WO2022144009A1 - 分布式语音处理系统及方法 - Google Patents

分布式语音处理系统及方法 Download PDF

Info

Publication number
WO2022144009A1
WO2022144009A1 PCT/CN2021/143983 CN2021143983W WO2022144009A1 WO 2022144009 A1 WO2022144009 A1 WO 2022144009A1 CN 2021143983 W CN2021143983 W CN 2021143983W WO 2022144009 A1 WO2022144009 A1 WO 2022144009A1
Authority
WO
WIPO (PCT)
Prior art keywords
sound
preprocessing
speech recognition
result
network
Prior art date
Application number
PCT/CN2021/143983
Other languages
English (en)
French (fr)
Inventor
毛建鑫
Original Assignee
乐鑫信息科技（上海）股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 乐鑫信息科技（上海）股份有限公司 filed Critical 乐鑫信息科技（上海）股份有限公司
Priority to US18/260,196 priority Critical patent/US20240062764A1/en
Publication of WO2022144009A1 publication Critical patent/WO2022144009A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00—Speech recognition
    • G10L15/28—Constructional details of speech recognition systems
    • G10L15/30—Distributed recognition, e.g. in client-server systems, for mobile phones or network applications
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00—Speech recognition
    • G10L15/02—Feature extraction for speech recognition; Selection of recognition unit
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00—Speech recognition
    • G10L15/28—Constructional details of speech recognition systems
    • G10L15/32—Multiple recognisers used in sequence or in parallel; Score combination systems therefor, e.g. voting systems

