called wearable smart devices. It is a general term for wearable devices that use wearable technology to intelligently design everyday wear, such as glasses, gloves, watches, clothing and shoes. A wearable device is a portable device that is directly worn on the body or integrated into the user's clothes or accessories. Wearable devices are not only a kind of hardware device, but also realize powerful functions through software support, data interaction, and cloud interaction. In a broad sense, wearable smart devices include full-featured, large-sized, complete or partial functions that can be achieved without relying on smart phones, such as smart watches or smart glasses, and only focus on a certain type of application function, and need to cooperate with other devices such as smart phones. Use, such as all kinds of smart bracelets and smart jewelry for physical sign monitoring.

In this application, the terminal 102 may be a terminal in the Internet of Things (IoT) system. IoT is an important part of the development of information technology in the future. Machine interconnection, an intelligent network of interconnection of things. In order to better illustrate the embodiments of the present application, first, the words involved in the present application will be described.

Device-related noise: The device-related noise is added during the process from the device receiving the voice to the voice signal processing. The device-related noise includes but not limited to channel noise, coding noise, microphone physical characteristics and quantity, and gain Related noise caused by, distance, environment, pre-processing algorithms, etc. It is understandable that if the user's voice is entered through the registered device, the voiceprint information in the voice will be affected by the device, and thus change, that is, it is no longer a "clean" voiceprint, but a device-related voiceprint is added. Voiceprint after the noise. If the channels of different devices are different, the noise may vary. For example, because the signal will undergo signal attenuation or signal delay during transmission, the sound signal will change during transmission. The sound signal is composed of signals of different frequencies. Therefore, in the sound transmission process, if the attenuation or delay of the signals of each frequency is not uniform, the received sound signal will be distorted. Will cause channel distortion. Or during signal transmission, it is necessary to convert analog signals into digital signals for transmission, and distortion may also be introduced. In the process of transmission, due to the limited bandwidth of the transmission channel, during encoding, the higher the bit rate, the more high-quality signals can be transmitted. However, when the signal is coded and decoded, it cannot be completely lossless. Therefore, the encoded and decoded speech signal will have a certain loss more or less. Different devices may have different effects on the sound signal.

Voiceprint ID: This voiceprint ID is used to distinguish registered voiceprint information. Each registered voiceprint information has a unique voiceprint ID. The voiceprint ID in this application can be a universal account independently owned by each user, such as this The voiceprint identification can be a user ID (identity), which can be used for a general account for performing device-related operations. If the device is a mobile phone, the general account can use complete services, such as downloading software, data synchronization, and mobile phone positioning. And other services. If the device is a TV, it can perform user-related data recommendation (favorite programs) and so on. Or the voiceprint identification may also be a dedicated account or identification for the voiceprint information identification function. One voiceprint identifier may correspond to at least one biological voiceprint template of the same user, and the voiceprint identifier is used to indicate the user. The voiceprint identification in the embodiment of the present application may be described by taking a user ID as an example.

Registered voiceprint information: information related to the user’s biological voiceprint after eliminating the noise related to the registered device. The registered voiceprint information includes at least one biological voiceprint template (that is, a preset biological voiceprint for matching or recognition). Pattern).

User's biological voiceprint: The user's own voiceprint, which has nothing to do with the device that records the voice, is equivalent to the voiceprint of the speaker that the listener hears face-to-face.

In this application, the devices included in the communication system corresponding to FIG. 1 are classified according to their functions, including: registered devices, storage devices, and voiceprint recognition devices. The registration device is used to register the user's voice. The registration device extracts registered voiceprint information from the voice, and the registered voiceprint information includes the user's own biological voiceprint obtained after separating the noise related to the registered device. (Ie, registered voiceprint information); the storage device (also called voiceprint sharing device) is used to store the user's registered voiceprint information and provide a registered voiceprint information sharing service. The voiceprint recognition device is a device that performs voiceprint recognition on the user's voice (may be a device that has not registered the user's voiceprint), but the voiceprint recognition function can be realized through the shared registered voiceprint information.

The registration device (or called a voiceprint registration device) may be any terminal among multiple terminals in the communication system. For example, the terminal may be a mobile phone, a tablet computer, or the like.

The storage device may be a server, a cloud server, or any terminal among multiple terminals, for example, it may be at least one terminal among multiple terminals, or it may also be each terminal among multiple terminals.

The voiceprint recognition device (or called the recognition device, or called the voiceprint recognition device) may be a terminal that has not registered the voiceprint among multiple terminals. For example, if the registered terminal is a mobile phone or a tablet computer, the voiceprint recognition device may be a wearable device, a vehicle-mounted terminal, a smart screen, a headset, a personal computer, etc.

In this application, the voiceprint recognition device is also referred to as the "first terminal", and the registered device is also referred to as the "second terminal".

It should be noted that the above-mentioned registration device, storage device and voiceprint recognition device are only for the convenience of explaining the division of devices from the functional level of each device, and it is not limited that each device can only perform one function. For example, a terminal can be either a registered device, a storage device, or a voiceprint recognition device. For example, a mobile phone can be used to register the user’s voiceprint, and it can also be used to store registered voiceprint information. For the recognition of the user's voiceprint; a terminal can be either a storage device or a voiceprint recognition device. For example, a personal computer can be used to store registered voiceprint information as well as a voiceprint recognition device.

In the embodiment of the present application, the registration device receives the voice input by the user, separates the noise related to the registration device itself in the voice, extracts the user's biological voiceprint information in the voice, and registers the voiceprint information to obtain the registered voiceprint information, Since the registered voiceprint information is the voiceprint information after the device-related noise is eliminated, and the device’s influence on the user’s biological voiceprint is eliminated, the registered voiceprint information can be shared with other devices, and the storage device stores the registered voice. The pattern information provides a basis for sharing the registered voiceprint information with other devices. When the user needs to use other devices (also known as voiceprint recognition devices) for voiceprint recognition, there is no need to use the voiceprint recognition device Voiceprint registration is performed, but the registered voiceprint information registered on the registered device can be shared. The voiceprint recognition device receives the voice input by the user, separates the noise in the voice related to the voiceprint recognition device itself, and extracts the The user’s biological voiceprint information (also called target voiceprint information) in the voice, the voiceprint recognition device obtains the registered voiceprint information from the storage device, and matches the registered voiceprint information with the target voiceprint information to perform voice Pattern recognition to identify the user's identity. In the embodiment of this application, the registered voiceprint information refers to the extracted biological voiceprint information of the user in the voice after eliminating the noise related to the registered device itself. Since the influence of the registered device on the biological voiceprint is eliminated, Share the registered voiceprint information with other voiceprint recognition devices. The user only needs to register the voiceprint on one device (registered device), and the voiceprint recognition function can be implemented on multiple devices, without the need for each device Voiceprint registration is performed, eliminating the complicated voiceprint registration process and realizing cross-device voiceprint recognition, which can greatly improve user experience, and both registered voiceprint information and target voiceprint information are separated from device-related noise The voiceprint information eliminates the influence of the device on the voiceprint. No matter which voiceprint device the user uses for voiceprint recognition, it has a high accuracy rate.

