WO2017000771A1

WO2017000771A1 - System for cancelling environment noise and application method thereof

Info

Publication number: WO2017000771A1
Application number: PCT/CN2016/085750
Authority: WO
Inventors: 施家琪; 刘鑫
Original assignee: 芋头科技（杭州）有限公司
Priority date: 2015-06-30
Filing date: 2016-06-14
Publication date: 2017-01-05
Also published as: CN106328151A; HK1231621A1; TWI581254B; CN106328151B; TW201701274A

Abstract

A system for cancelling environment noise and application method thereof. The method comprises: firstly detecting whether audio data is an onset frame of effective speech according to frequency spectrum energy, and counting the average noise variance as the initial value of a noise power spectrum if there is no effective speech; determining that a current signal is noise or an effective speech signal according to estimation of the noise power spectrum after detecting the onset frame of effective speech; performing smooth processing on the previous noise power spectrum and a power spectrum of a current frame, and updating the estimation of the noise power spectrum, if the current signal is noise; and updating a noise power spectrum of each point in a frequency spectrum of a current speech frame, calculating the estimated value of Fourier Transform of pure speech of the current frame by using a minimum mean square error determining method, and obtaining, by using inverse Fourier Transform, a current frame on which environment noise cancellation is performed, if the current signal is speech.

Description

Ring noise elimination system and application method thereof

Technical field

The present invention relates to the field of intelligent voice interaction, and in particular to a ring noise cancellation system and an application method thereof.

Background technique

With the expansion of the application field of robots and the rapid development of artificial intelligence technology, intelligent robots for ordinary families have begun to attract people's attention. The home-oriented intelligent robot is a kind of anthropomorphic machine system with different external sensing ability and processing ability, which is different from industrial control robot. Unlike the traditional industrial robots, the most important feature of home robots is that they are easy to interact. The most natural way of interaction is voice interaction. However, the voice interaction between human and machine has great uncertainty. One of the key problems is the daily environment. Noise has a great influence on effective speech. The noise in the daily environment will cause the robot to not recognize or misidentify the human voice, and finally the effect of voice interaction is greatly reduced. At present, most of the intelligent robots used in the home are unable to work normally in an environment with low signal-to-noise due to the lack of effective elimination of environmental noise. Other smart devices are applied to the voice enhancement technology in the traditional telephone communication field to be applied to the home smart device. The traditional environmental noise elimination and speech enhancement technologies applied to the telephone communication mostly provide a certain distance to the interaction distance between the speaker and the device. limit. In addition, some speech enhancement techniques applied in the field of traditional telephone communication have only a good effect on steady-state noise, and are less effective for some unsteady noise in the daily environment.

Home intelligence robot is a brand new field, and current environmental noise cancellation technology targets There is no deep research and application in this field. Traditional environmental noise cancellation technology is mostly used for telephone communication. The main purpose of these technologies is to prevent the influence of environmental background noise on communication quality. At present, the common environmental noise cancellation techniques used by smart devices based on voice interaction mostly come from the prior art of traditional telephone communication. These technologies include spectrum subtraction method, Wiener filtering method, and adaptive noise cancellation method. There are some problems with the application of these methods to the robotics field. The spectral subtraction method is a method of subtracting the mean value of the speech segment from the mean value of the speech segment to obtain the noise mean value, and then using the noise mean value to eliminate the noise. This method has a poor effect on the unsteady noise, and is easy to cause the noise to be eliminated. The speech is distorted, resulting in a decrease in speech recognition rate. The Wiener filter rule uses the transfer function of the Wiener filter to convolve the mean value of the noise amplitude with the amplitude of the speech segment to obtain the amplitude information of the signal after noise cancellation. The Wiener filtering method does not cause more serious speech distortion, and can effectively suppress noise with a small or stable range in the environment. However, the method estimates the noise mean by estimating the noise power spectrum by calculating the statistical average of the silent period. This estimation is based on the premise that the noise power spectrum does not change much before and after the utterance, so the change is larger. In the case of steady-state noise, this method cannot obtain better noise reduction effects. Another method of environmental noise cancellation that is commonly used on smart devices is directional microphone plus adaptive noise cancellation, which uses an omnidirectional microphone to collect ambient noise, a directional microphone to collect user speech, and then for both signals. Do adaptive noise reduction to obtain a pure speech signal. The above method has a good environmental noise elimination effect on a conventional telephone handset, but cannot be applied to a robot. This is because the above method relies on a directional microphone to collect user speech, which requires the user and the directional microphone to maintain a certain distance and direction, which is obviously not applicable in the home robot environment.