Definitions

  • the present application relates to the field of distributed speech processing, and in particular, to a distributed speech processing system and method.
  • Speech recognition technology and keyword recognition technology are becoming more and more mature, and they are more and more widely used in the market. For example, children's toys, educational products, smart home and other products have added the speech recognition function to realize the function of voice interactive control.
  • the voice recognition process is to collect the original voice into a device, and perform calculations on this device to obtain the recognition result.
  • the speech recognition cannot be successfully completed, which leads to the situation that the recognition cannot be recognized or the recognition effect is poor.
  • speech recognition is usually performed on the market through smart speakers or smart gateways. These devices serve as the control center and the only entry point for speech recognition.
  • the device must be connected to the Internet first, and then the corresponding cloud server must also be connected to the Internet.
  • These devices obtain speech recognition results from the cloud, and then complete speech recognition or voice control.
  • problems with this method For example, the failure of the equipment used as the only entrance for speech recognition, or the fluctuation of the network, etc., will cause the problem of speech recognition failure. In particular, when the network stability is poor, it is easy to cause a slow recognition response.
  • this kind of speech recognition method uploads the speech to the cloud, and the device needs to monitor the surrounding sound in real time, which may easily lead to user privacy and security issues.
  • Chinese patent (CN111415658A) discloses a decentralized voice control multi-device system and its control method.
  • the device first recognizes the wake-up word in the speech, then sends the recognized wake-up word to all devices in the system, and simultaneously receives wake-up words sent by other devices in the system.
  • the device filters all wake-up words and filters out the wake-up words that match the device.
  • the voice received by the device contains a wake-up word (ie, voice command) that is not supported by itself, it may cause voice control to fail.
  • Chinese patent (CN110136708A) discloses a distributed voice control system and control method based on Bluetooth mesh.
  • the control system includes a Bluetooth mesh network, a voice controller, and a Bluetooth node device; the voice controller includes voice acquisition, voice noise reduction, and voice identification, bluetooth module, and optional wifi module.
  • the voice controllers communicate with each other through Bluetooth and keep data synchronization in real time. Any voice controller can control the Bluetooth node device in the network; the Bluetooth node device communicates with the voice controller through the Bluetooth Mesh network, and receives Mesh data or its own Response to key events.
  • each voice controller collects voice, performs voice noise reduction and echo cancellation, and then performs local or online voice recognition, semantic understanding parses the information to be controlled, encapsulates it into Mesh data, and sends it to the Mesh network through the Bluetooth module middle. If the voice controller does not support the current control commands, the device may not recognize the voice commands that it does not support, and eventually the voice control fails.
  • the present invention discloses a distributed voice processing system and a processing method thereof.
  • a distributed voice processing system including a plurality of node devices, the plurality of node devices form a network, wherein each node device includes a processor, a memory, a communication module, and a sound processing module, and At least one node device in the plurality of node devices includes a sound collection module; wherein,
  • the sound collection module is configured to collect audio signals
  • the sound processing module is configured to preprocess the audio signal to obtain a first sound preprocessing result
  • the communication module is configured to send the sound preprocessing result to one or more node devices in the network;
  • the communication module is further configured to receive, from the network, one or more second sound preprocessing results from at least one other node device;
  • the sound processing module is further configured to perform speech recognition based on the first sound preprocessing result and/or the one or more second sound preprocessing results to obtain a first speech recognition result.
  • the communication module is further configured to send the first speech recognition result to one or more node devices in the network;
  • the communication module is further configured to receive from the network one or more second speech recognition results from at least one other node device; and
  • the sound processing module is further configured to perform speech recognition based on the first speech recognition result and the one or more second speech recognition results to obtain a final speech recognition result.
  • a processing method of a distributed voice processing system executed by a node device in a network, including:
  • the node device includes a sound acquisition module, perform the following steps: collect audio signals; preprocess the audio signals to obtain a first sound preprocessing result; and send the sound preprocessing result to the Mesh one or more node devices in the network;
  • Speech recognition is performed based on the first sound preprocessing result and/or the one or more second sound preprocessing results.
  • the processing method of the distributed speech processing system further comprises sending the first speech recognition result to one or more node devices in the network;
  • Speech recognition is performed based on the first speech recognition result and the one or more second speech recognition results to obtain a final speech recognition result.
  • the solution provided by the present application can extend the recognition distance without accessing the Internet, improve the recognition rate when people move, and easily realize cross-room voice control. At the same time, it can also take a step closer to making speech recognition closer to user habits and more adaptable to life usage scenarios.
  • the present invention performs distributed speech recognition through node devices in the network, and can realize speech recognition control over a long distance or across multiple rooms.
  • the technical solution of the present invention enables each node device in the network to participate in the process of speech recognition.
  • a decentralized design is realized, thereby reducing the recognition failure caused by the failure of key central nodes, and the design can realize
  • the devices in the network perform speech recognition concurrently, which can improve the efficiency of speech recognition; on the other hand, the information transmitted during the recognition process is sound preprocessing information, that is, non-original audio data, so the network bandwidth requirements are not high. Improved the stability of speech recognition.
  • the solution in the present invention transmits the sound preprocessing result, which can avoid recognition failure caused by unsupported instructions, and improve the stability and robustness of speech recognition.
  • Fig. 1 schematically shows a block diagram of one embodiment of a distributed speech processing system according to the present invention
  • FIG. 2 schematically shows a block diagram of another embodiment of a distributed speech processing system according to the present invention.
  • FIG. 3 schematically shows a block diagram of a node device according to an embodiment of the distributed speech processing system of the present invention
  • FIG. 4 schematically shows a flow chart of an embodiment of a distributed speech processing method according to the present invention
  • Fig. 5 schematically shows a flow chart of another embodiment of the distributed speech processing method according to the present invention.
  • FIG. 6 schematically shows a flow chart of a specific embodiment of the distributed speech processing method according to the present invention.
  • FIG. 7 schematically shows a flow chart of another specific embodiment of the distributed speech processing method according to the present invention.
  • FIG. 1 shows a block diagram of one embodiment of a distributed speech processing system 100 including a plurality of node devices 102 , 104 , 106 and 112 , 114 and 116 in a network 110 in accordance with the present invention.
  • the network 100 may be, for example, a wired, wireless and/or wired-wireless hybrid network used in homes and/or offices, including but not limited to wireless networks commonly employed in smart home scenarios.
  • a network 110 is formed between various devices, and communication can be performed in a wired or wireless manner.
  • the wired mode can adopt communication methods such as network cable or power line carrier
  • the wireless mode can adopt communication methods such as Wi-Fi, BLE, Zigbee, etc., to realize networking communication between various devices.
  • each node device has the ability to connect to other node devices.
  • An ad hoc network can be performed between each node device to form an ad hoc network or a group network.
  • Each device can also form a Mesh network, which can realize that any device node in the Mesh network can act as a router at the same time, that is, each node in the network can send and receive signals, and each node can communicate with one or more pairs. Wait for the nodes to communicate directly.
  • Fig. 2 schematically shows a block diagram of another embodiment of a distributed speech processing system 200 according to the present invention, wherein some node devices form a group, so that the system of the present invention can also transmit to the group in a broadcast or multicast manner information.
  • the node devices may be in one or more groups, and the group may be dynamic and user-defined, without requiring that the node devices between the groups must have a fixed hardware or communication connection relationship.
  • user 108 is located between device B and device C, and is within the pickup distance of device B and device C, but user 108 is far away from device A and other devices, and cannot directly receive user 108 through device A and other devices voice signal.
  • FIG. 3 schematically shows a block diagram of a node device 300 according to an embodiment of the distributed speech processing system of the present invention.
  • each node device 300 may include a processor 302 , a memory 304 , a communication module 306 , and a sound processing module 312 .
  • At least one node device of the plurality of node devices includes a sound collection module 308 .
  • the node device 300 may further include an output module 312, wherein the processor 302 may provide a ⁇ s-level precise clock; the communication module 306 may be wired (eg, network cable/power line carrier, etc.) or wireless (eg, Wi-Fi/ BLE/Zigbee, etc.) for networking communication with other devices; the memory 304 can record networking information and identify model parameters; the output module 312 can be, for example, a speaker, a switch device, etc.; the sound collection module 308 can be, for example, Either a single microphone, multiple microphones, or an array of microphones.
  • the processor 302 may provide a ⁇ s-level precise clock
  • the communication module 306 may be wired (eg, network cable/power line carrier, etc.) or wireless (eg, Wi-Fi/ BLE/Zigbee, etc.) for networking communication with other devices
  • the memory 304 can record networking information and identify model parameters
  • the output module 312 can be, for example, a speaker, a switch device
  • the sound acquisition module 308 may be configured to acquire audio signals.
  • the sound processing module 310 may be configured to preprocess the audio signal to obtain locally generated sound preprocessing results.
  • the communication module 306 may be configured to transmit the locally generated sound preprocessing results to one or more node devices in the network 110 .
  • the communication module 306 may also be configured to receive from the network 110 one or more sound preprocessing results from at least one other node device.
  • the locally generated sound preprocessing result may be referred to as the "first sound preprocessing result”
  • the sound preprocessing result received from other node devices through the network may be referred to as the "second sound preprocessing result”. result”.
  • the sound processing module 310 may also be configured to perform speech recognition based on locally generated sound preprocessing results and/or one or more sound preprocessing results received over the network 110 . In this way, the node device 300 can obtain the locally generated speech recognition result.
  • the speech recognition performed by the sound processing module of the node device may include, but is not limited to, wake word detection, keyword recognition, continuous speech recognition, and the like.
  • the speech recognition result obtained by the speech processing module of the node device performing speech recognition may include the device identification, the recognition result, the valid time of the recognition result, the recognition start time, and the sound quality.
  • the first voice recognition result may also include instruction information and device identification, so as to instruct the target device to perform corresponding operations.
  • the distributed speech recognition scheme of the present invention can utilize voice preprocessing results generated locally and from the network on the one hand, and speech recognition results generated locally and from the network on the other hand.
  • the node device of the present invention can arbitrate speech recognition results from different sources.
  • the communication module 306 may also be configured to transmit the locally generated speech recognition results to one or more node devices in the network.
  • the communication module 306 is also configured to receive from the network one or more speech recognition results from at least one other node device.
  • first speech recognition result the locally generated speech recognition result
  • second speech recognition result the speech recognition result received from other node devices through the network
  • the sound processing module 310 is further configured to perform speech recognition based on the first speech recognition result and the one or more second speech recognition results to obtain a final speech recognition result.
  • the sound processing module 310 is further configured to perform weighting processing on the first speech recognition result and one or more second speech recognition results to obtain a final speech recognition result.
