Information processing method and device for telephone speech recognition

By random velocity disturbance, noise addition, downsampling and channel simulation on non-phone channel voice data, training samples are generated, which solves the problem of low phone voice recognition accuracy and improves the performance ability of voiceprint recognition system in telephone channels and the performance of voiceprint recognition system under telephone channels and Great.

CN117765951BActive Publication Date: 2024-11-08NANJING LONGYUAN INFORMATION TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202311224332.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-09-21
Publication Date
2024-11-08
Estimated Expiration
2043-09-21

AI Technical Summary

Technical Problem

The accuracy of telephone voice recognition cannot reach the ideal level, mainly due to the diverse channel distortion and data privacy issues, which cannot be used directly as training samples, making it difficult to effectively train the phone voice.

Method used

By obtaining non-phone channel voice data, random velocity perturbation, adding noise and reverb, audio downsampling, and performing telephone channel transmission simulation, generating training samples for training of voiceprint recognition models, simulating distortion in telephone channels and Noise environment, enhances the robustness of the model.

Benefits of technology

Through specific data enhancement methods, the problem of model training in the absence of real telephone data is overcome, and the performance ability and robustness of the voiceprint recognition system in the telephone channel is improved, so as to better adapt to actual telephone communication. Distortion and noise conditions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117765951B_ABST
    Figure CN117765951B_ABST
Patent Text Reader

Abstract

The present application discloses an information processing method and device for telephone voice recognition, wherein the method includes obtaining non-telephone channel voice data, and performing random speed perturbation on the non-telephone channel voice data to obtain first voice data; adding noise and reverberation data to the first voice data to obtain second voice data; performing audio downsampling on the second voice data, and performing telephone channel transmission simulation on the downsampled data to obtain third voice data; based on the third voice data, training a preset voiceprint recognition model, and using the third voice data as a training sample to train the preset voiceprint recognition model. The data obtained by a specific data enhancement method is used for voiceprint recognition system model training, so as to realize the training of telephone channel voiceprint recognition model with a small amount of data, thereby overcoming the problem that it is difficult to implement model training for telephone data when no real telephone data is available.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of signal processing technology, and in particular to an information processing method and device for telephone voice recognition. Background Art

[0002] Due to diverse channel distortions and data privacy considerations, it cannot be used directly as training samples, making the task of recognizing real telephone speech through a model challenging.

[0003] In the related art, the accuracy of telephone voice recognition cannot reach a relatively ideal level. Summary of the invention

[0004] The main purpose of this application is to provide an information processing method and device for telephone voice recognition to overcome the defects of related technologies.

[0005] In order to achieve the above-mentioned objectives, in a first aspect, the present application provides an information processing method for telephone voice recognition, comprising: obtaining non-telephone channel voice data, and performing random speed perturbation on the non-telephone channel voice data to obtain first voice data; adding noise and reverberation data to the first voice data to obtain second voice data; performing audio downsampling on the second voice data, and simulating telephone channel transmission on the downsampled data to obtain third voice data; training a preset voiceprint recognition model based on the third voice data, and using the third voice data as training samples to train the preset voiceprint recognition model.

[0006] In some embodiments, simulating telephone channel transmission of the downsampled data includes: compressing the downsampled data based on different compression methods to obtain data under different compression methods; and using the downsampled data and the data under different compression methods as the third voice data.

[0007] In some embodiments, audio downsampling is performed on the second voice data, and telephone channel transmission simulation is performed on the downsampled data to obtain third voice data, including: clipping simulation is performed on the second voice data to obtain fourth voice data; audio downsampling is performed on the fourth voice data, and telephone channel transmission simulation is performed on the sampled data to obtain third voice data.

[0008] In some embodiments, simulating telephone channel transmission of the downsampled data to obtain the third voice data includes: performing telephone channel encoding on the downsampled data to obtain encoded audio data; performing network packet loss simulation on the encoded audio data to obtain fifth voice data; performing packet duplication simulation on the encoded audio data to obtain sixth voice data; performing packet damage simulation on the encoded audio data to obtain seventh voice data; performing packet disorder simulation on the encoded audio data to obtain eighth voice data; decoding the fifth voice data, the sixth voice data, the seventh voice data, and the eighth voice data, and using the decoded data as the third voice data.