Summary of the invention

In view of the above problems, the present invention provides a ring noise cancellation method for a home robot, which is characterized in that it comprises the following steps:

Step S1: an acquisition unit samples a sound signal in an environment in which the robot is currently located, acquires an audio signal that matches the sound signal, and acquires the audio signal by using Fourier transform analysis. a signal that matches the acquired spectrum signal, wherein the audio signal is composed of a plurality of audio frame signals;

Step S2, determining whether the currently-mentioned audio frame matches a preset voice frame according to a predetermined method;

Step S3, in a state in which the audio frame matches the voice frame, updating a noise power spectrum corresponding to each of the audio frames in the acquired spectrum;

Step S4: acquiring a gain of the noise power spectrum in the acquired spectrum, and performing Fourier transform calculation on the acquired spectrum according to the gain to form a calculation result output;

Step S5, performing inverse Fourier transform on the calculation result, and removing noise in the audio signal to obtain a valid audio signal.

The above method is characterized in that the step 3 further includes:

Step 31, in the state that the audio frame does not match the voice frame, the current noise power spectrum is read; in the state where the noise power spectrum is 0, step 33 is performed;

Step 32: Calculate a new power noise spectrum according to the noise power spectrum and the acquired spectrum processing, and assign the new power noise spectrum to the power noise spectrum;

Step 33: Calculate and update the noise power spectrum according to the noise variance in the environment in which the computing environment is located.

The above method is characterized in that, in the step 32, the noise power spectrum and the acquisition spectrum obtain a new power noise spectrum according to a smoothing processing method.

A ring noise cancellation system is applied to a home robot, and is characterized in that: an acquisition unit is configured to sample a sound signal in an environment in which the robot is currently located to obtain an audio signal that matches the sound signal, And acquiring, by the Fourier transform analysis, the acquired spectrum signal that matches the audio signal, wherein the sound signal includes a noise signal, and the audio signal is composed of a plurality of audio frame signals;

a determining unit, configured to connect the collecting unit to determine whether the currently-mentioned audio frame matches a preset voice frame according to a predetermined method; and form a judgment result output;

And an update unit, connected to the determining unit, in a state that the audio frame matches the voice frame, and a noise power spectrum corresponding to each of the audio frames in the acquired spectrum;

a calculation unit, connected to the update unit, acquiring a gain of the noise power spectrum in the acquired spectrum, and performing Fourier transform calculation on the acquired spectrum according to the gain to form a calculation result output;

And a processing unit, configured to receive the calculation result output by the calculation unit, perform inverse Fourier transform processing on the calculation result, and remove noise in the audio signal to obtain an effective audio signal.

The above system is characterized in that: the collecting unit comprises:

a buffer unit, configured to receive the sound signal acquired by the acquisition unit

a micro processing unit is connected to the buffer unit for performing a buffer sampling process on the sound signal in the buffer unit.

The above system is characterized in that, in the buffer unit, the sound signal is stored in the buffer unit in a predetermined audio packet format.

The above system is characterized in that the audio processing framework layer further comprises:

And a reading unit connected to the buffer unit for reading a predetermined audio data packet stored in the buffer unit.

The above system is characterized in that the audio data packet has a duration of 10 ms.