  • the sound processing module of the node device may be configured to assign a weight based on the sound quality of the first speech recognition result and the one or more second speech recognition results, and the higher the sound quality, the higher the weight assigned.
  • the sound processing module of the node device can be configured to assign weights based on the source device of the first speech recognition result and one or more second speech recognition results. If the source device is the node device, the assigned weight is greater.
  • a sound preprocessing result is an intermediate result produced during the recognition process from the original speech to the speech recognition result.
  • each of the first sound preprocessing result and the one or more second sound preprocessing results includes sound feature value, sound quality, and sound time information.
  • the communication module 306 of the node device receives from the network one or more second sound preprocessing results from at least one other node device, wherein the second sound preprocessing results include sound feature values, sound quality and sound
  • the time information may also include an increasing sequence number of the audio signal.
  • the sound feature value in the preprocessing result is the MFCC feature value or the PLP feature value of the audio signal.
  • Sound quality may include the signal-to-noise ratio and amplitude of the audio signal.
  • the sound time information may include a start time and an end time of the audio signal.
  • the sound time information may include the start time and duration of the audio signal.
  • the preprocessing techniques applicable to the present invention may include, but are not limited to, signal framing, pre-emphasis, Fast Fourier Transform (Fast Fourier Transform, FFT) and other preprocessing.
  • the preprocessing method can obtain audio parameters according to the audio signal, generate frequency domain signals, or perform Mel-Frequency Cepstral Coefficients (MFCC) algorithm or Perceptual Linear Predictive (PLP) algorithm extraction for characterization.
  • MFCC Mel-Frequency Cepstral Coefficients
  • PLP Perceptual Linear Predictive
  • the sound processing module 310 of the node device is further configured to, for each of the first sound preprocessing result and the one or more second sound preprocessing results, determine whether the sound quality exceeds a predetermined threshold, if If not, discard the speech preprocessing result.
  • the sound processing module 310 of the node device is further configured to select one or more sound preprocessing results with the highest sound quality from the first sound preprocessing result and the one or more second sound preprocessing results Perform speech recognition to obtain a first speech recognition result.
  • the first voice recognition result obtained by the voice processing module of the node device performing voice recognition may include instruction information, where the instruction information is a specific numerical value, such as 011, and the instruction information can, for example, be determined by the node device that supports the corresponding instruction. understand and execute.
  • the first voice recognition result obtained by the voice processing module of the node device performing the voice recognition may include instruction information, wherein different node devices support different ranges of instruction information.
  • the sound processing module of the node device may also be configured to select one or more sound preprocessing results with the highest sound quality from the first sound preprocessing result and one or more second sound preprocessing results to perform speech recognition to obtain The first speech recognition result.
  • the sound processing module of the node device can determine whether the sound quality of the first sound preprocessing result exceeds a predetermined threshold, and if it exceeds, select the first sound preprocessing result for speech recognition to obtain the first speech recognition result.
  • the communication module of the node device may send the first speech recognition result to one or more node devices in the network by means of unicast, multicast and/or broadcast.
  • the first voice recognition result is sent to a network node through the communication module of the node device. or multiple node devices.
  • the first voice recognition result is not sent to one or more node devices in the network.
  • the sound processing module of the node device may perform a time validity judgment on the obtained final speech recognition result, and if the validity time of the recognition result exceeds the valid time of the recognition result, the corresponding operation corresponding to the recognition result will not be performed.
  • the sound processing module of the node device can determine the device identification for the obtained final speech recognition result, and if the device identification is the node device, execute the corresponding operation corresponding to the identification result.
  • the voice processing module of the node device may determine the device identifier for the obtained final speech recognition result, and if the device identifier is the node device, output feedback information and send it to one or more other network devices through the communication module node device.
  • the voice processing module of the node device can determine the device ID for the final speech recognition result obtained, and if the device ID is the node device, output feedback information, where the output feedback information at least includes the recognition time, the recognition result and the maximum value of the incremented serial number. .
  • Fig. 4 schematically shows a flow chart of an embodiment of a distributed speech processing method 400 according to the present invention.
  • the distributed voice processing method is performed by node devices in the network.
  • an audio signal is acquired.
  • the audio signal is preprocessed to obtain a first sound preprocessing result.
  • the first sound preprocessing result is sent to one or more node devices in the network.
  • one or more second sound preprocessing results from at least one other node device are received from the network.
  • speech recognition is performed based on the first sound preprocessing result and/or the one or more second sound preprocessing results.
  • one or more sound preprocessing results with the highest sound quality may be selected from the first sound preprocessing result and the one or more second sound preprocessing results for speech recognition to obtain the first speech recognition result .
  • the method of the present invention may combine the local recognition result with the recognition result from the network to obtain the final speech recognition result.
  • the node device may send the first speech recognition result to one or more node devices in the network.
  • the node device may receive from the network one or more second speech recognition results from at least one other node device.
  • the node device may perform speech recognition based on the first speech recognition result and the one or more second speech recognition results to obtain a final speech recognition result.
  • a weighted average may be performed on the first speech recognition result and one or more second speech recognition results to obtain a final speech recognition result.
  • the solution of the present invention is also obtained by splicing fragments generated locally and from the sound preprocessing result of the network.
  • a distributed voice processing system comprising: a plurality of node devices forming a network, wherein each node device includes a processor, a memory, a communication module and a sound processing module, and At least one node device in the plurality of node devices includes a sound collection module; wherein the sound collection module is configured to collect audio signals; the sound processing module is configured to preprocess the audio signals to obtain a first sound preprocessing result; wherein the communication The module is further configured to receive from the network one or more second sound preprocessing results from at least one other node device; each of the first sound preprocessing results and the one or more second sound preprocessing results includes One or more data blocks; each data block in the one or more data blocks includes time information, the time information identifies the time when the sound processing module completes the preprocessing of the data block; the one or more data blocks Each data block further includes an increasing sequence number, which is allocated according to the time information in the data block; the sound processing module is further configured to splicing the first sound
  • the communication module is configured to send the first sound preprocessing result to one or more node devices in the network.
  • each data block in the first and/or second sound preprocessing result is configured to have the same duration.
  • the incremental sequence number is allocated to the data block when the sound processing module of each node device in the plurality of node devices preprocesses the audio signal.
  • the incremented sequence number is the result of the second sound preprocessing after the sound processing module of each node device in the plurality of node devices receives the second sound preprocessing result from at least one other node device from the network. data block allocation.
  • the sound processing module is configured to detect the time difference of the data blocks and assign the same incremental sequence number if the time difference is within a specified threshold.
  • the sound processing module is configured to select a data block with the best sound quality from the data blocks with the same increasing sequence number for splicing.
  • FIG. 5 schematically shows a flow chart of another embodiment of a distributed speech processing method 500 according to the present invention.
  • the method is performed by node devices in the network.
  • it is determined whether the node device includes a sound collection module. If so, go to step 504. If not, go to step 508.
  • audio signals are acquired.
  • the audio signal is preprocessed to obtain a first sound preprocessing result.
  • each of the first sound preprocessing result and the one or more second sound preprocessing results includes one or more A plurality of data blocks, wherein each data block in the one or more data blocks includes time information identifying the time at which the sound processing module completed preprocessing of the data block, and wherein each data block in the one or more data blocks
  • the block also includes an incremental sequence number, which is assigned according to the time information in the data block.
  • the data blocks of the first sound preprocessing result and/or one or more second sound preprocessing results are spliced in an ascending order of increasing sequence numbers to obtain a complete third sound preprocessing result.
  • the third sound preprocessing result is processed to obtain a final speech recognition result.
  • the sound preprocessing result obtained by splicing the data blocks of the first sound preprocessing result and/or one or more second sound preprocessing results is referred to as the "third sound preprocessing result" .
  • the distributed voice processing method further includes sending the first voice preprocessing result to one or more node devices in the network.
  • each data block in the first sound preprocessing result is configured to have the same duration.
  • the incremental sequence number is allocated to the data block when the sound processing module of each node device in the plurality of node devices preprocesses the audio signal, wherein the incremental sequence number is allocated according to time information.
  • the incremented sequence number is the result of the second sound preprocessing after the sound processing module of each node device in the plurality of node devices receives the second sound preprocessing result from at least one other node device from the network. Allocated by data blocks, where the incrementing sequence number is allocated according to the time information.
  • the sound processing module is configured to detect the time difference of the data blocks, and if the time difference is within a threshold, assign the same incremental sequence number.
  • the sound processing module is configured to select the data block with the best sound quality from the data blocks with the same increasing sequence number.
  • FIG. 6 schematically shows a flow chart of a specific embodiment of a distributed speech processing method 600 according to the present invention.
  • the node devices are self-organized, a group is established, and each node device in the network is allowed to perform speech recognition, and the method of exchanging identification information in the group is to recognize the speech originally generated by a node device.
  • the system is improved to a speech recognition system distributed on multiple node devices, thus solving the problems of relying on a single control center, relying on a network server, not being able to cross regions, and insecure privacy information in speech recognition, keyword recognition and voice control scenarios.
  • step 604 when the node device is powered on, it is found whether there is a group network. If no group network exists, then at step 606 a group network is created. If a group network already exists, the group network is joined at step 608 . After the node device accesses the group network, it first updates the function points of other devices in the network in step 610 to know whether the function points supported by other devices in the network have been modified. Broadcast in a group network. It should be understood that in the context of the present invention, "function point" is used to inform other node devices of other access groups which aspects of input and output functions they have.
  • a "group network” refers to a network formed by node devices that support broadcast and/or multicast, including but not limited to networks with various topologies (eg, Mesh topologies). Wi-Fi, BLE and ZigBee networks, and can be wired, wireless or hybrid.
  • the node device obtains the identification result through distributed identification, and in step 616, it is determined whether the device identification of the identification result is the present device. If the device identification is not the own device, the identification information is sent in step 622 .
  • the identification information may include identification device identification, identification time, identification result, and reliability of the identification result. If the device identification is the own device, the output is performed in step 618, and then the execution result information is sent to other node devices in the network in step 620.