This application mainly includes three stages, 1) the collection stage of registered voiceprint information; 2) the storage stage of registered voiceprint information, and 3) the speaker's voiceprint recognition stage. Please refer to Figure 2. In the collection stage of registered voiceprint information: Eliminate the device-related noise in the voice to obtain the user's biometric voiceprint, and record the biometric voiceprint under the existing user ID. The biological voiceprint is registered to obtain the registered voiceprint information; in the registered voiceprint information storage stage: the registered voiceprint information is stored in association with the user ID to provide a basis for the sharing of registered voiceprint information; in the speaker recognition stage, the registered voice The pattern information is matched with the target voiceprint information to obtain a matching result, which is used to indicate the identity of the user.

First, for 1) the registration voiceprint information collection stage, the execution subject of this stage is the registration device, or the execution subject of this stage is the processor, chip or chip system in the registration device. Please refer to Figure 3, the execution body of this stage takes the registered device as an example, and the registered device can perform the following steps: Step 301: The registered device receives the voice entered by the user. The voice does not pay attention to the text content of the voice itself, and is not limited to the text content of the voice, and is mainly used to extract voiceprint information. The processing may specifically collect voice through a voice input and output device. For details, refer to the specific introduction of FIG. 14, which is not limited here. Step 302: The registration device eliminates noise related to the registration device in the voice to obtain the user's biological voiceprint. The specific processing scheme can refer to the specific introduction of the subsequent noise elimination scheme, which will not be repeated here.

Step 303: The registration device registers the user's biological voiceprint to obtain the user's registered voiceprint information. "Registration" refers to the operation of recording the user's biological voiceprint under an existing voiceprint logo, or it can be understood as the operation of establishing the correspondence between the user's biological voiceprint and the voiceprint logo, or it can be understood In order to configure the user's voiceprint identification operation for the user's biological voiceprint. The voiceprint identifier may be a general account for performing related operations of the device. For example, the registered device uses a mobile phone as an example. After a user purchases a mobile phone, he will register a user ID, and the user ID can use services such as downloading software and mobile phone location. The voiceprint identification may also be a dedicated account for the voiceprint recognition function, for example, the user's mobile phone number may be used as the dedicated account. In order to distinguish the voiceprint information that has been "recorded" from the voiceprint information that has not been "recorded", the voiceprint information that has been "recorded" or the biological voiceprint that has been configured with voiceprint identification is called "registered voiceprint" information". The registration device establishes the correspondence between the user's biological voiceprint and the user ID, and completes the registration process of the user's biological voiceprint. The registered voiceprint information includes the biological voiceprint of the user in step 302.

The registered device can be used as a storage device to store the registered voiceprint information. Optionally, the registration device may also send the registered voiceprint information and the corresponding user ID to other storage devices, and the storage device stores the registered voiceprint information. The storage device may be a cloud server, a server or other terminals.

In this application, in the collection stage of registered voiceprint information, the received user’s voice is extracted to eliminate device-related noise in the voice and obtain the user’s "clean" biological voiceprint (when the biological voiceprint speaks face-to-face with a person). The voiceprint of is the closest), because the device’s impact on the voice is reduced, the extracted biological voiceprint information can be used as registered voiceprint information shared by multiple devices.

Then, for the stage of 2) registered voiceprint information storage (or registered voiceprint information sharing), the execution subject of storing the registered voiceprint information may be a storage device, or may also be a memory in the storage device. For example, if the storage device is a server, the execution subject that stores the registered voiceprint information may be the server, or may also be a memory in the server. The storage device receives the registered voiceprint information and the corresponding user ID sent by the registration device, and the storage device stores the registered voiceprint information and the corresponding user ID. Optionally, the registration device may also send information such as the registration time of the registered voiceprint information, the identification of the registered device, and the voice intensity received by the registered device to the storage device, and the storage device stores the foregoing information in association with the user ID. The registration time is used to record the time when the voiceprint information is registered, and can be used to prompt the user to update regularly according to the registration time. The identification of the registered device is used for the storage device to identify the registered device. When the storage device receives the user ID, it can be identified whether it is sent by the registered device or by an unregistered device (voiceprint recognition device). The voice intensity information can be used to indicate the distance between the speaker and the registered device when the registered voiceprint information is collected. The user ID can correspond to multiple biological voiceprint templates of the same user, and multiple biological voiceprint templates can correspond to the same or different ones. Voice strength information. Each time the registration device performs collection, noise elimination, and registration, one biological voiceprint template can be obtained through a process of performing step 301 to step 303; the registration device can obtain multiple biological voiceprint templates by performing the above process multiple times. When multiple biological voiceprint templates correspond to different voice intensity information, multiple biological voiceprint templates of the same user can be covered in multiple situations (the voiceprint is recorded in different distances), and the accuracy of voiceprint recognition can be improved.

Optionally, in an example, referring to FIG. 4, the registration device sends the registered voiceprint information and user ID to the cloud (or server), and the cloud (or server) stores the registered voiceprint information and user ID in association. In this example, the user's registered voiceprint information is stored in the cloud (or server), and the registered voiceprint information is distinguished by the user ID, which can save the storage space of the terminal and can be applied to a wider range of application scenarios, not only indoor home application scenarios , It can also include outdoor scenes, such as the application scene of cellular vehicle network, as long as the terminals that can connect to the network can share the registered voiceprint information.

Optionally, in the second example, referring to Figure 5, the registered device sends the registered voiceprint information and user ID to the cloud (or server), and any of the multiple terminals (such as terminal 2, can also be (Referred to as the third terminal) can download the registered voiceprint information to the local according to its own storage resources, the registered voiceprint information can be stored in a third terminal (such as a tablet), and the third terminal can be in a local area network For example, in a home scene, the terminals included in the home scene include mobile phones, smart screens, smart lamps, and tablets. The multiple terminals can use wireless fidelity (WiFi), Bluetooth, infrared, and network cards. When the voiceprint recognition device needs to perform voiceprint recognition, the voiceprint recognition device can quickly obtain the registered voiceprint information from the third terminal.