In summary, a ring noise elimination system designed by the present invention and an application method thereof are a complete software technical solution for environmental noise elimination of a home intelligent robot based on voice interaction as an interaction basis, and solve the traditional environmental noise elimination range. Larger unsteady noise is not a good problem.

DRAWINGS

Embodiments of the present invention are described more fully with reference to the accompanying drawings. However, the attached drawings are for illustration and illustration only and are not intended to limit the scope of the invention.

FIG. 1 is a schematic diagram of an algorithm flow of the present invention.

detailed description

In order to make the technical solutions and advantages of the present invention easier to understand, the following detailed description will be made with reference to the accompanying drawings. It should be noted that the specific embodiments described herein are merely illustrative The present invention is not intended to limit the invention.

The core idea of the present invention is to first detect whether the audio data is a valid speech start frame according to the spectral energy, and if there is no valid speech, the statistical average noise variance is used as the initial value of the noise power spectrum. When a valid speech start frame is detected, it is determined whether the current signal is a noise or a valid speech signal based on an estimate of the noise power spectrum. If it is noise, smooth the previous noise power spectrum and the power spectrum of the current frame, update the estimation of the noise power spectrum, and if it is speech, update the noise power spectrum of each point in the current speech frame spectrum, and then The minimum mean square error discriminant method is used to calculate the estimate of the pure Fourier transform of the current frame, and then the inverse Fourier transform is used to obtain the current frame after the ambient sound is removed.

Therefore, the present invention relates to a ring noise cancellation method, which is applied to a home robot, which includes the following steps:

Step S1: an acquisition unit samples a sound signal in a current environment of the robot, acquires an audio signal that matches the sound signal, and acquires an acquired spectrum signal that matches the audio signal by performing Fourier transform analysis on the audio signal. Wherein the audio signal is composed of a plurality of audio frame signals;

Step S2, determining whether the current audio frame matches a preset voice frame according to a predetermined method;

Step S3, updating a noise power spectrum corresponding to each audio frame in the acquired spectrum in a state in which the audio frame matches the voice frame;

Step S4: acquiring a gain of a noise power spectrum in the acquired spectrum, and performing Fourier transform calculation on the acquired spectrum according to the gain to form a calculation result output;

In step S5, an inverse Fourier transform is performed on the calculation result to remove noise in the audio signal to obtain an effective audio signal.

Wherein, in step S3, the method further includes:

Step 31, in the state in which the audio frame does not match the voice frame, the current noise power spectrum is read; in the state where the noise power spectrum is 0, step 33 is performed;

Step 32: Calculate a new power noise spectrum according to the noise power spectrum and the acquisition spectrum processing, and assign a new power noise spectrum to the power noise spectrum;

Step 33: Calculate and update the noise power spectrum according to the noise variance in the environment in which the calculation is performed.

In step 32, the noise power spectrum and the acquisition spectrum are subjected to a smoothing process to obtain a new power noise spectrum.

In addition, the present invention designs a ring noise cancellation system according to the ring noise elimination method, which is applied to a home robot, and includes:

The acquisition unit is configured to sample the sound signal in the environment in which the robot is currently located to obtain an audio signal that matches the sound signal, and obtain an acquired spectrum signal that matches the audio signal by performing Fourier transform analysis on the audio signal, where The sound signal includes a noise signal, and the audio signal is composed of a plurality of audio frame signals;

a determining unit, connected to the collecting unit, configured to determine whether the current audio frame matches a predetermined voice frame according to a predetermined method; and form a judgment result output;

The updating unit and the connection judging unit further acquire a noise power spectrum corresponding to each audio frame in the spectrum when the judgment result is that the audio frame matches the speech frame;

a calculation unit, connected to the update unit, obtains a gain of the noise power spectrum in the acquired spectrum, and performs Fourier transform calculation on the acquired spectrum according to the gain to form a calculation result output;

The processing unit is connected to the calculation unit for receiving the calculation result output by the calculation unit, performing inverse Fourier transform processing on the calculation result, and removing noise in the audio signal to obtain an effective audio signal.