[0009] In some embodiments, clipping simulation is performed on the second voice data to obtain fourth voice data: data in a specific time interval in the second voice data is determined based on a random mask; and plosive sound simulation is performed on the data in the specific time interval to obtain fourth voice data.

[0010] In some embodiments, the method further includes: extracting telephone voice data; inputting the telephone voice data into a trained voiceprint recognition model to determine the voiceprint feature vector in the telephone voice data.

[0011] In some embodiments, after extracting the telephone voice data, the method also performs echo elimination on the telephone voice data, including: based on a high-pass filter, attenuating the data in the telephone voice data whose frequency is lower than a preset cutoff frequency, and transmitting the telephone voice signal whose frequency is higher than or equal to the preset cutoff frequency; based on a preset echo elimination rule, adaptively filtering the existing telephone voice signal whose frequency is lower than the preset cutoff frequency; suppressing the echo signal of a preset frequency band in the data after adaptive filtering to obtain a signal after echo elimination.

[0012] In some embodiments, the adaptive filtering of data having a frequency lower than a preset cutoff frequency based on a preset echo cancellation rule includes dynamically adjusting the tap coefficients of the high-pass filter with the goal of minimizing the mean square value of the difference between the telephone voice signal output by the high-pass filter and the expected signal.

[0013] In a second aspect, the present application also provides an information processing device for telephone voice recognition, comprising a first processing unit, configured to obtain non-telephone channel voice data, and perform random speed perturbation on the non-telephone channel voice to obtain first voice data; a second processing unit, configured to add noise and reverberation data to the first voice data to obtain second voice data; a third processing unit, configured to perform audio downsampling on the second voice data, and simulate telephone channel transmission of the downsampled data to obtain third voice data; a training unit, configured to train a preset voiceprint recognition model based on the third voice data, so as to train the preset voiceprint recognition model using the third voice data as a training sample.

[0014] In some embodiments, the device also includes: a voice extraction unit configured to extract telephone voice data; and a model recognition unit configured to input the telephone voice data into a trained voiceprint recognition model to determine the voiceprint feature vector in the telephone voice data.

[0015] In a third aspect, the present application also provides an electronic device, comprising: a memory for storing a computer program product; a processor for executing the computer program product stored in the memory, and when the computer program product is executed, the method in any of the above embodiments is implemented.

[0016] In a fourth aspect, the present application further provides a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, implement the method in any of the above embodiments.

[0017] The present application provides an information processing method and device for telephone voice recognition, wherein the method includes obtaining non-telephone channel voice data, and performing random speed perturbation on the non-telephone channel voice to obtain first voice data; adding noise and reverberation data to the first voice data to obtain second voice data; performing audio downsampling on the second voice data, and simulating telephone channel transmission on the downsampled data to obtain third voice data; based on the third voice data, training a preset voiceprint recognition model, and using the third voice data as a training sample to train the preset voiceprint recognition model. The data obtained by a specific data enhancement method is used for voiceprint recognition system model training, so as to realize the training of telephone channel voiceprint recognition model with a small amount of data, thereby overcoming the problem that it is difficult to implement model training for telephone data when no real telephone data is available. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings constituting a part of this application are used to provide a further understanding of this application, so that other features, purposes and advantages of this application become more obvious. The schematic embodiment drawings and their descriptions of this application are used to explain this application and do not constitute an improper limitation on this application. In the drawings:

[0019] Figure 1 is a flow chart of the information processing method for telephone speech recognition of the present application;

[0020] Figure 2 It is a flow chart of the present application for simulating telephone channel transmission of downsampled data;

[0021] Figure 3 It is an application diagram of the coding and decoding process of the analog telephone channel of the present application;

[0022] Figure 4 is another flow chart of the present application for simulating telephone channel transmission of downsampled data;

[0023] Figure 5 It is another flow chart of the present application for simulating telephone channel transmission of downsampled data based on the clipping technology;

[0024] Figure 6 is a data enhancement workflow diagram of random mask-based packet loss and clipped audio simulation of the present application;

[0025] Figure 7 is a schematic diagram of the audio clipping simulation process of the present application;

[0026] Figure 8 is another flow chart of the information processing method for telephone speech recognition of the present application;

[0027] Fig. 9 It is an application schematic diagram of the information processing method for telephone speech recognition of the present application;

[0028] Fig.10 It is a flow chart of echo cancellation in voice IP phone in this application;

[0029] Fig.11 It is a structural schematic diagram of an application embodiment of the electronic device of the present application. DETAILED DESCRIPTION

[0030] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0031] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0032] In the present application, the terms "upper", "lower", "left", "right", "front", "back", "top", "bottom", "inner", "outer", "middle", "vertical", "horizontal", "lateral", "longitudinal" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings. These terms are mainly used to better describe the present application and its embodiments, and are not used to limit the indicated devices, elements or components to have a specific orientation, or to be constructed and operated in a specific orientation.