Optionally, in the third example, as shown in Figure 6, the registered voiceprint information can be distributed and stored in each of multiple terminals (such as smart screens, personal computers, vehicle-mounted terminals, and tablet computers, etc.) . If each terminal sends a user ID to the cloud, each terminal can receive the registered voiceprint information corresponding to the user ID from the cloud (or server). Alternatively, the voiceprint data can also be shared between various terminals through Bluetooth, WiFi, infrared, network card, etc. The voiceprints of the voiceprint devices can be synchronized and shared regularly or non-periodically between various terminals. Users can also configure not to update the voiceprint and not share the voiceprint on each device. If the terminals cannot share the registered voiceprint data through the above-mentioned communication methods, the registered voiceprint information can also be shared when the terminals are connected to other devices.

Each terminal has stored the registered voiceprint information corresponding to the user ID. At this time, if voiceprint recognition needs to be performed, the voiceprint recognition device is also a storage device. The voiceprint recognition device can obtain the registered voiceprint information from the local. The registered voiceprint information needs to be obtained from the cloud or from other devices, that is, the registered voiceprint information is shared, and voice recognition can be performed quickly.

Finally, for 3) the speaker’s voiceprint recognition stage, the main body of this stage is the voiceprint recognition device (also called the first terminal), or it can also be the processor, chip or chip system in the first terminal. Take the first terminal as an example. The first terminal receives the voice input by the user, separates the noise related to the first terminal in the voice, and extracts the user’s biological voiceprint information (also called target voiceprint information), and the first terminal Obtain the registered voiceprint information from the storage device, and match the registered voiceprint information with the target voiceprint information to perform cross-device voiceprint recognition. If the registered voiceprint information matches the target voiceprint information, the voiceprint recognition is successful. If the voiceprint information does not match the target voiceprint information, the voiceprint recognition fails.

Referring to Figure 7, in the voiceprint recognition stage of the speaker, the execution subject takes the first terminal as an example for description. The first terminal (such as a smart speaker) can perform the following steps: step 701, the first terminal receives the user input Voice, the process can be specifically realized by voice collection of voice input and output devices, which will not be expanded here. Step 702: The first terminal eliminates noise related to the recognition device in the voice to obtain target voiceprint information, where the target voiceprint information is the biological voiceprint of the user. For details of this process, refer to the subsequent description. Step 703: The first terminal obtains the registered voiceprint information of the user. The registered voiceprint information can be obtained from a registered device or other equipment, or obtained from a registered device or other device in advance and stored in the memory of the first terminal, and the registered voiceprint information is read from the memory when it is used . For the specific process of obtaining this information from the registered device or other devices, please refer to the following description.

In a possible implementation manner, the registered voiceprint information and the ID corresponding to the registered voiceprint information are stored by a storage device or a registration device. The storage device may be a cloud server, a server, or the storage device may be a terminal, or , The storage device can also be multiple terminals.

The voiceprint recognition device sends a user ID to the storage device or the registration device to request the user's registered voiceprint information; the voiceprint recognition device receives the registered voiceprint information and the corresponding user ID sent by the storage device or the registration device.

In another possible implementation manner, if the storage device is a terminal or multiple terminals in a communication system, the storage device can share (or synchronize) the registered voiceprint information and the corresponding user ID with other terminals. For example, it can be synchronized regularly, or multiple terminals can be synchronized when connected to the same LAN. For example, if the storage device is a tablet computer, the communication system includes 3 terminals, a tablet computer, a TV, and a mobile phone. When the 3 terminals are connected in communication (such as via WiFi), the tablet computer will store the registered voiceprint information and the corresponding ID Transmitted to mobile phones and TVs, that is, the voiceprint recognition device receives the registered voiceprint information and the corresponding ID sent by the storage device. In this implementation, the voiceprint recognition device does not need to send a user ID to the storage device to request the registration of voiceprint information.

In another possible implementation manner, the first terminal sends the user ID and the identification of the first terminal to the cloud server, and the cloud server confirms the storage location of the registered voiceprint information corresponding to the user ID according to the user ID (that is, in which device it is stored) In), the cloud server confirms that the registered voiceprint information is stored in the first device according to the user ID, the cloud server sends the user ID and the identification of the first terminal to the first device, and the first device is in communication connection with the first terminal. The first device sends the registered voiceprint information to the first terminal according to the identifier of the first terminal, and the first terminal receives the registered voiceprint information sent by the first device.

Step 704: The first terminal matches the target voiceprint information with the registered voiceprint information to obtain a matching result, and the matching result is used to indicate the identity of the user. If the target voiceprint information matches the registered voiceprint information, it is determined that the user and the user corresponding to the registered voiceprint information are the same user, that is, the user identity is true (or the preset user), if the target voiceprint information matches the registered voiceprint information If the voiceprint information does not match, it is determined that the user and the user corresponding to the registered voiceprint information are different users, that is, the user identity is false. Optionally, if the user identity is a preset user, the first terminal executes the voice control instruction; if the user identity is false (or not a preset user), it is determined that the user corresponds to the registered voiceprint information The users of are different users, and the first terminal does not need to execute the voice control instruction.

Regarding the above three stages, an application scenario is used as an example. Please refer to Figure 8. The terminals that the user S usually uses include mobile phones, tablet computers, computers, smart screens, and vehicle-mounted terminals. User S can register his voiceprint through a registered device (such as a mobile phone). The mobile phone receives the voice input by user S, such as "Hello, Xiaoyi". The input voice does not focus on the text itself, and the mobile phone extracts the The biological voiceprint information in the voice, the biological voiceprint information is the voiceprint information after the noise related to the mobile phone is separated, the biological voiceprint information is associated with the user ID of the user S, and the mobile phone associates the user ID with the user ID. The biological voiceprint information is sent to the cloud, which serves as a storage center for registered voiceprint information. When user S wants to control the vehicle-mounted terminal by voice, log in the user ID on the vehicle-mounted terminal (you can log in in advance, you do not need to log in every time), the vehicle-mounted terminal sends the user ID to the cloud through the cellular network, and the cloud will communicate with the user ID The registered voiceprint information corresponding to the user ID is sent to the vehicle terminal, and the vehicle terminal receives the voice input by user S. If the voice is "play music", the vehicle terminal receives the voice input by user S and separates the voice from the vehicle. For terminal-related noise, extract the target voiceprint information in the voice. The vehicle-mounted terminal will match the registered voiceprint information received from the cloud with the target voiceprint information. When the registered voiceprint information matches the target voiceprint information, the vehicle-mounted The terminal confirms that the user identity of the user S is the preset user, and executes the voice command of "play music".

It is understandable that the classification according to the purpose of voiceprint recognition technology can be divided into "speaker confirmation" and "speaker identification". "Speaker confirmation" refers to judging whether the subject is a designated person, and "speaker identification" refers to identifying which of the recorded speakers is the subject.