In the present invention, the acquisition unit includes:

a buffer unit for receiving a sound signal acquired by the acquisition unit;

A micro processing unit is connected to the buffer unit for performing a pulse sampling process on the sound signal in the buffer unit.

In the present invention, in the buffer unit, the sound signal is stored in the buffer unit in a predetermined audio packet format.

In the present invention, the audio processing framework layer further includes:

The reading unit is connected to the buffer unit for reading a predetermined audio data packet stored in the buffer unit.

The following is specifically described in conjunction with the embodiments.

In view of the problems that can only be encountered in robotic environment noise reduction, the line proposes an environmental noise elimination technology solution for smart home robots:

A software system layer framework and an audio noise reduction processing algorithm for implementing audio collection and processing in a preset embedded system are applied to a home intelligent robot, wherein the software system includes:

An acquisition unit, a software driver for collecting audio stream data from a specific audio device;

The collection unit further includes:

a buffer unit and a micro processing unit, the buffer unit is configured to receive the sound signal acquired by the collecting unit, the micro processing unit is connected to the buffer unit, and is used for resampling the sound signal in the buffer unit, and a noise reduction component formed by the reading unit Processing the software layer, the reading unit is connected to the buffer unit for reading a predetermined audio data packet stored in the buffer unit. The micro processing unit reads the audio data stream from the buffer unit, and performs necessary re-sampling processing on the audio data stream, and then cuts the audio data packet into 10 milliseconds to be filled into the buffer unit, wherein the reading unit is used for the buffer unit. Reading the preprocessed audio data packet and then transmitting it to the algorithm for environmental noise cancellation processing;

The determining unit is connected to the collecting unit for determining whether the current audio frame matches a predetermined voice frame according to a predetermined method, forming a judgment result output, and adapting different underlying drivers and an adaptation software program of the upper application software;

And an update unit, connected to the determining unit, and updating a noise power spectrum corresponding to each audio frame in the acquired spectrum in a state that the audio frame matches the voice frame;

The calculating unit obtains the gain of the noise power spectrum in the acquired spectrum, and performs Fourier transform calculation on the acquired spectrum according to the gain to form a calculation result output;

The processing unit is connected to the calculation unit for accepting the calculation result output by the calculation unit, performing inverse Fourier transform processing on the calculation result, and removing noise in the audio to obtain an effective audio signal.

The application interface sends the obtained valid audio signal to the upper application.

As shown in FIG. 1 , for the application of the above system, a calculation method of ring noise cancellation is also involved:

Step S1: The home intelligent robot collects the sound signal through the audio device, acquires the audio data that matches the sound signal, and then detects whether the collected audio data is a valid voice start frame according to the energy of the audio data, if the detected result is displayed as For invalid speech data, the average noise variance is counted as the initial value of the noise power spectrum. If the detected result shows valid speech, proceed to the next step.

Wherein, in step S1, the home intelligence robot obtains a spectrum of the collected audio data by performing discrete Fourier transform on the collected audio data, and the discrete Fourier transform is performed by transforming the signal from the time domain to the frequency domain. Then study the spectral structure and variation of the signal.

Step S2, judging whether the collected audio data is noise according to the estimated noise power spectrum in step S1, and if it is noise, smoothing the power spectrum of the noise and the previously calculated noise power spectrum, after finishing After the smoothing process, the estimation of the noise power spectrum is updated. If the noise is not detected but the normal command voice, the next step is performed.

Step S3, updating the noise power spectrum of each point in the spectrum of the speech frame collected to the audio data, and then calculating the pure tone Fourier change estimate of the audio data, in the present invention, by using the minimum mean square error discriminant method A pure tone Fourier change estimate for the speech frame is calculated.

Step S4: Acquire a noise power gain in the acquired spectrum, and perform Fourier transform calculation on the acquired spectrum according to the gain to form a calculation result output.