  • the execution result information may include device identification, identification time, identification result, execution result, and the like.
  • FIG. 7 schematically shows a flow chart of another specific embodiment of a distributed speech processing method 700 according to the present invention.
  • the distributed speech processing method in this embodiment includes three inputs and one output.
  • the three inputs are: the sound collected by the local microphone in step 702 and the sound preprocessing collected from the network in step 708 information, and speech recognition information collected from the network at step 714; one output is the speech recognition result output at step 720.
  • the distributed speech processing method 700 is divided into three stages: a preprocessing stage, an analysis decision stage, and a recognition arbitration stage.
  • the sound collected by the local microphone will first undergo the sound preprocessing in step 704 to obtain preprocessing information, and then in step 706 the preprocessing information is sent to the group network.
  • the preprocessing information includes, for example, feature information of the collected sound that can be used for recognition of the model.
  • the preprocessing information also includes information such as the signal-to-noise ratio and amplitude of the collected sound, which can be used to evaluate the sound quality.
  • the preprocessing information also includes, for example, an incrementing sequence number of the information.
  • the preprocessing information may also include start time information and end time information.
  • the sound quality of the voice preprocessing information collected by the network and the voice preprocessing information obtained locally is sorted, and the best quality preprocessing information is analyzed and selected and sent to the subsequent speech recognition step 712 .
  • speech recognition is performed to output local recognition information.
  • the local or network identification information may include, for example, but not limited to, the result of the recognition, the device identification of the speech recognition, the valid time of the recognition result, the recognition start time, and the translator or more of the sound quality.
  • the identification information collected from the network in step 714 is analyzed and judged in step 716, and the expired information is removed according to the aging of the identification information.
  • Recognition arbitration is then performed at step 718 along with the output of the local speech recognition at step 712.
  • the recognition arbitration in step 718 sorts the voice recognition results according to the voice quality carried by the network voice recognition results and the local voice recognition results, so as to select a better voice recognition result to generate the final voice recognition result. For example, a specified number of speech recognition results with higher sound quality may be selected for weighting to obtain the final recognition result.
  • device A, device B, and device C are powered on in sequence. After device A is powered on first, it finds that there is no group network, so it creates a group network. After device B and device C are powered on and find that a group network already exists, they join the group network. After device B and device C are connected to the group network, first update the function points of other devices in the network (ie device A) whether there is any modification, and at the same time put the function points of this device (ie device B and device C) in the group Broadcast to inform other node devices of other access groups of their own input and output functions.
  • the user is between device B and device C and makes a voice signal.
  • Device B and Device C collect the audio signal sent by the user.
  • Device A cannot collect the audio signal sent by the user because it is far away from the user and exceeds the pickup distance of the sound collection module of Device A.
  • the device B and the device C preprocess the received audio signal, and the obtained preprocessing result contains at least sound feature information of the collected audio signal that can be applied to the speech recognition model.
  • the preprocessing results also contain information such as the signal-to-noise ratio and amplitude of the collected sound, which can be used to assess sound quality.
  • the preprocessing result also contains the incremental sequence number of the audio signal. Taking device B as an example, the preprocessed data sent by it includes N pieces of data, and the N pieces of data all have increasing serial numbers.
  • the preprocessing result also includes start time information and end time information, and the start time information is used to distinguish different sound information.
  • Device B and device C preprocess the collected speech, obtain relevant preprocessing information, and send it to the network.
  • the communication module of device A also receives the sound preprocessing results from device B and device C, respectively, from the network.
  • the communication module of device B receives the sound preprocessing result from device C from the network.
  • the communication module of device C receives the sound preprocessing result from device B from the network.
  • the signal quality of the device B has exceeded the predetermined threshold.
  • the quality of the collected audio signal in the second sound preprocessing result obtained from the device C and received by the device B through the network is better.
  • device B selects the sound preprocessing result with the highest sound quality (ie, the second sound preprocessing result of device C here) for subsequent speech recognition.
  • the signal quality of device B has exceeded a predetermined threshold based on the first sound preprocessing result obtained by its own sound acquisition module and sound preprocessing module. Even if the quality of the collected audio signal in the second sound preprocessing result obtained from device C is better, device B will still use its own sound collection module and the preprocessed signal obtained by the sound preprocessing module to perform subsequent speech recognition.
  • device A is a television
  • device B is an air conditioner
  • device C is a desk lamp.
  • Devices A, B, and C can support a part of common command information.
  • these three devices jointly support general-purpose command information with command values ranging from 000 to 099, such as the wake-up command "Hello A”.
  • the three devices also support different types of commands. For example, device A supports the instruction information "increase TV volume (111)", while device B and device C do not support this instruction.
  • the device A does not collect the audio signal through the sound collection module of the device A because the distance from the user exceeds the pickup distance. But device A receives the second sound preprocessing result from device B and device C through the network. Device A selects the second sound preprocessing result with the highest sound quality for subsequent speech recognition according to the sound quality ranking in the second sound preprocessing result.
  • the device B performs speech recognition to obtain the first speech recognition result as "Hello A", and determines that the device identification in the first speech recognition result is device A, so the The first speech recognition result is forwarded to other devices (ie, device A and device C) in the network.
  • the device A performs speech recognition to obtain the first speech recognition result as "Hello A", and the device identification in the first speech recognition result is determined to be the device A, so it is not The first speech recognition result is then forwarded to other devices in the network.
  • device A obtains the first speech recognition result (“Hello A”) based on its own sound processing module.
  • Device A receives a second speech recognition result (“Hello C") from device B over the network.
  • Device A receives another second speech recognition result (“Hello A”) from device C over the network.
  • a weighting process is performed for the three speech recognition results, wherein the weighting is assigned considering two factors, ie, the sound quality in the recognition results and the source device of the recognition results. The higher the sound quality, the greater the weight assigned to the corresponding speech recognition result. If the source device of the identification result is this device, the higher the weight is assigned.
  • device A in this example assigns different weight values (B: 0.6, C: 0.4) to the second speech recognition results received from device B and device C according to the sound quality.
  • the speech recognition result is given a higher weight value (A: 0.8). Therefore, the final weighted result for these three speech recognition results is "Hello A”: 1.2, “Hello C”: 0.6. And thus the final speech recognition result is "Hello A”.
  • device A determines the temporal validity of the speech recognition result based on the final speech recognition result ("Hello A"). If it is found that it is still within the valid time range, it is further judged whether the device identification is the device, and the device identification in the speech recognition result is found to be "A", that is, the device, so the operation corresponding to the instruction information is performed. At the same time, device A sends feedback information to other devices (i.e., device B and device C) in the network. The feedback information includes at least the identification time, the identification result and the maximum value of the incremented sequence number. Device B and Device C receive the feedback information and know that the recognition result has been completed. Therefore, Device B and Device C respectively stop the voice recognition work and the sending work of their own devices.
  • the sound preprocessing result includes sound feature value, sound quality information, and sound time information.
  • the sound feature value is the MFCC feature value or the PLP feature value of the audio signal, that is, the sound feature value obtained by the MFCC algorithm, and is used to represent the content of the speech information.
  • the sound quality information includes the signal-to-noise ratio and amplitude of the audio signal.
  • the sound time information includes the start time and end time of the audio signal, or the start time and duration of the audio signal.
  • the sound preprocessing result also contains an incrementing sequence number.
  • the first sound preprocessing data sent by it may include N pieces of data, wherein each piece of data has an increasing sequence number.
  • the sound time information may include start time information and/or end time information of the audio signal to distinguish different sound information.
  • Device B and Device C send the obtained sound preprocessing results to all devices in the network.
  • Device A receives the sound preprocessing results from device B and device C respectively through the network.
  • Device B receives the sound preprocessing result sent by device C through the network.
  • Device C receives the sound preprocessing result sent by device B through the network.
  • Device A prioritizes the voice preprocessing information sent by device B and the voice preprocessing information sent by device C received from the network, so as to select a voice preprocessing result for subsequent speech recognition.
  • Device B prioritizes the locally obtained voice preprocessing information and the voice preprocessing information received from the network and sent from device C to select a voice preprocessing result for subsequent voice recognition.
  • Device C prioritizes the locally obtained sound preprocessing information and the sound preprocessing information received from the network and sent from device B to select a sound preprocessing result for subsequent speech recognition.
  • the solution of the present invention can be applied to a continuous speech recognition scenario.
  • a continuous speech recognition scenario referring also to Figure 2, it is assumed that the user walks from device B to device C while uttering the voice command "turn on the kitchen light".
  • device A is a kitchen lamp
  • the voice collected by device B is "turn on the kitchen”
  • the voice collected by device C is "room light”.
  • Devices B and C preprocess the voice information, and the feature information corresponding to the preprocessing information "Open Kitchen” obtained by device B and the preprocessing information described above; the features corresponding to the preprocessing information "room lamp” obtained by device C information and the preprocessing information described earlier. And both B and C send their respective preprocessing information to group group1.
  • the preprocessed data may include N blocks, and each of the N blocks of data has an increasing sequence number.
  • the duration of each block of data may be, for example, 30 ms, and each block of data is assigned an increasing sequence number.
  • the incrementing sequence number is related to the point in time when the device completes the preprocessing. It is generally believed that if the user issues a voice command indoors, the time points when each device (A, B, C) in the network completes the preprocessing is close). For different devices, the sequence numbers of the preprocessed data blocks are the same at the same/close time point (actually there may be a time difference of about 10ms).
  • the device will send each data block to the group after processing each data block. After receiving each data block, the device will select the optimal data block from the data blocks with the same serial number (the method is similar to preprocessing sorting), and The data blocks with different serial numbers are spliced to form a complete preprocessing result.
  • the voice command "Hello A” can be divided into 10 pieces of data (serial number 000-009) after voice preprocessing, device A will select a data block from the received data blocks with serial number 000; A data block is selected from the received data blocks with a sequence number of 000; ...; and a final preprocessing result (the first preprocessing result) is formed by splicing the plurality of data blocks in the sequence of the sequence numbers.
  • each device can only collect part of the voice information.
  • device A may collect the serial number of For data blocks of 000-006, device B can collect data blocks with serial numbers of 003-009.
  • both devices A and B can receive data blocks of 000-009, so that the voice data (preprocessing) can be completed. result) splicing.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Human Computer Interaction (AREA)
  • Physics & Mathematics (AREA)
  • Acoustics & Sound (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Telephonic Communication Services (AREA)