The application scenario corresponding to Figure 8 above is the "speaker confirmation" application scenario, which is to extract the target voiceprint information in the voice received by the vehicle terminal, and determine the registration corresponding to the target voiceprint information and the user ID Whether the voiceprint information is the voiceprint of the same person, after determining the identity of the user S, the vehicle-mounted terminal can execute the voice command of the user S. If the target voiceprint information and the registered voiceprint information do not match, the vehicle-mounted terminal does not need to execute the voice command of the user S. The application scenarios corresponding to FIG. 8 are only examples. The application scenarios of this application include but are not limited to account login (such as bank account login), identity verification (such as security door voice recognition, identity recognition in financial securities transactions) and other application scenarios.

This application can also be applied to the application scenario of "speaker identification". Optionally, each registered voiceprint information corresponds to a user ID. For example, in a family scenario, each user ID corresponds to user data. The user data can be different information for different application scenarios. For example, the user data can be The user's favorite program, the temperature of the air conditioner, etc. The user data takes the temperature of an air conditioner as an example, and the corresponding relationship between user ID and user data can be shown in Table 1 below:

Table 1

用户IDUser ID	用户user	空调温度Air conditioning temperature
1A1A	ff	25摄氏度25 degrees Celsius
2D2D	gg	20摄氏度20 degrees Celsius
3C3C	cc	27摄氏度27 degrees Celsius

Match the target voiceprint information with the multiple registered voiceprint information corresponding to the voiceprint identifier, determine the target registered voiceprint information that matches the target voiceprint information; obtain the target registered voiceprint information correspondence The user data associated with the user ID; then perform the operation corresponding to the user data.

For example, please refer to Figure 9. In a "speaker recognition" application scenario, a family includes 3 family members, such as user f (such as father), user g (such as mother) and user c (such as child). Take a tablet as a registered device as an example. 3 means that all family members can register voiceprints through the tablet. The tablet can register a preset number of voiceprint information. The preset number can be set by the user, or it can be System settings of the tablet. In this application scenario, a terminal can register the voiceprint information of 3 users as an example. User f logs in to the user ID (or pre-login), and the tablet computer receives the voice input for registration by user f. The voice can be any The text content of the voice (such as hello, Xiaoyi), the text content of the voice is not limited, the tablet computer separates the noise related to the tablet computer in the voice, and extracts the biological voiceprint information of the user f in the voice. The biological voiceprint information is associated with the user ID (such as "1A"), and the registration is performed to obtain the first registered voiceprint information of the user f. In the same way, the tablet computer receives the voice for registration input by user g, and obtains the second registered voiceprint information of user g. The second registered voiceprint information corresponds to the user ID "such as 2D", and the tablet computer receives the input from user c The voice used for registration, the third registered voiceprint information of user c is obtained, and the user ID corresponding to the third registered voiceprint information "such as 3C", the tablet computer sends the user ID corresponding to each registered voiceprint information to the cloud , Stored by the cloud.

If the user needs to use a smart speaker and wants to control the air conditioner through the smart speaker (the smart speaker has been associated with the user ID in the tablet smart home application), the smart speaker receives the voice "Xiaoyi, turn on the air conditioner" input by the user f , The smart speaker separates the noise related to the smart speaker in the voice, extracts the user’s biological voiceprint information (target voiceprint information) in the voice, and the smart speaker sends the user ID to the cloud, and the cloud uses the three user IDs And the corresponding 3 registered voiceprint information are sent to the smart speaker, or the smart speaker has pre-stored the 3 registered voiceprint information and the corresponding user ID, and the smart speaker combines the target voiceprint information with the received 3 registrations The voiceprint information is matched. If the target voiceprint information matches the first registered voiceprint information of the three registered voiceprint information (such as the registered voiceprint information of the user f), the voice command is executed. Optionally, the smart speaker determines the user ID (such as "1A") corresponding to the first registered voiceprint information, and the smart speaker can determine the relevant user data corresponding to the voiceprint identifier. For example, the relevant user data is "temperature 25°C". ". It is understandable that the smart speaker recognizes the user ID (such as "1A") corresponding to the first registered voiceprint information, indicating that the voice is the voice input by user f, and the historical user data recorded by the smart speaker can be: user ID (1A) corresponds to "temperature 25°C", which means that user f often adjusts the temperature of the air conditioner to 25°C, and the smart speaker sends control commands to the air conditioner according to the relevant information to adjust the temperature of the air conditioner to 25°C.

Use voiceprint recognition technology to identify specific users and selectively execute commands. At the same time, by identifying different users, smart devices can also provide personalized services for different users and expand the applications of smart devices.

The above application scenario is a home application scenario. This application can also be applied to a work scenario. The registered voiceprint information of multiple members of a working group corresponds to multiple IDs. The voiceprint recognition technology in this application can be used to identify the speaker as a working group For which user (that is, to identify the speaker), different users have different permissions. If the speaker is identified as the user d corresponding to the user ID, the voiceprint recognition device directly executes the user data of the user d's permissions.

In this example, the “speaker identification” can be performed through the voiceprint recognition, and the speaker identification is performed to determine the user data corresponding to the speaker. The user data includes, but is not limited to, those associated with the “speaker” Historical information, or user data corresponding to the "speaker" permission. The device can be intelligently customized or intelligently recommended based on user data. For example, in an application scenario, the smart speaker can perform operations corresponding to the historical information based on the historical information (such as the temperature 25°C) associated with the "speaker" (the user corresponding to the user ID "1A"), without requiring the user Perform habitual operations every time to improve user experience.

Optionally, in step 702 of the implementation manner corresponding to FIG. 7, the specific manner in which the first terminal extracts target voiceprint information in the voice that is not related to the first terminal, and in step 302 of the implementation manner corresponding to FIG. 3, The specific method for the registration device to extract the biological voiceprint information in the voice that is not related to the device may include: 1. The method of machine learning; 2. The method of signal processing.

1. The way of machine learning.

The first terminal (or voiceprint recognition device) uses the voice as the input of a voiceprint extraction model, and outputs the target voiceprint information through the voiceprint extraction model, and the voiceprint extraction model is collected from multiple devices The corpus is learned.

The voiceprint extraction model includes, but is not limited to, Gaussian mixture model (GMM), Gaussian mixture model-universal background model (GMM-UBM), i-vector, x-vector, dnn-ivector, Algorithms such as deep neural network (DNN), speech analysis, speech factorization, clustering, transformation, etc.

Machine learning includes: the training phase of the voiceprint extraction model and the application phase of the voiceprint extraction model.