[0033] In addition, some of the above terms may be used to express other meanings in addition to indicating orientation or positional relationship. For example, the term "on" may also be used to express a certain dependency or connection relationship in some cases. For those of ordinary skill in the art, the specific meanings of these terms in this application can be understood according to specific circumstances.

[0034] In addition, the term "plurality" shall mean two or more.

[0035] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0036] Reference below Figure 1 , Figure 1 A flowchart of an embodiment of the task generation method of the present application is shown as follows: Figure 1 As shown, the process includes the following steps:

[0037] Step 101: Acquire non-telephone channel voice data, and perform random speed disturbance on the non-telephone channel voice to obtain first voice data.

[0038] In this embodiment, after the voice data is speed-changed by a random speed perturbation method, the speed perturbation effect existing in the telephone channel can be simulated by changing the playback speed of the voice signal. In the training process of the voiceprint recognition system, in order to increase the diversity of training data, the non-telephone channel voice data is changed to make it closer to the voice under the telephone channel. Therefore, by the random speed perturbation method, the voice data can be randomly speed-changed in each training iteration.

[0039] Specifically, the first round of data multiplication is performed based on y = F1(x, v), where x represents the non-telephone channel voice data, v represents the random speed disturbance parameter, and F represents the function or algorithm of the speed disturbance. By combining the non-telephone channel voice data x with the random speed disturbance parameter v, the data y after the random speed disturbance method can be obtained for the first round of data multiplication. Exemplarily, different techniques such as time domain speed change, frequency domain speed change or other methods can be used to achieve random speed disturbance.

[0040] Furthermore, in each training iteration, a speed disturbance parameter v can be randomly generated, which is used to control the speed change degree of the voice data. Then, the non-telephone channel voice data x is combined with the speed disturbance parameter v to perform speed change processing, wherein the speed change can be performed in the time domain or the frequency domain. The specific speed change algorithm can adopt linear interpolation method, phase stretching method, synchronous resampling method and other methods, and finally the data y after speed disturbance processing is obtained.

[0041] This embodiment uses a random speed perturbation method to simulate the speed perturbation effect of voice data under different telephone channels, thereby increasing the diversity of training data and improving the performance of the voiceprint recognition system on telephone channel data.

[0042] Step 102: adding noise and reverberation data to the first speech data to obtain second speech data.

[0043] In this embodiment, noise and reverberation data in a natural scene are randomly added to the audio data obtained in step 101 to perform a second round of data multiplication.

[0044] Specifically, the second voice data is obtained based on y = F2(x1, v), where x1 represents the non-telephone channel voice data after the first round of data enhancement, v represents the noise and reverberation disturbance parameters, and F represents the disturbance function or algorithm.

[0045] In this embodiment, the second round of data multiplication involves adding noise and reverberation data in natural scenes to the non-telephone channel voice data. In this way, the environmental noise and reverberation effects existing in real scenes can be simulated, so that the voiceprint recognition model has better robustness. Among them, natural scene noise refers to background noise data obtained from the actual recording environment, such as street noise, wind noise, traffic noise, etc. By superimposing natural scene noise on non-telephone channel voice data, the model can perform voiceprint recognition more accurately in a noisy environment. Reverberation data refers to adding appropriate reverberation effects to voice data by simulating different scenes such as rooms and environments. Reverberation data can increase the time domain and frequency domain characteristics of the voice signal, so that the model can better adapt to different acoustic environments.

[0046] Step 103: Perform audio downsampling on the second voice data, and simulate telephone channel transmission on the downsampled data to obtain third voice data.

[0047] In this embodiment, the data in step 102 is sent to the telephone channel encoding and decoding module for telephone channel transmission simulation, and a third round of data multiplication is performed. In telephone communication, voice data is compressed, encoded and decoded to reduce file size and transmission bandwidth. This process may introduce some specific distortion effects, such as bit depth reduction, quantization error, compression loss, etc. These distortion effects have a negative impact on audio quality and clarity.