Abstract

提供一种分布式语音处理系统及方法。该系统包括：网络中的多个节点设备，其中每个节点设备包括处理器、存储器、通信模块及声音处理模块，至少一个节点设备包括声音采集模块；其中声音采集模块配置为采集音频信号；声音处理模块配置为对音频信号进行预处理以得到第一声音预处理结果；通信模块配置为将第一声音预处理结果发送到网络中的一个或多个节点设备；通信模块还配置为从网络接收来自至少一个其他节点设备的一个或多个第二声音预处理结果；及声音处理模块还配置为基于第一声音预处理结果和/或一个或多个第二声音预处理结果执行语音识别。本申请实现了去中心化的设计,提高了语音识别的效率、稳定性和鲁棒性。

Description

分布式语音处理系统及方法 技术领域
本申请涉及分布式语音处理领域，尤其涉及一种分布式语音处理系统及方法。
背景技术
语音识别技术和关键词识别技术日趋成熟，且在市场中的应用越来越广泛，比如儿童玩具、教育产品、智能家居等产品中都加入了语音识别功能，实现语音交互控制的功能。
当前的语音识别有两种常见的方法，一种是基于单个设备的本地识别，另一种是基于本地识别结合服务器云端的识别。
对于第一种方法，智能家居市场上常用的单个设备实现本地语音控制，其语音识别过程是将原始语音采集到一个设备，在这个设备上进行计算获得识别结果。这种方法在应用过程中，如果用户在空间环境中移动或需要跨房间识别时，受拾音距离的限制，常常不能顺利完成语音识别，从而导致出现不能识别或识别效果差的情况。
对于第二种方法，市面上通常通过智能音箱或者智能网关等进行语音识别。这些设备作为控制中枢，同时也是语音识别的唯一入口。设备必须先连接到互联网，然后接入相应的云端服务器也需要连接到互联网。这些设备从云端获取语音识别结果，进而完成语音识别或语音控制。这种方法存在问题，例如作为语音识别的唯一入口的设备发生故障、或者网络出现波动等，都会造成语音识别失败的问题。特别是，当网络稳定性较差时，容易导致识别响应速度慢。此外，此类语音识别方法将语音上传至云端，并且设备需要实时监听周围环境声音，容易导致出现用户隐私安全问题。
与此同时，这两种方法都存在不能跨房间的语音识别控制等问题。
除了上述两种方法之外，还存在另外一种本地中心识别方法，其将多个点的原始语音采集后，传输到一个中心设备进行语音识别。该方法可以 用于解决拾音距离短、跨房间识别难、人员移动等问题。但这个方法严重依赖中心设备，当中心设备出现故障时会导致整个系统的语音识别功能失效。并且，由于直接传输原始语音数据对网络要求很高，数据传输的时延大，其实际的识别效果欠佳。
中国专利(CN111415658A)公开了一种去中心化语音控制多设备系统及其控制方法。在该方案中，设备首先识别出语音中的唤醒词，然后将识别出的唤醒词向系统内所有设备发送，并同时接收系统内其他设备发送的唤醒词。设备对所有的唤醒词进行筛选，筛选出与本设备匹配的唤醒词。在该方案中，如果设备接收到的语音中包含自身不支持的唤醒词(即语音指令)，则可能导致语音控制失败。
中国专利(CN110136708A)公开了一种基于蓝牙mesh的分布式语音控制系统及控制方法，该控制系统包括蓝牙Mesh网络、语音控制器、蓝牙节点设备；语音控制器包括语音采集，语音降噪，语音识别，蓝牙模块，及可选的wifi模块。语音控制器之间通过蓝牙互相通信并实时保持数据同步，任一语音控制器都可控制网络内蓝牙节点设备；蓝牙节点设备通过蓝牙Mesh网络与语音控制器通信，根据接收Mesh数据或其自身的按键事件进行响应操作。在该方案中，每一语音控制器采集语音，进行语音降噪和回音消除，然后进行语音本地或者在线识别，语义理解解析出要控制的信息，封装成Mesh数据，通过蓝牙模块发送至Mesh网络中。如果语音控制器不支持当前的控制指令，则可能会导致设备无法识别自己不支持的语音指令，最终导致语音控制失败。
综上，现有技术中需要一种改进的分布式语音处理方案，以解决现有技术中存在的上述问题。应理解，上述所列举的技术问题仅作为示例而非对本发明的限制，本发明并不限于同时解决上述所有技术问题的技术方案。本发明的技术方案可以实施为解决上述或其他技术问题中的一个或多个。
发明内容
为克服现有技术存在的缺陷，本发明公开了一种分布式语音处理系统及其处理方法。
在本发明的一方面，提供一种分布式语音处理系统，包括多个节点设备，所述多个节点设备组成网络，其中每个节点设备包括处理器、存储器、通信模块以及声音处理模块，且所述多个节点设备中的至少一个节点设备包括声音采集模块；其中，
所述声音采集模块配置为采集音频信号；
所述声音处理模块配置为对所述音频信号进行预处理以得到第一声音预处理结果；
所述通信模块配置为将所述声音预处理结果发送到所述网络中的一个或多个节点设备；
所述通信模块还配置为从所述网络接收来自至少一个其他节点设备的一个或多个第二声音预处理结果；及
所述声音处理模块还配置为基于所述第一声音预处理结果和/或所述一个或多个第二声音预处理结果执行语音识别以得到第一语音识别结果。
优选地，所述通信模块还配置为将所述第一语音识别结果发送到所述网络中的一个或多个节点设备；
优选地，所述通信模块还配置为从所述网络接收来自至少一个其他节点设备的一个或多个第二语音识别结果；及
优选地，所述声音处理模块还配置为基于所述第一语音识别结果和所述一个或多个第二语音识别结果执行语音识别以得到最终语音识别结果。
在本发明的另一方面，提供一种分布式语音处理系统的处理方法，由网络中的节点设备执行，包括：
若所述节点设备包括声音采集模块，则执行下述步骤：采集音频信号；对所述音频信号进行预处理以得到第一声音预处理结果；以及将所述声音预处理结果发送到所述Mesh网络中的一个或多个节点设备；
从所述Mesh网络接收来自至少一个其他节点设备的一个或多个第二声 音预处理结果；及
基于所述第一声音预处理结果和/或所述一个或多个第二声音预处理结果执行语音识别。
优选地，所述分布式语音处理系统的处理方法还包括将所述第一语音识别结果发送到所述网络中的一个或多个节点设备；
从所述网络接收来自至少一个其他节点设备的一个或多个第二语音识别结果；及
基于所述第一语音识别结果和所述一个或多个第二语音识别结果执行语音识别以得到最终语音识别结果。
本申请提供的方案可以在不接入互联网的情况下，扩展识别距离，提高人员移动情况下的识别率，轻松实现跨房间语音控制。同时，还能更近一步让语音识别更贴近用户习惯，更适应生活使用场景。
此外，本发明通过网络中的节点设备进行分布式地语音识别，可以实现超远距离或跨多个房间的语音识别控制。本发明的技术方案使得网络中的每个节点设备都参与到语音识别的过程中，一方面，实现了去中心化的设计，从而减少因关键中心节点故障导致的识别失败，并且该设计可以实现网络中的设备并发式地进行语音识别，可以提高语音识别的效率；另一方面，在识别过程中传输的信息是声音预处理信息，即非原始音频数据，因此对网络的带宽要求不高，提高了语音识别的稳定性。通过传输非原始音频数据，可以产生两个优点：首先，相对于直接传输原始数据的语音识别方法，本发明中的方案所需传输的数据量下降；其次，相比于直接传输识别结果的语音识别方法，本发明中的方案传输的是声音预处理结果，可以避免因为不支持指令而导致的识别失败，提高了语音识别的稳定性和鲁棒性。
附图说明
在下文中，将基于实施例参考附图进一步解释本申请。
图1示意性地示出根据本发明的分布式语音处理系统的一个实施例的框 图；
图2示意性地示出根据本发明的分布式语音处理系统的另一实施例的框图；
图3示意性地示出根据本发明的分布式语音处理系统的一个实施例的节点设备的框图；
图4示意性地示出根据本发明的分布式语音处理方法的一个实施例的流程图；
图5示意性地示出根据本发明的分布式语音处理方法的另一实施例的流程图；
图6示意性地示出根据本发明的分布式语音处理方法的一个具体实施例的流程图；以及
图7示意性地示出根据本发明的分布式语音处理方法的另一具体实施例的流程图。
具体实施方式
以下将结合附图和具体的实施方式，对本申请的分布式语音识别处理系统及其处理方法进行详细说明。应理解，附图所示以及下文所述的实施例仅仅是说明性的，而不作为对本申请的限制。此外应理解，在本公开中，例如“第一”、“第二”、“第三”等序数词，除非明确指明或由技术上下文确定，仅用于指示技术方案中的不同或相同元素，而不意味着对这些元素的顺序或重要性的任何限制。
图1示出了根据本发明的分布式语音处理系统100的一个实施例的框图，该系统包括网络110中的多个节点设备102、104、106以及112、114和116。应理解，网络100可以例如是家庭和/或办公室使用的有线、无线和/或有线无线混合网络，包括但不限于在智能家居场景中通常所采用的无线网络。各个设备之间组成网络110，可以采用有线或者无线的方式进行通信。其中，有线的方式可以采用网线或者电力线载波等通信方式，无线的方式可以采用Wi-Fi，BLE，Zigbee等通信方式，实现各个设备之间的组网通信。
在一个具体实施例中，各个节点设备均具有连接其他节点设备的能力。各个节点设备之间可以进行自组网，构成自组织网络或者群组网络。各个设备之间还可以组成Mesh网络，可以实现Mesh网络中的任何设备节点都可同时作为路由器，即网络中的每个节点都能发送和接收信号，每个节点都能与一个或多个对等节点进行直接通信。
图2示意性地示出根据本发明的分布式语音处理系统200的另一实施例的框图，其中部分节点设备形成群组，从而本发明的系统也可以按广播或多播方式向群组发送消息。应理解，节点设备可以处于一个或多个群组中，且群组可以是动态的、可由用户自定义的，而不必要求群组之间的节点设备必须具有固定的硬件或通信连接关系。
在图1和图2所示的系统中，用户与节点设备间可能具有不同距离。例如用户108处于设备B和设备C之间，且处在设备B和设备C的拾音距离之内，但是用户108离设备A和其他设备较远，无法通过设备A和其他设备直接接收用户108发出的语音信号。
图3示意性地示出根据本发明的分布式语音处理系统的一个实施例的节点设备300的框图。如图3所示，每个节点设备300可以包括处理器302、存储器304、通信模块306以及声音处理模块312。多个节点设备中的至少一个节点设备包括声音采集模块308。可选地，节点设备300还可以包括输出模块312，其中处理器302可以提供μs级精准的时钟；通信模块306可以采用有线(比如，网线/电力线载波等)或者无线(比如，Wi-Fi/BLE/Zigbee等)的任何方式，用于与其他设备进行组网通信；存储器304可以记录组网信息和识别模型参数；输出模块312例如可以是扬声器、开关量装置等；声音采集模块308例如可以是单个麦克风、多个麦克风，或者是麦克风阵列。
声音采集模块308可以配置为采集音频信号。声音处理模块310可以配置为对音频信号进行预处理以得到本地产生的声音预处理结果。通信模块306可以配置为将本地产生的声音预处理结果发送到网络110中的一个或多个节点设备。通信模块306还可以配置为从网络110接收来自至少一个其他 节点设备的一个或多个声音预处理结果。应理解，在本申请的上下文中，本地产生的声音预处理结果可以称为“第一声音预处理结果”，通过网络从其他节点设备接收的声音预处理结果可以称为“第二声音预处理结果”。声音处理模块310还可以配置为基于本地产生的声音预处理结果和/或一个或多个通过网络110接收的声音预处理结果执行语音识别。以此方式，节点设备300可以得到本地产生的语音识别结果。
节点设备的声音处理模块执行的语音识别可以包括但不限于唤醒词检测、关键词识别、连续语音识别等。作为非限制性示例，节点设备的声音处理模块执行语音识别得到的语音识别结果可以包括设备标识，识别结果，识别结果的有效时间，识别开始时间，声音质量。第一语音识别结果还可以包括指令信息和设备标识，以指示目标设备进行相应的操作。