As shown in FIG. 10, in the training phase of the voiceprint extraction model, a large amount of corpus for learning and reference data for reference are obtained. The large amount of corpus includes corpus collected by different types of equipment. For example, the different types of equipment include but are not limited to smart home appliances (such as smart screens, TVs, smart washing machines, smart air conditioners, etc.), lighting systems, smart speakers, robots, etc.; the terminal can also be a vehicle-mounted terminal, user equipment, mobile phone , Tablet computers, personal computers, virtual reality terminal devices, augmented reality terminal devices, wearable devices and other devices with voice recording capabilities. The reference data is voiceprint information that has nothing to do with the device (or has a weak correlation).

Receive a large amount of corpus input by the same user through different types of equipment, and input a large amount of corpus into the voiceprint model (such as the GMM-UBM model). The voiceprint data output by the GMM-UBM model is compared with the reference data. The reference data It is the user's biological voiceprint data. If the difference between the output voiceprint data and the reference data is greater than or equal to the threshold, the output voiceprint data is re-input to the GMM-UBM model, the voiceprint data is output through the GMM-UBM model, and then the The output voiceprint data is compared with the reference data. Through continuous iterative training, if the difference between the voiceprint data output from the GMM-UBM model and the reference data is less than the threshold, it indicates that the voiceprint data output from the model can be close The reference data is obtained, and the voiceprint extraction model is obtained. The voiceprint extraction model is used to separate device-related noise and extract the user's biological voiceprint. In this example, the voiceprint extraction model is pre-trained through machine learning, and the biological voiceprint of the user in the speech is directly extracted through the trained voiceprint extraction model, which has good robustness.

2. The way of signal processing.

In this application, a filter may be used to process the voice signal to perform frequency response compensation. The frequency response compensation is used to eliminate noise in the voice related to the recognition device to obtain the user's biological voiceprint information. The filter is the most basic signal processing device, which extracts the required signal from the mixed signals. In this application, the main function of the filter is to eliminate various noises that affect signal processing. The filter generates different gains according to different frequencies, so that specific signals are highlighted, attenuated the highlighted signals, and weaker signals are enhanced , So as to achieve the purpose of eliminating equipment noise.

For example, the filter is shown in the following expression:

The above formula (1) is a finite impulse response filter, where n is the point in time, N is the unit impulse response length of the digital filter, the coefficient a and the coefficient x are convolved to produce the filtered output y, and k starts from 0, and always Take N.

Or, the filter is shown in the following expression:

The above formula (2) is an infinite impulse response filter, where n refers to the point in time, N and P are the unit impulse response length of the digital filter, k starts from 0 and continues to N, and the coefficients a and x Convolution; j starts from 0 and continues to P, and the sum of the convolution of the coefficients b and y produces the filtered output y.

If in the above equation (1), the frequency response of different devices is adjusted to a level or the same by adjusting the coefficients a and x, if in the above equation (2), the frequency response of different devices is all tending to be level or Consistent, so as to filter out the noise related to the device in the speech.

Please refer to Figure 11A. Figure 11A includes three curves, the upper limit curve 1101, the lower limit curve 1102 and the curve 1103 without frequency response compensation. The frequency response curve 1103 has a peak between 600hz-1.2KHz. , The peak exceeds the upper limit curve 1101, and between 300 Hz and 500 Hz, the frequency response curve 1103 is an upward trend curve. Please refer to Figure 11B. Figure 11B includes three curves, the upper limit curve 1101, the lower limit curve 1102 and the frequency response compensated curve 1104. The frequency response tends to be horizontal, or, in the above formula (2), the frequency response of different devices tends to be horizontal by adjusting the coefficients a and b. In the frequency range of 300hz-2.5KHz, the frequency response tends to be consistent (such as -2dBr) , So as to filter out the noise related to the device in the voice, and get the user's biological voiceprint.

In this example, the voice signal is processed by the filter, the prominent signal is attenuated, and the weaker signal is enhanced, thereby compensating the frequency response, so as to achieve the purpose of removing equipment-related noise. The filter is filtered. It directly filters the noise signal, and the output signal is the user's biological voiceprint signal, which is simple and fast to implement.

In an alternative implementation, the biological voiceprint of each person is not static, but will change, such as different time periods of the day, or different health conditions (such as health and sickness), or different Factors such as the age of the user, these factors will cause the biological voiceprint of the same person to change. In order to improve the robustness of the system, multiple voiceprints can be registered for the same user, that is, the same user corresponds to multiple biological voiceprint templates One user ID corresponds to one registered voiceprint information, and the registered voiceprint information includes multiple biological voiceprint templates.

The storage device can generate a voiceprint model (ie, voiceprint template) based on multiple biological voiceprints of the same user. Please refer to Figure 12. Take the x-vector system as an example. The x-vector system is a speaker based on DNN. recognition system. By training the DNN, the speaker's speech is mapped to a fixed-dimensional embedding vector (embeddings), called x-vector.

The x-vector network receives the voiceprint information of the same user. The voiceprint information is the user's voiceprint for removing device-related noise. The x-vector network can capture user's voiceprint information by using shorter speech, and has stronger robustness in short speech. An input can correspond to an x-vector vector, this vector is the user's voiceprint information (because the device-related information is removed before, the voiceprint information has nothing to do with the device), or it becomes the user's voiceprint model. If multiple device-independent voiceprint information is input through multiple devices, or multiple device-independent voiceprint information is input through one device, there will be multiple x-vector vectors. Through linear discriminant analysis, LDA) performs dimensionality reduction. The multiple vectors can cover more user pronunciation conditions (such as user voiceprint information in different time periods, voiceprint information in healthy or unhealthy states, etc.), that is, one user corresponds to multiple The biological voiceprint model (or template), the multiple biological voiceprint models form a voiceprint information template library, which can further enhance the voiceprint recognition effect. The number of templates corresponding to the same user can be 10-30, and it can be updated regularly or irregularly, replacing the old template with a new template to improve the robustness of the biological voiceprint template.

It can be understood that the registration device can register one biological voiceprint template at a time, or multiple biological voiceprint templates can be registered multiple times. Or each of multiple registered devices can register a biological voiceprint template. Optionally, the registration device can send one or more biometric voiceprint templates to the storage device or recognition device through one message at a time, or it can send multiple biometric voiceprint templates to the storage device or the recognition device through multiple messages, respectively.述Recognition equipment. The sending channels include, but are not limited to, wireless and wired methods, which can be specifically implemented through a transceiver. For details, refer to the corresponding description in FIG. 14. The identification device may obtain the one or more biological voiceprint templates from the registration device or the storage device. For the recognition device, the one or more biological voiceprint templates are used to perform voiceprint recognition matching, and belong to the registered voiceprint information that can be used by the recognition device, although the registered voiceprint information is not all The identification device is generated and registered by itself, but from other devices. Through this solution, device-independent voiceprint information sharing is realized, flexible cross-device voiceprint recognition is realized, and user experience is improved.