[0048] Specifically, the data obtained from the second round of enhancement is subjected to audio downsampling, and the audio can be downsampled to an 8K sampling rate.

[0049] Further, the third voice data is obtained based on y = F3 (x2, v), wherein x2 represents the non-telephone channel voice data, and v represents the data encoding and decoding mode and parameters. In the third round of data multiplication, the audio signal is simulated by the telephone channel encoding and decoding module, and the simulated audio undergoes a similar compression encoding and decoding process, which reduces its quality and introduces certain distortion. The distortion effects of this simulation may include blurring of the sound, increased noise, loss of details, etc.

[0050] The third round of data multiplication simulates the encoding and decoding process of the telephone channel to cause the audio to be damaged to a certain extent. In this way, the distortion effects commonly seen in the encoding and decoding process of telephone channel transmission are introduced, thereby increasing the robustness of the voiceprint recognition model to low-quality or damaged audio.

[0051] By introducing these distortion effects, the voiceprint recognition model can better learn to handle low-quality or damaged audio during training. This data multiplication technology can improve the robustness of the model, enabling it to better adapt to the distortion and noise conditions commonly seen in actual telephone communications.

[0052] Step 104: Based on the third voice data, a preset voiceprint recognition model is trained to use the third voice data as a training sample for training the preset voiceprint recognition model.

[0053] In this embodiment, the audio after encoding and decoding is sent to the voiceprint recognition model for training.

[0054] Reference below Figure 2 , Figure 2 A schematic diagram of a process of performing telephone channel transmission simulation on downsampled data in the present application is shown, and the process includes:

[0055] Step 201: compressing the downsampled data based on different compression methods to obtain data under different compression methods.

[0056] Step 202: Use the downsampled data and the data in different compression modes as the third voice data.

[0057] In this embodiment, in actual telephone communications, voice data is compressed, encoded, and decoded to reduce file size and transmission bandwidth. This process may introduce some specific distortion effects, such as bit depth reduction, quantization error, compression loss, etc. These distortion effects have a negative impact on audio quality and clarity.

[0058] In the third round of data multiplication in this embodiment, the signal is simulated by the telephone channel codec module, and the simulated audio undergoes a similar compression encoding and decoding process, which reduces its quality and introduces certain distortion. Such simulated distortion effects may include blurring of the sound, increased noise, loss of details, etc. The distortion effect and simulation process may vary according to the application scenario and requirements. The specific implementation of this step can be carried out according to the telephone channel codec standard, such as G.711, G.729, etc.

[0059] Specifically, refer to Figure 3 The exemplary analog telephone channel coding and decoding process may include highly impaired simulation, moderately impaired simulation, slightly impaired simulation and hybrid coding modules, which respectively enhance the original data with different degrees of signal loss to make it more consistent with the telephone channel data in real scenarios. The telephone channel coding and decoding module covers the codecs used in fixed telephone communications, satellite / radio transmission and IP voice.

[0060] Figure 3In the method, after the speed-perturbed data is multiplied to add noise and reverberation data in natural scenes, it is downsampled, and then the channel encoding and decoding module is used to simulate different degrees of damage and mixed coding. The simulated damaged data of different degrees, the downsampled data, and the mixed coding data are decoded and input into the preset voiceprint recognition model to train the preset model voiceprint recognition model.

[0061] Reference below Figure 4 , Figure 4 A schematic diagram of a process of performing telephone channel transmission simulation on downsampled data in the present application is shown, and the process includes:

[0062] Step 401: performing telephone channel coding on the downsampled data to obtain coded audio data;

[0063] Step 402: Performing a network packet loss simulation on the encoded audio data to obtain fifth voice data;

[0064] Step 403: performing packet repetition simulation on the encoded audio data to obtain sixth voice data;

[0065] Step 404: performing packet damage simulation on the encoded audio data to obtain seventh voice data;

[0066] Step 405: performing packet disorder simulation on the encoded audio data to obtain eighth voice data;

[0067] Step 406: Decode the fifth voice data, the sixth voice data, the seventh voice data, and the eighth voice data, and use the decoded data as the third voice data.

[0068] Reference below Figure 5 , Figure 5 A schematic diagram of a process of performing audio downsampling on the second voice data and performing telephone channel transmission simulation on the downsampled data to obtain the third voice data in the present application is shown. The process includes:

[0069] Step 501: Perform clipping simulation on the second voice data to obtain fourth voice data.