在一个或多个实施例中，本发明的分布式语音识别方案一方面可以利用本地产生和来自网络的声音预处理结果，另一方面还可以利用本地产生和来自网络的语音识别结果。
在本发明的一个实施例中，本发明的节点设备可以对不同来源的语音识别结果进行仲裁。作为非限制性示例，通信模块306还可以配置为将本地产生的语音识别结果发送到网络中的一个或多个节点设备。通信模块306还配置为从网络接收来自至少一个其他节点设备的一个或多个语音识别结果。应理解，在本申请的上下文中，本地产生的语音识别结果可以称为“第一语音识别结果”，通过网络从其他节点设备接收的语音识别结果可以称为“第二语音识别结果”。声音处理模块310还配置为基于第一语音识别结果和一个或多个第二语音识别结果执行语音识别以得到最终语音识别结果。此外，在一个具体实施例中，声音处理模块310还配置为对第一语音识别结果和一个或多个第二语音识别结果执行加权处理以得到最终语音识别结果。
例如，节点设备的声音处理模块可以配置为基于第一语音识别结果和一个或多个第二语音识别结果的声音质量分配权重，声音质量越高则分配的权重越大。又如，节点设备的声音处理模块可以配置为基于第一语音识别结果和一个或多个第二语音识别结果的来源设备分配权重，如果来源设 备为本节点设备，则分配的权重越大。
在本发明的上下文中，声音预处理结果是从原始语音到语音识别结果的识别过程中产生的中间结果。在一个具体实施例中，第一声音预处理结果和一个或多个第二声音预处理结果的每个包括声音特征值、声音质量及声音时间信息。在一个具体实施例中，节点设备的通信模块306从网络接收来自至少一个其他节点设备的一个或多个第二声音预处理结果，其中第二声音预处理结果包括声音特征值、声音质量及声音时间信息，还可以包括音频信号的递增序号。其中，预处理结果中的声音特征值为音频信号的MFCC特征值或PLP特征值。声音质量可以包括音频信号的信噪比和幅度。声音时间信息可以包括音频信号的开始时间和结束时间。声音时间信息可以包括音频信号的开始时间和持续时间。本领域技术人员应理解，本发明的实施不限于此。相反，本领域技术人员可以基于现有以及将来开发出的语音识别和处理技术，可以采用任何合适的声音预处理结果以实现本发明的方案。
本领域技术人员应理解，适用于本发明的预处理技术可以包括但不限于信号分帧、预加重、快速傅里叶变换(Fast Fourier Transform,FFT)等预处理。预处理方法可以根据音频信号获得音频参数、生成频域信号或进行梅尔频率倒谱特征(Mel-Frequency Cepstral Coefficients，MFCC)算法或感知线性预测(Perceptual Linear Predictive，PLP)算法提取，用于表征语音信息的内容。
在一个具体实施例中，节点设备的声音处理模块310还配置为对于第一声音预处理结果和一个或多个第二声音预处理结果中的每个，判断其声音质量是否超过预定阈值，若否，则丢弃该语音预处理结果。
在一个具体实施例中，节点设备的声音处理模块310还配置为在第一声音预处理结果和一个或多个第二声音预处理结果中，选取一个或多个声音质量最高的声音预处理结果进行语音识别以得到第一语音识别结果。
作为示例而非限制，节点设备的声音处理模块执行语音识别得到的第一语音识别结果可以包括指令信息，其中，指令信息是具体数值，如011， 该指令信息例如可由支持相应指令的节点设备所理解和执行。此外，节点设备的声音处理模块执行语音识别得到的第一语音识别结果可以包括指令信息，其中，不同的节点设备支持不同范围的指令信息。
此外，节点设备的声音处理模块还可以配置为在第一声音预处理结果和一个或多个第二声音预处理结果中，选取一个或多个声音质量最高的声音预处理结果进行语音识别以得到第一语音识别结果。
在一个实施例中，节点设备的声音处理模块可以判断第一声音预处理结果的声音质量是否超过预定阈值，若超过，则选择第一声音预处理结果进行语音识别以得到第一语音识别结果。
作为示例实施例，节点设备的通信模块可以通过单播、多播和/或广播的方式将第一语音识别结果发送到网络中的一个或多个节点设备。
在一个实施例中，对于第一语音识别结果中的设备标识与本节点设备的设备标识不一致的第一语音识别结果，通过节点设备的通信模块将该第一语音识别结果发送到网络中的一个或多个节点设备。相反，对于第一语音识别结果中的设备标识与本节点设备的设备标识一致的第一语音识别结果，则不将该第一语音识别结果发送到网络中的一个或多个节点设备。
在一个实施例中，节点设备的声音处理模块可以对得到的最终语音识别结果进行时间有效性判断，如果超过识别结果的有效时间，则不执行识别结果所对应的相应操作。此外，节点设备的声音处理模块可以对于得到的最终语音识别结果判断设备标识，如果设备标识为本节点设备，则执行识别结果所对应的相应操作。
作为非限制性示例，节点设备的声音处理模块可以对于得到的最终语音识别结果判断设备标识，如果设备标识为本节点设备，则输出反馈信息并通过通信模块发送到网络中的其他一个或多个节点设备。
此外，节点设备的声音处理模块可以对于得到的最终语音识别结果判断设备标识，如果设备标识为本节点设备，则输出反馈信息，其中输出反馈信息至少包括识别时间、识别结果和递增序号的最大值。
图4示意性地示出根据本发明的分布式语音处理方法400的一个实施例 的流程图。该分布式语音处理方法由网络中的节点设备执行。在步骤402，判断节点设备是否包括声音采集模块。如果是，则转到步骤404。如果否，则转到步骤410。在步骤404，采集音频信号。在步骤406，对音频信号进行预处理以得到第一声音预处理结果。在步骤408，将第一声音预处理结果发送到网络中的一个或多个节点设备。在步骤410，从网络接收来自至少一个其他节点设备的一个或多个第二声音预处理结果。在步骤412，基于第一声音预处理结果和/或一个或多个第二声音预处理结果执行语音识别。
在一个实施例中，可以对第一声音预处理结果和一个或多个第二声音预处理结果中的每个，判断其声音质量是否超过预定阈值，若否，则丢弃该语音预处理结果。
在另一个实施例中，可以在第一声音预处理结果和一个或多个第二声音预处理结果中选取一个或多个声音质量最高的声音预处理结果进行语音识别以得到第一语音识别结果。
作为非限制性示例，本发明的方法可以结合本地识别结果与来自网络的识别结果以得到最终语音识别结果。例如，节点设备可以将第一语音识别结果发送到网络中的一个或多个节点设备。节点设备可以从网络接收来自至少一个其他节点设备的一个或多个第二语音识别结果。节点设备可以基于第一语音识别结果和一个或多个第二语音识别结果执行语音识别以得到最终语音识别结果。
在另一个具体实施例中，可以对第一语音识别结果和一个或多个第二语音识别结果执行加权平均以得到最终语音识别结果。
在一个或多个实施例中，本发明的方案还利用本地产生的以及来自网络的声音预处理结果的片段进行拼接以获得。
在一个实施例中，提供了一种分布式语音处理系统，包括：多个节点设备，该多个节点设备组成网络，其中每个节点设备包括处理器、存储器、通信模块以及声音处理模块，且所述多个节点设备中的至少一个节点设备包括声音采集模块；其中声音采集模块配置为采集音频信号；声音处理模块配置为对音频信号进行预处理以得到第一声音预处理结果；其中，通信 模块还配置为从该网络接收来自至少一个其他节点设备的一个或多个第二声音预处理结果；该第一声音预处理结果和该一个或多个第二声音预处理结果中的每个包括一个或多个数据块；该一个或多个数据块中的每个数据块包括时间信息，该时间信息标识该声音处理模块完成该数据块预处理的时间；该一个或多个数据块中的每个数据块还包括递增序号，该递增序号依据数据块中的时间信息分配；该声音处理模块还配置为按照递增序号递增顺序拼接该第一声音预处理结果和/或一个或多个第二声音预处理结果的数据块，以得到完整的第三声音预处理结果；以及该声音处理模块还配置为对该第三声音预处理结果进行处理以得到最终的语音识别结果。
在一个实施例中，通信模块配置为将第一声音预处理结果发送到该网络中的一个或多个节点设备。
在一个实施例中，第一和/或第二声音预处理结果中的每个数据块配置为具有相同的时长。
在一个实施例中，递增序号是多个节点设备中的每个节点设备的声音处理模块对该音频信号进行预处理时为数据块分配的。
在一个实施例中，递增序号是多个节点设备中的每个节点设备的声音处理模块对从网络接收来自至少一个其他节点设备的第二声音预处理结果后，为第二声音预处理结果的数据块分配的。
在一个实施例中，声音处理模块配置为检测数据块的时间差，如果时间差在指定阈值以内，则分配相同的递增序号。
在一个实施例中，声音处理模块配置为从相同递增序号的数据块中选择声音质量最优的数据块进行拼接。
图5示意性地示出根据本发明的分布式语音处理方法500的另一实施例的流程图。该方法由网络中的节点设备执行。在步骤502，判断节点设备是包括声音采集模块。如果是，则转到步骤504。如果否，则转到步骤508。在步骤504，采集音频信号。在步骤506，对该音频信号进行预处理以得到第一声音预处理结果。在步骤508，从网络接收来自至少一个其他节点设备的一个或多个第二声音预处理结果，其中第一声音预处理结果和一个或多 个第二声音预处理结果中的每个包括一个或多个数据块，其中一个或多个数据块中的每个数据块包括时间信息，该时间信息标识声音处理模块完成数据块预处理的时间，且其中一个或多个数据块中的每个数据块还包括递增序号，该递增序号依据数据块中的时间信息分配。在步骤510，按照递增序号递增顺序拼接第一声音预处理结果和/或一个或多个第二声音预处理结果的数据块，以得到完整的第三声音预处理结果。在步骤512，对第三声音预处理结果进行处理以得到最终的语音识别结果。
应理解，在本发明的上下文中，将第一声音预处理结果和/或一个或多个第二声音预处理结果的数据块拼接所得的声音预处理结果称为“第三声音预处理结果”。
在一个实施例中，该分布式语音处理方法还包括，将第一声音预处理结果发送到网络中的一个或多个节点设备。
在一个实施例中，第一声音预处理结果中的每个数据块配置为具有相同的时长。
在一个实施例中，递增序号是多个节点设备中的每个节点设备的声音处理模块对音频信号进行预处理时为数据块分配的，其中递增序号根据时间信息分配。
在一个实施例中，递增序号是多个节点设备中的每个节点设备的声音处理模块对从网络接收来自至少一个其他节点设备的第二声音预处理结果后，为第二声音预处理结果的数据块分配的，其中递增序号根据时间信息分配。
在一个实施例中，声音处理模块配置为检测数据块的时间差，如果时间差在阈值以内，则分配相同的递增序号。
在一个实施例中，声音处理模块配置为从相同递增序号的数据块中选择声音质量最优的数据块。
图6示意性地示出根据本发明的分布式语音处理方法600的一个具体实施例的流程图。在该实施例中，对节点设备进行自组网，建立群组，让每台组网的节点设备都进行语音识别，在群组内交换识别信息的方法，将原 本由一个节点设备的语音识别系统改进为分布在多个节点设备上的语音识别系统，从而解决了语音识别、关键词识别和语音控制场景中，依赖单一控制中心，依赖网络服务器，不能跨区域，隐私信息不安全等问题。
在步骤604，节点设备上电的时候，发现是否存在群组网络。如果不存在群组网络，则在步骤606创建群组网络。如果已经存在群组网络，则在步骤608加入该群组网络。节点设备在接入群组网络后，先在步骤610更新网络中其他设备功能点，以获知网络中其他设备所支持的功能点是否有修改，同时或之后，在步骤612将自身的设备功能点在群组网络中进行广播。应理解在本发明的上下文中，“功能点”用于告知其他接入群组的其他节点设备自己具有哪些方面的输入和输出功能。应理解，在本发明的上下文中，“群组网络”是指在其中支持广播和/或多播的节点设备所构成的网络，包括但不限于具有各种拓扑结构(例如Mesh拓扑结构)的Wi-Fi、BLE和ZigBee网络，并且可以是有线、无线或混合网络。