Corresponding to the methods given in the foregoing method embodiments, the embodiments of the present application also provide corresponding devices, including corresponding modules for executing the foregoing embodiments. The module can be software, hardware, or a combination of software and hardware. As shown in FIG. 13, an embodiment of the present application also provides a device 1300, which may be a terminal or a component of a terminal (for example, an integrated circuit, a chip, etc.). The voiceprint recognition device includes voice input and output. A module 1301 (or a voice input and output unit), a processing module 1302 (or a processing unit), and a transceiver module 1303 (or a transceiver unit).

In a possible design, the device 1300 can perform the function of the recognition device in the above method embodiment: the voice input and output module 1301 is used to receive the voice input by the user; the processing module 1302 is used to eliminate the voice input and output module 1301 received The noise in the voice related to the recognition device is used to obtain target voiceprint information, the target voiceprint information is the user's biological voiceprint; the transceiver module 1303 is used to obtain the user's registered voiceprint information, and the registered voiceprint information includes cancellation and registration equipment At least one biological voiceprint template obtained from related noises; the processing module 1302 is also used to match the target voiceprint information obtained by the processing module 1302 with the registered voiceprint information obtained by the transceiver module 1303 to obtain a matching result, and the matching result is used To indicate the identity of the user.

Further, the voice input and output module 1301 is configured to execute step 701 in the embodiment corresponding to FIG. 7 above. For specific implementation, please refer to the specific description in step 701, which will not be repeated here. The processing module 1302 is configured to execute step 702 and step 704 in the embodiment corresponding to FIG. 7. For specific implementation, please refer to the specific description of step 702 and step 704, which will not be repeated here. The transceiver module 1303 is configured to perform step 703 in the embodiment corresponding to FIG. 7. For specific implementation, please refer to the specific description of step 703, which is not repeated here.

In another possible design, the device 1300 can perform the function of the registered device in the above method embodiment: the voice input and output module 1301 is used to receive the voice input by the user; the processing module 1302 is used to eliminate the voice input and output module 1301 receiving The noise associated with the registered device in the voice of the user can be used to obtain the user's registered voiceprint information. Optionally, the transceiver module 1303 is configured to send the registered voiceprint information and the corresponding voiceprint identifier obtained by the processing module 1302 to other storage devices.

Further, the voice input and output module 1301 is configured to execute step 301 in the embodiment corresponding to FIG. 3 above. For specific implementation, please refer to the specific description in step 301, which will not be repeated here. The processing module 1302 is configured to execute step 302 and step 303 in the embodiment corresponding to FIG. 3, and details are not described here.

In another implementation, the device may be a chip or an integrated circuit. At this time, the transceiver module 1303 may be a communication interface, the processing module 1302 may be a logic circuit, and the voice input and output module 1301 may be an audio circuit. Optionally, the communication interface may be an input/output interface or a transceiver circuit. The input and output interface may include an input interface and an output interface. The transceiver circuit may include an input interface circuit and an output interface circuit.

In an implementation manner, the processing module 1302 may be a processing device, and the functions of the processing device may be partially or fully implemented by software. Optionally, the functions of the processing device may be partially or fully realized by software. At this time, the processing device may include a memory and a processor, where the memory is used to store a computer program, and the processor reads and executes the computer program stored in the memory to perform corresponding processing and/or steps in any method embodiment. Optionally, the processing device may only include a processor. The memory for storing the computer program is located outside the processing device, and the processor is connected to the memory through a circuit/wire to read and execute the computer program stored in the memory.

It can be understood that each functional component in FIG. 13 can be implemented by software, hardware or a combination of the two, which is not specifically limited.

In addition, FIG. 14 is a schematic structural diagram of an apparatus 1400 provided by an embodiment of the application. The device may be a terminal, or may also be an integrated circuit, chip, or chip system in the terminal. The device uses a terminal as an example. The terminal may include, but is not limited to, a terminal in a smart home, a lighting system, a smart speaker, a robot, etc.; the terminal may also be a vehicle-mounted terminal, user equipment, mobile phone, tablet computer, personal computer, etc.

As shown in FIG. 14, the apparatus 1400 includes a processor 1401, a transceiver 1402, a memory 1403, and a voice input and output device 1404. Among them, the processor 1401, the transceiver 1402, the memory 1403, and the voice input and output device 1404 can communicate with each other through internal connection paths, and transfer control signals and/or data signals. Among them, the memory 1403 is used to store a computer program, and the processor 1401 is used to call and run the computer program from the memory 1403 to control the transceiver 1402 to send and receive signals. The processor 1401, which is the control center of the device, uses various interfaces and lines to connect the various parts of the entire mobile phone, by running or executing software programs and/or modules stored in the memory 1403, and calling and storing in the memory 1403 Data, perform various functions of the phone and process data.

The voice input and output device 1404 is used for the audio interface between the user and the mobile phone. The voice input unit may be an audio circuit, or may be a voice recognizer. The audio circuit may include a speaker 14041 and a microphone 14042. The microphone 14042 converts the collected sound signals into electrical signals, which are received by the audio circuit and converted into audio data, and then the audio data is output to the processor 1401 for processing. Device related noise to get the user’s biological voiceprint.

Optionally, the device may also include an antenna. The transceiver 1402 transmits or receives wireless signals through an antenna. The transceiver 1402 can be used to send or receive registered voiceprint information and corresponding voiceprint identifiers to other devices. Optionally, the processor 1401 and the memory 1403 may be combined into one processing device, and the processor 1401 is configured to execute the program code stored in the memory 1403 to implement the foregoing functions. Optionally, the memory 1403 may also be integrated in the processor 1401. Alternatively, the memory 1403 is independent of the processor 1401, that is, located outside the processor 1401. Optionally, the transceiver 1402 includes but is not limited to a radio frequency (RF) circuit, a communication interface, a WiFi module, a Bluetooth module module, and so on.

Optionally, the device may further include a display unit 1405 which can be used to display information input by the user or information provided to the user and various images. The display unit can be configured with a display panel in the form of a liquid crystal display (Liquid Crystal Display, LCD), an organic light-emitting diode (Organic Light-Emitting Diode, OLED), etc.

In a possible design, the device 1400 can be used to perform the functions performed by the recognition device in the method embodiment: the voice input and output device 1404 is used to receive the voice input by the user; the processor 1401 is used to eliminate the voice in the voice input and output device 1404. Recognize the noise related to the device to obtain target voiceprint information, the target voiceprint information is the user's biological voiceprint; the transceiver 1402 is used to obtain the user's registered voiceprint information, and the registered voiceprint information includes eliminating noise related to the registered device The obtained at least one biological voiceprint template; the processor 1401 is further configured to match the target voiceprint information with the registered voiceprint information to obtain a matching result, and the matching result is used to indicate the identity of the user.