[0070] Clipping is a common form of audio distortion that limits the signal once a threshold is exceeded. This distortion can occur in a variety of situations. For example, when a sensor records a signal, the signal may be clipped due to exceeding the data range it can measure. In addition, clipping can also occur when a signal is digitized, or any time an analog or digital signal conversion occurs. This distortion phenomenon is particularly common in situations where gain is present or overshoot and undershoot are present. The specific audio clipping simulation process is as follows: Figure 7As shown, when the audio signal is higher than the limiting threshold A, the audio is truncated.

[0071] As an optional implementation method of this embodiment, performing clipping simulation on the second voice data to obtain fourth voice data includes: determining data in a specific time interval in the second voice data based on a random mask; and performing plosive sound simulation on the data in the specific time interval to obtain fourth voice data.

[0072] In this embodiment, the original audio is subjected to a random mask module to select a specific time interval. The selected interval is subjected to a plosive sound simulation through an audio clipping simulation;

[0073] Step 502: Perform audio downsampling on the fourth voice data, and perform telephone channel coding on the downsampled data to obtain encoded audio data.

[0074] Step 503: Performing a network packet loss simulation on the encoded audio data to obtain fifth voice data;

[0075] Step 504: performing packet repetition simulation on the encoded audio data to obtain sixth voice data;

[0076] Step 505: performing packet damage simulation on the encoded audio data to obtain seventh voice data;

[0077] Step 506: performing packet disorder simulation on the encoded audio data to obtain eighth voice data;

[0078] Step 507: Decode the fifth voice data, the sixth voice data, the seventh voice data, and the eighth voice data, and use the decoded data as the third voice data.

[0079] In this embodiment, reference Figure 6 The data enhancement workflow diagram of packet loss and clipping audio simulation based on random mask is shown in the figure. The audio passes through the random mask module to select a specific time interval, and the audio clipping simulation is used to simulate the plosive sound of the selected interval. At the same time, the network packet loss simulation is used to increase the simulation of the frame loss signal. In addition, it also includes a packet duplication simulation module, a packet damage simulation module, and a packet disorder simulation module. The audio is then sent to the decoder module for decoding and sent to the back end of the voiceprint training model.

[0080] Reference below Figure 8 , Figure 8 A flow chart of an embodiment of an information processing method for telephone speech recognition in the present application is shown, comprising:

[0081] Step 801: Extract telephone voice data.

[0082] In this embodiment, reference Fig. 9 , the voice data of the phone can be obtained from the Internet and resampled to 8K sampling rate, and then the effective audio sound (i.e. human voice) can be extracted based on the VAD module.

[0083] Step 802: Input the telephone voice data into the trained voiceprint recognition model to determine the voiceprint feature vector in the telephone voice data.

[0084] In this embodiment, after the human voice is extracted, the audio data is sent to the voiceprint recognition model trained in the above embodiment, so that the model outputs the voiceprint feature vector of the telephone channel speech. .

[0085] As an optional implementation of this embodiment, after extracting the telephone voice data, the method further performs echo elimination on the telephone voice data, including: based on a high-pass filter, weakening the data in the telephone voice data whose frequency is lower than a preset cutoff frequency, and transmitting the telephone voice signal with a frequency higher than or equal to the preset cutoff frequency; based on a preset echo elimination rule, adaptively filtering the telephone voice signal in the telephone voice signal whose frequency is lower than the preset cutoff frequency; suppressing the echo signal of a preset frequency band in the telephone voice signal after adaptive filtering to obtain a signal after echo elimination.

[0086] In this optional implementation, reference Figure 8 and Fig. 9 The goal of echo cancellation in voice IP phones is to reduce the echo generated in telephone communications. When the audio signal experiences a certain delay in transmission, an echo effect will occur. According to the ITU-T P.861 standard, the frequency range of human voice in telephone communications is usually 300Hz-3400Hz. In order to reduce the impact of echo, a high-pass filter (HPF) can be used to process the audio signal. The high-pass filter reduces the part below a certain cutoff frequency (fc) to reduce communication delays and reduce the generation of echoes. Specifically, the high-pass filter can pass input signals with frequencies greater than or equal to the cutoff frequency (fc) while attenuating input signals with frequencies less than or equal to the cutoff frequency (fc).