在步骤614，节点设备通过分布式识别获取识别结果，并在步骤616判断该识别结果的设备标识是否为本设备。如果设备标识不是本设备，则在步骤622发送识别信息。作为非限制性示例，识别信息可以包括识别设备标识、识别时间、识别结果、识别结果的可信度。如果设备标识是本设备，则在步骤618执行输出，并随后在步骤620将执行结果信息发送到网络中的其他节点设备。作为非限制性示例，执行结果信息可以包括设备标识、识别时间、识别结果，执行结果等。
图7示意性地示出根据本发明的分布式语音处理方法700的另一具体实施例的流程图。如图7所示，该实施例中的分布式语音处理方法包括三个输入和一个输出，三个输入分别是：在步骤702通过本地麦克风采集的声音，在步骤708从网络收集的声音预处理信息，以及在步骤714从网络收集的语音识别信息；一个输出是在步骤720输出的语音识别结果。
在该实施例中，分布式语音处理方法700分为三个阶段：预处理阶段、分析决策阶段，以及识别仲裁阶段。
在预处理阶段，本地麦克风采集的声音会先经过步骤704的声音预处理， 获得预处理信息，然后在步骤706将预处理信息发送到群组网络。预处理信息例如包含有所采集声音的可用于识别模型辨识的特征信息。预处理信息例如还包含有所采集声音的信噪比和幅度等可用于评估声音质量的信息。预处理信息例如还包含有该信息的递增序号。作为示例而非限制，预处理信息还可以包含开始时间信息和结束时间信息。
在步骤710的分析决策阶段，对网络收集的声音预处理信息和本地获取到的声音预处理信息的声音质量进行排序，分析筛选出最优质的预处理信息送入后续的语音识别步骤712。在步骤712，执行语音识别以输出本地识别信息。本地或者网络识别信息例如可以包括但不限于识别的结果、语音识别的设备标识、识别结果的有效时间、识别开始时间和声音质量中的译者或多者。
在识别仲裁阶段，先针对在步骤714从网络收集到的识别信息在步骤716进行分析判断，根据识别信息的时效去掉过期的信息。然后与步骤712的本地语音识别的输出一起在步骤718进行识别仲裁。步骤718的识别仲裁根据网络语音识别结果和本地语音识别结果所携带的声音质量进行排序，以选取较佳的语音识别结果来产生最终的语音识别结果。例如，可以选取指定数量的几个声音质量较稿高的语音识别结果进行加权以获取最后的识别结果。
下面通过场景示例，进一步说明本发明的原理。在第一个场景中，参考图2所示，其中设备A、设备B、设备C依次上电。设备A首先上电后，发现不存在群组网络，因此创建群组网络。设备B和设备C上电后，发现群组网络已存在，则加入该群组网络。设备B和设备C在接入群组网络后，先更新网络中其他设备的功能点(即设备A)是否有修改，同时将本设备(即设备B和设备C)的功能点在群组里面进行广播，告知其他接入群组的其他节点设备自己具有的输入和输出功能。
用户处于设备B和设备C之间并发出语音信号。设备B和设备C采集到用户发出的音频信号，设备A因为与用户的的距离较远，超出设备A的声音采集模块的拾音距离，因此不能采集到用户发出的音频信号。
设备B和设备C对接收到的音频信号进行预处理，所获得的预处理结果至少包含所采集音频信号的可应用于语音识别模型的声音特征信息。预处理结果还包含所采集声音的信噪比和幅度等可用于评估声音质量的信息。预处理结果还包含有该音频信号的递增序号。以设备B为例，其发送的预处理数据包括N块数据，N块数据均带有递增的序号预处理结果还包含开始时间信息和结束时间信息，开始时间信息用于区别不同的声音信息。
设备B和设备C将采集到的语音进行预处理，获得相关的预处理信息发送到网络中。设备A的通信模块还从网络中接收分别来自设备B和设备C的声音预处理结果。设备B的通信模块从网络中接收来自设备C的声音预处理结果。设备C的通信模块从网络中接收来自设备B的声音预处理结果。
设备B基于自身的声音采集模块和声音预处理模块所获得的第一声音预处理结果，其信号质量已超过预定阈值。但设备B通过网络接收到的从设备C处获得的第二声音预处理结果中的采集的音频信号质量更好。设备B根据第一和第二声音预处理结果中的声音质量，选择声音质量最高的声音预处理结果(即，此处为设备C的第二声音预处理结果)进行后续的语音识别。
在图2所示的场景中，例如另一情形下，设备B基于自身的声音采集模块和声音预处理模块所获得的第一声音预处理结果，其信号质量已超过预定阈值。即使从设备C处获得的第二声音预处理结果中的采集的音频信号质量更好，设备B依然会使用自身的声音采集模块和声音预处理模块获得的预处理信号进行后续的语音识别。
在图2所示的场景中，例如另一情形下，假设设备A是电视，设备B是空调，设备C是台灯。设备A、B、C可以支持一部分共同的指令信息。例如，这三个设备共同支持指令数值范围为000-099的通用型指令信息，如唤醒指令“Hello A”。此外，这三个设备也支持不同类型的指令。例如设备A支持指令信息“提高电视音量(111)”，而设备B和设备C则不支持该指令。
在图2所示的场景中，例如另一情形下，设备A因为距离用户超过拾音距离，没有通过设备A的声音采集模块采集到音频信号。但是设备A通过网络接收到来自设备B和设备C的第二声音预处理结果。设备A根据第二声音预处理结果中的声音质量排序，选择声音质量最高的第二声音预处理结果进行后续的语音识别。
在图2所示的场景中，例如一种情形下，设备B进行语音识别获得第一语音识别结果为“Hello A”，判别该第一语音识别结果中的设备标识为设备A，因此将该第一语音识别结果转发至网络中的其他设备(即设备A和设备C)。
在图2所示的场景中，例如另一种情形下，设备A进行语音识别获得第一语音识别结果为“Hello A”，判别该第一语音识别结果中的设备标识为设备A，因此不再将该第一语音识别结果转发至网络中的其他设备。
在图2所示的场景中，例如，设备A基于自身的声音处理模块获得第一语音识别结果(“Hello A”)。设备A通过网络从设备B处接收到第二语音识别结果(“Hello C”)。设备A通过网络从设备C处接收到另一个第二语音识别结果(“Hello A”)。对于这三个语音识别结果执行加权处理，其中权重的分配考虑两个因素，即识别结果中的声音质量和识别结果的来源设备。声音质量越高，对应的语音识别结果分配的权重则越大。识别结果的来源设备为本设备，则分配的权重越大。例如本示例中的设备A，根据声音质量，对从设备B和设备C接收的第二语音识别结果分别赋予不同的权重值(B：0.6，C：0.4)，对来自设备A本身的第一语音识别结果赋予更高的权重值(A：0.8)。因此，对于这三个语音识别结果最终的加权结果为“Hello A”：1.2，“Hello C”：0.6。并从而得到最终语音识别结果为“Hello A”。
在图2所示的场景中，例如，设备A基于最终语音识别结果(“Hello A”)，判断该语音识别结果的时间有效性。如果发现仍在有效时间范围内，则进一步判断设备标识是否为本设备，发现该语音识别结果中的设备标识为“A”，即本设备，因此执行相应于指令信息的操作。同时，设备A向网 络中的其他设备(即设备B和设备C)发送反馈信息。反馈信息至少包括识别时间、识别结果和递增序号的最大值。设备B和设备C收到反馈信息，获知本次识别结果已经执行完成，因此设备B和设备C分别停止本设备的语音识别工作以及发送工作。
如图2所示，当用户在设备B和设备C之间发出音频信号“Hello A”，设备B和设备C将通过各自的声音采集模块采集到该用户的音频信号。并且，设备B和设备C通过各自的声音处理模块进行预处理以得到声音预处理结果。因其与用户的距离超过声音采集模块的拾音距离，设备A和网络中的其他设备无法通过设备本身的声音采集模块采集到声音。其中，声音预处理结果包含声音特征值，声音质量信息，及声音时间信息。其中，声音特征值是音频信号的MFCC特征值或PLP特征值，即通过MFCC算法获得的声音特征值，用于表征语音信息的内容。声音质量信息包括音频信号的信噪比和幅度。声音时间信息包括音频信号的开始时间和结束时间，或音频信号的开始时间和持续时间。声音预处理结果还包含递增序号。以设备B为例，其发送的第一声音预处理数据可以包括N块数据，其中每块数据均带有递增的序号。声音时间信息可以包括音频信号的开始时间信息和/或结束时间信息，以区别不同的声音信息。
如图2所示，设备B和设备C将获得的声音预处理结果发送到网络中的所有设备。设备A通过网络分别收到设备B和设备C发出的声音预处理结果。设备B通过网络收到设备C发出的声音预处理结果。设备C通过网络收到设备B发出的声音预处理结果。设备A对从网络接收到的设备B发出的声音预处理信息和设备C发出的声音预处理信息进行优先级排序，以选择进行后续语音识别的声音预处理结果。设备B对本地获得的声音预处理信息以及从网络接收到的从设备C发出的声音预处理信息进行优先级排序，以选择进行后续语音识别的声音预处理结果。设备C对本地获得的声音预处理信息以及从网络接收到的从设备B发出的声音预处理信息进行优先级排序，以选择进行后续语音识别的声音预处理结果。
在另一实施例中，本发明的方案可以应用于连续语音识别场景。在该 场景中，同样参考图2，假设用户在发出语音指令“打开厨房灯”的同时从设备B走到设备C。在本实施例中，设备A为厨房灯，设备B采集到的语音是“打开厨”，设备C采集到的语音是“房灯”。
设备B、C对语音信息进行预处理，B得到的预处理信息“打开厨”对应的特征信息以及前面所述的各项预处理信息；设备C得到的预处理信息“房灯”对应的特征信息以及前面所述的各项预处理信息。并B，C均向群组group1发送各自的预处理信息。
如上文所述，预处理数据可以包括N个块，N块数据均带有递增序号，每块数据时长例如可以为30ms，每块数据均分配有递增序号。递增序号与设备完成预处理的时间点有关。一般认为，若用户在室内发出语音指令，网络内的各个设备(A，B，C)完成预处理的时间点是接近的)。对于不同设备而言，相同/相近时间点(实际可能会有10ms左右的时间差)完成预处理的数据块的序列号相同。
在实际应用中，设备处理完每块数据后都会向群组内发送，设备收到各个数据块后会从相同序列号的数据块中选择最优的数据块(方法类似预处理排序)，并将不同序列号的数据块拼接后形成完整的预处理结果。
如语音指令“Hello A”在语音预处理后可以分为10块数据(序列号000-009)，设备A会从收到的多个序列号为000的数据块中选择出一个数据块；从收到的多个序列号为000的数据块中选择出一个数据块；……；并按序列号顺序对多个数据块进行拼接已形成最终的预处理结果(第一预处理结果)。
这样就可以解决人员移动的问题，若用户从发出语音命令的过程中从设备A附近移动到设备B附近，每个设备只能采集到部分语音信息，例如，设备A可以会采集到序列号为000-006的数据块，设备B可以采集到序列号为003-009的数据块，通过上述方法，设备A、B均可以收到到000-009的数据块，从而可以完成语音数据(预处理结果)的拼接。
应理解，以上分布式语音识别系统和方法仅作为示例提供，而非对本发明的限制。本领域技术人员应理解，可以将本发明的原理应用于不同于以上 所述分布式语音识别系统和方法而不脱离本发明的范围。虽然出于本公开的目的已经描述了本申请各方面的各种实施例，但是不应理解为将本公开的教导限制于这些实施例。在一个具体实施例中公开的特征并不限于该实施例，而是可以和不同实施例中公开的特征进行组合。例如，在一个实施例中描述的根据本申请的方法的一个或多个特征和/或操作，亦可单独地、组合地或整体地应用在另一实施例中。关于系统/设备实施例的描述同样适用于方法实施例，反之亦然。本领域技术人员应理解，还存在可能的更多可选实施方式和变型，可以对上述系统进行各种改变和修改，而不脱离由本申请权利要求所限定的范围。