Optionally, the transceiver 1402 is configured to receive the registered voiceprint information and the voiceprint identifier of the registered voiceprint information sent by the storage device or the registered device, and the voiceprint identifier is used to indicate the user. Optionally, the transceiver 1402 is also used to send the user's voiceprint identification to the storage device or the registration device to request the user's registered voiceprint information.

Optionally, the processor 1401 is specifically configured to input the voice into the voiceprint extraction model, and eliminate noise related to the recognition device in the voice through the voiceprint extraction model to obtain target voiceprint information. Optionally, the voiceprint extraction model is obtained by learning corpus collected by multiple devices. Optionally, the processor 1401 is further configured to perform frequency response compensation on the voice through a filter, and the frequency response compensation is used to eliminate noise related to the recognition device in the voice to obtain target voiceprint information. Optionally, the processor 1401 is further configured to obtain user data associated with the user identity; and perform operations corresponding to the user data.

In a possible design, the device 1400 is used to perform the functions performed by the registered device in the above method embodiment: the voice input and output device 1404 is used to receive the voice input by the user; the processor 1401 is used to: The noise related to the registered device is used to obtain the user's registered voiceprint information, and the registered voiceprint information specifically includes the user's biological voiceprint.

Optionally, the transceiver 1402 is configured to send the registered voiceprint information and the voiceprint identifier corresponding to the registered voiceprint information to the storage device or the recognition device, and the voiceprint identifier is used to indicate the user. Optionally, the processor 1401 is specifically configured to input the voice into the voiceprint extraction model, and eliminate noise related to the registered device in the voice through the voiceprint extraction model to obtain registered voiceprint information. Optionally, the voiceprint extraction model is obtained by learning corpus collected by multiple devices. Optionally, the processor 1401 is further configured to perform frequency response compensation on the voice through a filter, and the frequency response compensation is used to eliminate noise related to the recognition device in the voice to obtain registered voiceprint information.

It can be understood that the processor in the embodiment of the present application may be an integrated circuit chip with signal processing capability. In the implementation process, the steps of the foregoing method embodiments can be completed by hardware integrated logic circuits in the processor or instructions in the form of software. The above-mentioned processor can be general processing, digital signal processor (digital signal processor, DSP), application specific integrated circuit (ASIC), field programmable gate array (field programmable gate array, FPGA) or other programmable Logic devices, discrete gates or transistor logic devices, discrete hardware components.

The solution described in this application can be implemented in various ways. For example, these technologies can be implemented in hardware, software, or a combination of hardware. For hardware implementation, the processing unit used to execute these technologies at a communication device (for example, a base station, a terminal, a network entity, or a chip) can be implemented in one or more general-purpose processors, DSPs, digital signal processing devices, ASICs, Programmable logic device, FPGA, or other programmable logic device, discrete gate or transistor logic, discrete hardware component, or any combination of the foregoing. The general-purpose processor may be a microprocessor. Alternatively, the general-purpose processor may also be any traditional processor, controller, microcontroller, or state machine. The processor can also be implemented by a combination of computing devices, such as a digital signal processor and a microprocessor, multiple microprocessors, one or more microprocessors combined with a digital signal processor core, or any other similar configuration. accomplish.

It can be understood that the memory in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memory. Among them, the non-volatile memory can be read-only memory (ROM), programmable read-only memory (programmable ROM, PROM), erasable programmable read-only memory (erasable PROM, EPROM), and electrically available Erase programmable read-only memory (electrically EPROM, EEPROM) or flash memory. The volatile memory may be random access memory (RAM), which is used as an external cache. By way of exemplary but not restrictive description, many forms of RAM are available, such as static random access memory (static RAM, SRAM), dynamic random access memory (dynamic RAM, DRAM), and synchronous dynamic random access memory (synchronous DRAM, SDRAM), double data rate synchronous dynamic random access memory (double data rate SDRAM, DDR SDRAM), enhanced synchronous dynamic random access memory (enhanced SDRAM, ESDRAM), synchronous connection dynamic random access memory (synchlink DRAM, SLDRAM) ) And direct memory bus random access memory (direct rambus RAM, DR RAM). It should be noted that the memories of the systems and methods described herein are intended to include, but are not limited to, these and any other suitable types of memories.

The present application also provides a computer-readable medium on which a computer program is stored, and when the computer program is executed by a computer, the function of any of the foregoing method embodiments is realized. This application also provides a computer program product, which, when executed by a computer, realizes the functions of any of the foregoing method embodiments.

It can be understood that the “embodiment” mentioned throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of the present application. Therefore, the various embodiments throughout the specification do not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It can be understood that, in the various embodiments of the present application, the size of the sequence number of the above-mentioned processes does not imply the order of execution, and the execution order of each process should be determined by its function and internal logic, and should not correspond to the embodiments of the present application. The implementation process constitutes any limitation.

It can be understood that in this application, "when", "if" and "if" all mean that the device will make corresponding processing under certain objective circumstances. It is not a time limit, and it does not require the device to be implemented. There must be a judgmental action, and it does not mean that there are other restrictions. The use of the singular element in this application is intended to mean "one or more" rather than "one and only one", unless otherwise specified. In this application, unless otherwise specified, "at least one" is intended to mean "one or more", and "multiple" is intended to mean "two or more". In addition, the terms "system" and "network" in this article are often used interchangeably in this article.

The term "at least one of..." or "at least one of..." as used herein means all or any combination of the listed items, for example, "at least one of A, B and C", It can mean: A alone exists, B alone exists, C exists alone, A and B exist at the same time, B and C exist at the same time, and there are six cases of A, B and C at the same time, where A can be singular or plural, and B can be Singular or plural, C can be singular or plural.

It can be understood that in the embodiments of the present application, "B corresponding to A" means that B is associated with A, and B can be determined according to A. However, it should also be understood that determining B based on A does not mean that B is determined only based on A, and B can also be determined based on A and/or other information. The term "and/or" in this article is only an association relationship describing the associated objects, which means that there can be three relationships, for example, A and/or B, which can mean: A alone exists, A and B exist at the same time, exist alone In the three cases of B, A can be singular or plural, and B can be singular or plural. The character "/" generally indicates that the associated objects before and after are in an "or" relationship.

Those of ordinary skill in the art can understand that, for the convenience and conciseness of the description, the specific working process of the system, device, and unit described above can refer to the corresponding process in the foregoing method embodiment, and will not be repeated here.