[0087] In the voice IP phone, the echo cancellation method can be implemented through the following steps. The specific process is as follows: Fig. 9As shown: There are two IP phone devices in the figure, marked as Phone 1 and Phone 2. Phone 1's microphone is connected to the transmitter (Tx-1), and the receiver (Rx-1) is used as a speaker, and the speaker is connected to the receiver (Rx-1). Phone 2 also has a corresponding transmitter (Tx-2) and receiver (Rx-2). A high-pass filter is connected to the signal transmitter (Tx-1) to send out the voice signal with a frequency higher than 300Hz-3400Hz, thereby attenuating the low-frequency part while retaining the high-frequency information. In the echo cancellation module, a filter bank is used to stabilize and filter the telephone signal. These filters can appropriately process the voice signal to achieve echo cancellation.

[0088] The adaptive echo cancellation module is used to perform adaptive filtering operations on audio signals in the lower frequency band to achieve echo cancellation. The module dynamically adjusts the filter tap coefficient h^(n) according to the error signal e(n) using the minimum mean square error criterion, so that the output signal y^(n) of the filter is close to the expected signal y(n).

[0089] The residual echo cancellation module is used to suppress the echo signal in the higher frequency band and output the signal after the echo is cancelled.

[0090] As an optional implementation method of this embodiment, the adaptive filtering of the signal with a frequency lower than a preset cutoff frequency in the telephone voice signal based on a preset echo cancellation rule includes: dynamically adjusting the tap coefficient of the high-pass filter with the goal of minimizing the mean square value of the difference between the telephone voice signal output by the high-pass filter and the expected signal.

[0091] In this optional implementation, during the adaptive process, the adaptive algorithm continuously adjusts the filter tap coefficient h^(n) according to the error signal e(n) using the minimum mean square error criterion to minimize the difference between the filter output signal y^(n) and the desired signal y(n). The minimum mean square error criterion requires that the mean square value of the difference between the filter output and the desired signal be minimized, thereby achieving the goal of reducing the echo effect. Therefore, by dynamically adjusting the filter tap coefficient, the filter output signal is matched with the desired signal as much as possible.

[0092] Specifically, in the adaptive process, the adaptive algorithm dynamically adjusts the filter tap coefficient h^ according to the error signal e (n) through the minimum mean square error criterion, and finally makes the filter output signal y^(n) infinitely close to y(n). According to the minimum mean square error criterion, which requires that the mean square value of the difference between the filter output and the expected signal is minimized, the objective function (performance function) is defined as:

[0093] ,

[0094] in is the autocorrelation function of the original audio signal, and is the cross-correlation function between the input signal and the reference signal. Select the optimal weight coefficient h ^ to minimize the objective function ε(n). When the following formula is equal to 0, the obtained filter optimal weight vector is also the Wiener optimal solution. When the weight coefficient of the adaptive filter satisfies the above formula, the mean square error reaches the minimum value.

[0095]

[0096] This optional implementation can effectively reduce the echo problem in voice IP phones by using a high-pass filter and an adaptive echo cancellation algorithm. This method can provide a clearer and more understandable communication experience, improve call quality, and make the recognized voice data more accurate.

[0097] The present application also provides an information processing device for telephone voice recognition, which includes: a first processing unit, configured to obtain non-telephone channel voice data, and perform random speed perturbation on the non-telephone channel voice to obtain first voice data; a second processing unit, configured to add noise and reverberation data to the first voice data to obtain second voice data; a third processing unit, configured to perform audio downsampling on the second voice data, and simulate telephone channel transmission on the downsampled data to obtain third voice data; a training unit, configured to train a preset voiceprint recognition model based on the third voice data, so as to train the preset voiceprint recognition model using the third voice data as a training sample.

[0098] As an optional implementation method of this embodiment, simulating telephone channel transmission of the downsampled data includes: compressing the downsampled data based on different compression methods to obtain data under different compression methods; and using the downsampled data and the data under different compression methods as the third voice data.

[0099] As an optional implementation method of this embodiment, the second voice data is audio downsampled, and the second voice data is clipped to obtain fourth voice data; the fourth voice data is audio downsampled, and the sampled data is simulated for telephone channel transmission to obtain third voice data.