In step S5, the ambient sound collected in the audio data is removed to obtain a pure sound, so that only the robot can clearly obtain the command content. In the present invention, the environment in the audio data is removed by discrete Fourier transform. sound.

Therefore, in a home-based intelligent robot based on Android embedded intelligent operating system based on voice interaction, the robot needs to eliminate the ambient noise after speech acquisition to improve the accuracy of speech recognition.

Exemplary embodiments of the specific structure of the specific embodiments are given by way of illustration and the accompanying drawings, and other transitions are possible in accordance with the spirit of the invention. Although the above invention proposes a prior preferred embodiment, these are not intended to be limiting.

Various changes and modifications will no doubt become apparent to those skilled in the <RTIgt; Accordingly, the appended claims are to cover all such modifications and modifications The scope and content of any and all equivalents are intended to be within the scope and spirit of the invention.

Claims

A ring noise elimination method is applied to a home robot, which is characterized in that it comprises the following steps:

Step S1: an acquisition unit samples a sound signal in an environment in which the robot is currently located, acquires an audio signal that matches the sound signal, and acquires and analyzes the audio signal by discrete Fourier transform analysis. An acquisition spectrum signal matched by an audio signal, wherein the audio signal is composed of a plurality of audio frame signals;

Step S2, determining whether the currently-mentioned audio frame matches a preset voice frame according to a predetermined method;

Step S3, in a state in which the audio frame matches the voice frame, updating a noise power spectrum corresponding to each of the audio frames in the acquired spectrum;

Step S4: acquiring a gain of the noise power spectrum in the acquired spectrum, and performing Fourier transform calculation on the acquired spectrum according to the gain to form a calculation result output;

Step S5, performing inverse Fourier transform on the calculation result, and removing noise in the audio signal to obtain a valid audio signal.
The method according to claim 1, wherein the step 3 further comprises:

Step 31, in the state that the audio frame does not match the voice frame, the current noise power spectrum is read; in the state where the noise power spectrum is 0, step 33 is performed;

Step 32: Calculate a new power noise spectrum according to the noise power spectrum and the acquired spectrum processing, and assign the new power noise spectrum to the power noise spectrum;

Step 33: Calculate and update the noise according to the noise variance in the environment in which the computing environment is located power spectrum.
The method according to claim 2, wherein in the step 32, the noise power spectrum and the acquired spectrum obtain a new power noise spectrum according to a smoothing method.
A ring noise cancellation system for use in a home robot, characterized in that it comprises:

An acquisition unit, configured to sample a sound signal in an environment in which the robot is currently located to obtain an audio signal that matches the sound signal, and acquire the audio signal by using Fourier transform analysis a matched acquisition spectrum signal, wherein the sound signal includes a noise signal, and the audio signal is composed of a plurality of audio frame signals;

a determining unit, configured to connect the collecting unit to determine whether the currently-mentioned audio frame matches a preset voice frame according to a predetermined method, and form a judgment result output;

And an update unit, connected to the determining unit, in a state that the audio frame matches the voice frame, and a noise power spectrum corresponding to each of the audio frames in the acquired spectrum;

a calculation unit, connected to the update unit, acquiring a gain of the noise power spectrum in the acquired spectrum, and performing Fourier transform calculation on the acquired spectrum according to the gain to form a calculation result output;

And a processing unit, configured to receive the calculation result output by the calculation unit, perform inverse Fourier transform processing on the calculation result, and remove noise in the audio signal to obtain an effective audio signal.
The system of claim 4, wherein the processing unit comprises:

a buffer unit, configured to receive the sound signal acquired by the collection unit;

a micro processing unit is coupled to the buffer unit for performing resampling processing on the sound signal in the buffer unit.
The system according to claim 5, wherein in said buffer unit, said sound signal is stored in said buffer unit in a predetermined audio packet format.
The system of claim 6 wherein the processing unit further comprises:

And a reading unit connected to the buffer unit for reading a predetermined audio data packet stored in the buffer unit.
The system of claim 6 wherein said audio data packet has a duration of 10 ms.