Claims (20)

  1. 一种分布式语音处理系统，其特征在于，包括：
    网络中的多个节点设备，其中每个节点设备包括处理器、存储器、通信模块以及声音处理模块，且所述多个节点设备中的至少一个节点设备包括声音采集模块；
    其中所述声音采集模块配置为采集音频信号；
    所述声音处理模块配置为对所述音频信号进行预处理以得到第一声音预处理结果；
    所述通信模块配置为将所述第一声音预处理结果发送到所述网络中的一个或多个节点设备；
    所述通信模块还配置为从所述网络接收来自至少一个其他节点设备的一个或多个第二声音预处理结果；及
    所述声音处理模块还配置为基于所述第一声音预处理结果和/或所述一个或多个第二声音预处理结果执行语音识别。
  2. 根据权利要求1所述的分布式语音处理系统，其特征在于，所述通信模块还配置为将所述第一语音识别结果发送到所述网络中的一个或多个节点设备；
    所述通信模块还配置为从所述网络接收来自至少一个其他节点设备的一个或多个第二语音识别结果；及
    所述声音处理模块还配置为基于所述第一语音识别结果和所述一个或多个第二语音识别结果执行语音识别以得到最终语音识别结果。
  3. 根据权利要求1所述的分布式语音处理系统，其特征在于，所述第一声音预处理结果和所述一个或多个第二声音预处理结果的每个包括声音特征值、声音质量及声音时间信息。
  4. 根据权利要求3所述的分布式语音处理系统，其特征在于，所述声音特征值为所述音频信号的MFCC特征值或PLP特征值。
  5. 根据权利要求3所述的分布式语音处理系统，其特征在于，所述声音质量包括所述音频信号的信噪比和幅度。
  6. 根据权利要求3所述的分布式语音处理系统，其特征在于，所述声音时间信息包括所述音频信号的开始时间和结束时间，或所述音频信号的开始时间和持续时间。
  7. 根据权利要求3所述的分布式语音处理系统，其特征在于，所述第一声音预处理结果和所述一个或多个第二声音预处理结果的每个还包括所述音频信号的递增序号。
  8. 根据权利要求3所述的分布式语音处理系统，其特征在于，所述声音处理模块还配置为对于所述第一声音预处理结果和所述一个或多个第二声音预处理结果中的每个，判断其声音质量是否超过预定阈值，若否，则丢弃该语音预处理结果。
  9. 根据权利要求3所述的分布式语音处理系统，其特征在于，所述声音处理模块还配置为在所述第一声音预处理结果和所述一个或多个第二声音预处理结果中，选取一个或多个声音质量最高的声音预处理结果进行语音识别以得到所述第一语音识别结果。
  10. 根据权利要求2所述的分布式语音处理系统，其特征在于，所述声音处理模块还配置为对所述第一语音识别结果和所述一个或多个第二语音识别结果执行加权处理以得到所述最终语音识别结果。
  11. 一种分布式语音处理方法，由网络中的节点设备执行，其特征在于，包括：
    若所述节点设备包括声音采集模块，则执行下述步骤：
    采集音频信号；
    对所述音频信号进行预处理以得到第一声音预处理结果；及
    将所述第一声音预处理结果发送到所述网络中的一个或多个节点设备；
    从所述网络接收来自至少一个其他节点设备的一个或多个第二声音预处理结果；及
    基于所述第一声音预处理结果和/或所述一个或多个第二声音预处理结果执行语音识别。
  12. 根据权利要求11所述的分布式语音处理方法，其特征在于，还包括将所述第一语音识别结果发送到所述网络中的一个或多个节点设备；
    从所述网络接收来自至少一个其他节点设备的一个或多个第二语音识别结果；及
    基于所述第一语音识别结果和所述一个或多个第二语音识别结果执行语音识别以得到最终语音识别结果。
  13. 根据权利要求11所述的分布式语音处理方法，其特征在于，所述第一声音预处理结果和所述一个或多个第二声音预处理结果的每个包括声音特征值、声音质量及声音时间信息。
  14. 根据权利要求13所述的分布式语音处理方法，其特征在于，所述声音特征值为所述音频信号的MFCC特征值或PLP特征值。
  15. 根据权利要求13所述的分布式语音处理方法，其特征在于，所述声音质量包括所述音频信号的信噪比和幅度。
  16. 根据权利要求13所述的分布式语音处理方法，其特征在于，所述声音时间信息包括所述音频信号的开始时间和结束时间，或所述音频信号的开始时间和持续时间。
  17. 根据权利要求13所述的分布式语音处理方法，其特征在于，所述第一声音预处理结果和所述一个或多个第二声音预处理结果的每个还包括所述音频信号的递增序号。
  18. 根据权利要求13所述的分布式语音处理方法，其特征在于，对于所述第一声音预处理结果和所述一个或多个第二声音预处理结果中的每个，判断其声音质量是否超过预定阈值，若否，则丢弃该语音预处理结果。
  19. 根据权利要求13所述的分布式语音处理方法，其特征在于，在所述第一声音预处理结果和所述一个或多个第二声音预处理结果中选取一个或多个声音质量最高的声音预处理结果进行语音识别以得到所述第一语音识别结果。
  20. 根据权利要求12所述的分布式语音处理方法，其特征在于，对所述第一语音识别结果和所述一个或多个第二语音识别结果执行加权处理以得 到所述最终语音识别结果。
PCT/CN2021/143983 2020-12-31 2021-12-31 分布式语音处理系统及方法 WO2022144009A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US18/260,196 US20240062764A1 (en) 2020-12-31 2021-12-31 Distributed speech processing system and method

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011628865.4A CN112652310A (zh) 2020-12-31 2020-12-31 分布式语音处理系统及方法
CN202011628865.4 2020-12-31

Publications (1)

Publication Number Publication Date
WO2022144009A1 true WO2022144009A1 (zh) 2022-07-07

Family

ID=75366909

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/143983 WO2022144009A1 (zh) 2020-12-31 2021-12-31 分布式语音处理系统及方法

Country Status (3)

Country Link
US (1) US20240062764A1 (zh)
CN (1) CN112652310A (zh)
WO (1) WO2022144009A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112652310A (zh) * 2020-12-31 2021-04-13 乐鑫信息科技（上海）股份有限公司 分布式语音处理系统及方法

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5430730A (en) * 1993-09-14 1995-07-04 Rolm Company Method for building a sub-network in a distributed voice messaging system
CN105206281A (zh) * 2015-09-14 2015-12-30 胡旻波 基于分布式麦克风阵列网络的语音增强方法
CN107925818A (zh) * 2015-10-15 2018-04-17 华为技术有限公司 用于声音处理节点装置的声音处理节点
US20180190292A1 (en) * 2016-02-03 2018-07-05 Hua Xu Voice recognition system and construction method thereof
CN110046222A (zh) * 2019-03-04 2019-07-23 视联动力信息技术股份有限公司 一种智能问答方法和系统
CN110136708A (zh) * 2019-04-23 2019-08-16 深圳合一智控科技有限公司 一种基于蓝牙Mesh的分布式语音控制系统及控制方法
CN110310637A (zh) * 2019-06-25 2019-10-08 重庆信络威科技有限公司 一种基于分布式多麦克风及蓝牙Mesh的语音控制方法及系统
CN110838286A (zh) * 2019-11-19 2020-02-25 腾讯科技（深圳）有限公司 一种模型训练的方法、语种识别的方法、装置及设备
CN112652310A (zh) * 2020-12-31 2021-04-13 乐鑫信息科技（上海）股份有限公司 分布式语音处理系统及方法

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN206516350U (zh) * 2017-01-20 2017-09-22 浙江小尤鱼智能技术有限公司 一种基于分布式语音控制的智能家居系统
CN109545242A (zh) * 2018-12-07 2019-03-29 广州势必可赢网络科技有限公司 一种音频数据处理方法、系统、装置及可读存储介质
CN111415657A (zh) * 2019-01-07 2020-07-14 成都启英泰伦科技有限公司 一种去中心化设备、多设备系统及其语音控制方法
CN112071331B (zh) * 2020-09-18 2023-05-30 平安科技（深圳）有限公司 语音文件修复方法、装置、计算机设备及存储介质

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5430730A (en) * 1993-09-14 1995-07-04 Rolm Company Method for building a sub-network in a distributed voice messaging system
CN105206281A (zh) * 2015-09-14 2015-12-30 胡旻波 基于分布式麦克风阵列网络的语音增强方法
CN107925818A (zh) * 2015-10-15 2018-04-17 华为技术有限公司 用于声音处理节点装置的声音处理节点
US20180190292A1 (en) * 2016-02-03 2018-07-05 Hua Xu Voice recognition system and construction method thereof
CN110046222A (zh) * 2019-03-04 2019-07-23 视联动力信息技术股份有限公司 一种智能问答方法和系统
CN110136708A (zh) * 2019-04-23 2019-08-16 深圳合一智控科技有限公司 一种基于蓝牙Mesh的分布式语音控制系统及控制方法
CN110310637A (zh) * 2019-06-25 2019-10-08 重庆信络威科技有限公司 一种基于分布式多麦克风及蓝牙Mesh的语音控制方法及系统
CN110838286A (zh) * 2019-11-19 2020-02-25 腾讯科技（深圳）有限公司 一种模型训练的方法、语种识别的方法、装置及设备
CN112652310A (zh) * 2020-12-31 2021-04-13 乐鑫信息科技（上海）股份有限公司 分布式语音处理系统及方法

Also Published As

Publication number Publication date
CN112652310A (zh) 2021-04-13
US20240062764A1 (en) 2024-02-22

Similar Documents

Publication Publication Date Title
US10735479B2 (en) Methods, systems and apparatus to enable voice assistant device communication
CN106653008B (zh) 一种语音控制方法、装置及系统
CN105788599B (zh) 语音处理方法、路由器及智能语音控制系统
CN107622652A (zh) 家电系统的语音控制方法与家电控制系统
US11765782B2 (en) Network connectivity analyzer and device peer-assisting
JP2019204074A (ja) 音声対話方法、装置及びシステム
CN112035086B (zh) 音频播放方法及装置
CN108470568B (zh) 智能设备控制方法及装置、存储介质、电子设备
CN112037789A (zh) 设备唤醒方法、装置、存储介质及电子装置
CN108156497B (zh) 一种控制方法、控制设备及控制系统
WO2017210991A1 (zh) 一种语音过滤的方法、装置及系统
CN108711424B (zh) 一种分布式语音控制方法及系统
US9774743B2 (en) Silence signatures of audio signals
CN109949801A (zh) 一种基于耳机的智能家居设备语音控制方法及系统
WO2022144009A1 (zh) 分布式语音处理系统及方法
EP3761653A1 (en) Playing control method and apparatus for device group, and playing system
CN112151013A (zh) 智能设备交互方法
CN112838967B (zh) 主控设备、智能家居及其控制装置、控制系统及控制方法
CN109979495B (zh) 基于人脸识别的音频进度智能跟随播放方法及系统
CN110493092B (zh) 基于远场语音和iot的万能遥控以及家电遥控方法
CN112820287A (zh) 分布式语音处理系统及方法
CN113674738A (zh) 一种全屋分布式语音的系统和方法
CN112002319A (zh) 智能设备的语音识别方法及装置
US9843683B2 (en) Configuration method for sound collection system for meeting using terminals and server apparatus
CN218217617U (zh) 一种实现多蓝牙麦克风音箱的网络会议管理系统

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21914737

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 18260196

Country of ref document: US

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21914737

Country of ref document: EP

Kind code of ref document: A1