It can be understood that the systems, devices, and methods described in this application can also be implemented in other ways. For example, the device embodiments described above are merely illustrative, for example, the division of the units is only a logical function division, and there may be other divisions in actual implementation, for example, multiple units or components may be combined or It can be integrated into another system, or some features can be ignored or not implemented. In addition, the displayed or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection through some interfaces, devices or units, and may be in electrical, mechanical or other forms.

If the functions described in this embodiment are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solution of the present application essentially or the part that contributes to the existing technology or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including Several instructions are used to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (read-only memory, ROM), random access memory (random access memory, RAM), magnetic disks or optical disks and other media that can store program codes. .

The same or similar parts in the various embodiments of this application may be referred to each other. In each embodiment of this application, and each implementation method/implementation method/implementation method in each embodiment, if there is no special description and logical conflict, between different embodiments and each implementation manner/implementation method in each embodiment/ The terms and/or descriptions between the implementation methods/implementation methods are consistent and can be cited each other. The technical features in different embodiments and various implementation modes/implementation methods/implementation methods in each embodiment are based on their inherent The logical relationship can be combined to form a new embodiment, implementation, implementation method, or implementation method. The implementation manners of the application described above do not constitute a limitation on the protection scope of the application.

The above are only specific implementations of this application, but the protection scope of this application is not limited to this. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in this application. Should be covered within the scope of protection of this application.

Claims

A voiceprint recognition device in a recognition device, which is characterized by comprising: a processor, a voice input and output device and a transceiver connected to the processor;

The voice input and output device is used to receive the voice input by the user;

The processor is configured to eliminate noise related to the recognition device in the voice to obtain target voiceprint information, where the target voiceprint information is the biological voiceprint of the user;

The transceiver is configured to obtain registered voiceprint information of the user, where the registered voiceprint information includes at least one biological voiceprint template obtained by eliminating noise related to a registered device;

The processor is further configured to match the target voiceprint information with the registered voiceprint information to obtain a matching result, and the matching result is used to indicate the user identity.
The device of claim 1, wherein:

The transceiver is configured to receive the registered voiceprint information and the voiceprint identifier of the registered voiceprint information sent by a storage device or the registration device, where the voiceprint identifier is used to indicate the user.
The device of claim 2, wherein:

The transceiver is further configured to send the voiceprint identifier of the user to the storage device or the registration device to request the user's registered voiceprint information.
The device according to any one of claims 1 to 3, characterized in that:

The processor is specifically configured to input the voice into a voiceprint extraction model, and eliminate noise related to the recognition device in the voice through the voiceprint extraction model to obtain the target voiceprint information.
The device according to claim 4, wherein the voiceprint extraction model is obtained by learning corpus collected by multiple devices.
The device according to any one of claims 1 to 3, characterized in that:

The processor is further configured to perform frequency response compensation on the voice through a filter, and the frequency response compensation is used to eliminate noise in the voice related to the recognition device to obtain the target voiceprint information.
The device according to claims 1-6, wherein the processor is further configured to:

Obtaining user data associated with the user identity;

Perform an operation corresponding to the user data.
A voiceprint registration device in a registration device, characterized by comprising: a processor and a voice input and output device connected to the processor;

The voice input and output device is used to receive the voice input by the user;

The processor is configured to eliminate noise related to the registered device in the voice to obtain registered voiceprint information of the user, where the registered voiceprint information includes the biological voiceprint of the user.
The device according to claim 8, wherein the device further comprises a transceiver connected to the processor;

The transceiver is configured to send the registered voiceprint information and a voiceprint identifier corresponding to the registered voiceprint information to a storage device or an identification device, where the voiceprint identifier is used to indicate the user.
The device according to claim 8 or 9, characterized in that:

The processor is specifically configured to input the voice into a voiceprint extraction model, and eliminate noise related to the registered device in the voice through the voiceprint extraction model to obtain the registered voiceprint information.
The device according to claim 10, wherein the voiceprint extraction model is obtained by learning corpus collected by multiple devices.
The device according to claim 8 or 9, characterized in that:

The processor is further configured to perform frequency response compensation on the voice through a filter, and the frequency response compensation is used to eliminate noise related to the recognition device in the voice to obtain the registered voiceprint information.
A method for cross-device voiceprint recognition, applied to a recognition device, is characterized in that it includes:

Receive the voice input by the user;

Removing noise related to the recognition device in the voice to obtain target voiceprint information, where the target voiceprint information is a biological voiceprint of the user;

Acquiring registered voiceprint information of the user, where the registered voiceprint information includes at least one biological voiceprint template obtained by eliminating noise related to the registered device;

The target voiceprint information is matched with the registered voiceprint information to obtain a matching result, and the matching result is used to indicate the identity of the user.
The method according to claim 13, wherein said obtaining registered voiceprint information of said user comprises:

Receiving the registered voiceprint information and a voiceprint identifier corresponding to the registered voiceprint information sent by a storage device or the registration device, where the voiceprint identifier is used to indicate the user.
The method according to claim 14, wherein before the receiving the registered voiceprint information and the voiceprint identifier corresponding to the registered voiceprint information sent by the storage device or the registration device, the method further comprises :

Send the voiceprint identifier to the storage device or the registration device to request the user's registered voiceprint information.
The method according to any one of claims 13-15, wherein the removing noise related to the recognition device in the speech to obtain target voiceprint information comprises:

The voice is input into a voiceprint extraction model, and noise related to the recognition device in the voice is eliminated through the voiceprint extraction model to obtain the target voiceprint information.
The method according to claim 16, wherein the voiceprint extraction model is obtained by learning corpus collected by multiple devices.
The method according to any one of claims 13-17, wherein after the matching the target voiceprint information with the registered voiceprint information to obtain a matching result, the method further comprises:

Obtaining user data associated with the user identity;

Perform an operation corresponding to the user data.
A method for cross-device voiceprint recognition, applied to registered devices, is characterized in that it includes:

Receive the voice input by the user;

Eliminate noise related to the registered device in the voice to obtain registered voiceprint information; the registered voiceprint information includes the biological voiceprint of the user.
The method according to claim 19, wherein the method further comprises:

Send the registered voiceprint information and the voiceprint identifier corresponding to the registered voiceprint information to a storage device or an identification device.
The method according to claim 19 or 20, wherein the removing noise related to the registration device in the voice to obtain biological voiceprint information comprises:

The voice is input into a voiceprint extraction model, and noise related to the registered device in the voice is eliminated through the voiceprint extraction model to obtain the registered voiceprint information.
The method according to claim 21, wherein the voiceprint extraction model is obtained by learning corpus collected by multiple devices.
The method according to claim 19 or 20, wherein the removing noise related to the registration device in the voice to obtain biological voiceprint information comprises:

Perform frequency response compensation on the voice through a filter, and the frequency response compensation is used to eliminate noise related to the recognition device in the voice to obtain the registered voiceprint information.