[0100] As an optional implementation method of this embodiment, the simulating telephone channel transmission of the downsampled data to obtain the third voice data includes: performing telephone channel encoding on the downsampled data to obtain encoded audio data; performing network packet loss simulation on the encoded audio data to obtain fifth voice data; performing packet duplication simulation on the encoded audio data to obtain sixth voice data; performing packet damage simulation on the encoded audio data to obtain seventh voice data; performing packet disorder simulation on the encoded audio data to obtain eighth voice data; decoding the fifth voice data, the sixth voice data, the seventh voice data, and the eighth voice data, and using the decoded data as the third voice data.

[0101] As an optional implementation method of this embodiment, clipping simulation is performed on the second voice data to obtain fourth voice data: data in a specific time interval in the second voice data is determined based on a random mask; and plosive sound simulation is performed on the data in the specific time interval to obtain fourth voice data.

[0102] As an optional implementation of this embodiment, the device also includes: an extraction unit configured to extract telephone voice data; input the telephone voice data into a trained voiceprint recognition model to determine the voiceprint feature vector in the telephone voice data.

[0103] As an optional implementation method of this embodiment, after extracting the telephone voice data, the device also performs echo elimination on the telephone voice data, including: based on a high-pass filter, weakening the data in the telephone voice data whose frequency is lower than a preset cutoff frequency, and transmitting the telephone voice signal with a frequency higher than or equal to the preset cutoff frequency; based on a preset echo elimination rule, adaptively filtering the existing telephone voice signal with a frequency lower than the preset cutoff frequency; suppressing the echo signal of a preset frequency band in the data after adaptive filtering to obtain a signal after echo elimination.

[0104] As an optional implementation method of this embodiment, the adaptive filtering of data with a frequency lower than a preset cutoff frequency based on a preset echo cancellation rule includes: dynamically adjusting the tap coefficients of the high-pass filter with the goal of minimizing the mean square value of the difference between the telephone voice signal output by the high-pass filter and the expected signal.

[0105] Below, reference Fig.11 An electronic device according to an embodiment of the present disclosure is described. Fig.11 This is a schematic diagram of the structure of an application embodiment of the electronic device provided by the present application. Fig.11 An electronic device according to an embodiment of the present disclosure is described.

[0106] like Fig.11 As shown, the electronic device includes one or more processors and memory.

[0107] The processor may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions.

[0108] The memory may include one or more computer program products, and the computer program product may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, a random access memory (RAM) and / or a cache memory (cache), etc. The non-volatile memory may include, for example, a read-only memory (ROM), a hard disk, a flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor may execute the program instructions to implement the task generation method of each embodiment of the present application described above and / or other desired functions.

[0109] In one example, the electronic device may further include: an input device and an output device, and these components are interconnected via a bus system and / or other forms of connection mechanisms (not shown).

[0110] In addition, the input device may also include, for example, a keyboard, a mouse, and the like.

[0111] The output device can output various information to the outside, including the determined distance information, direction information, etc. The output device can include, for example, a display, a speaker, a printer, a communication network and a remote output device connected thereto, and the like.

[0112] Of course, to simplify, Fig.11 Only some of the components in the electronic device related to the present application are shown, and components such as a bus, an input / output interface, etc. are omitted. In addition, the electronic device may further include any other appropriate components according to specific application conditions.

[0113] In addition to the above-mentioned methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the steps of the task generation method according to various embodiments of the present application described in the above part of this specification.

[0114] The computer program product may be written in any combination of one or more programming languages ​​to write program codes for performing the operations of the embodiments of the present application, including object-oriented programming languages, such as Java, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0115] In addition, an embodiment of the present application may also be a computer-readable storage medium on which computer program instructions are stored. When the computer program instructions are executed by a processor, the processor executes the steps of the task generation method according to various embodiments of the present application described in the above part of this specification.

[0116] The computer readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can include, for example, but is not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0117] A person of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk, etc., various media that can store program codes.

[0118] The basic principles of the present application are described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, effects, etc. mentioned in the present application are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. are required by each embodiment of the present application. In addition, the specific details disclosed above are only for the purpose of illustration and ease of understanding, not for limitation, and the above details do not limit the present application to being implemented by adopting the above specific details.

[0119] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An information processing method for telephone speech recognition, characterized in that: include: Acquire non-telephone channel voice data, and perform random speed perturbation on the non-telephone channel voice data to obtain first voice data; adding noise and reverberation data to the first speech data to obtain second speech data; Performing audio downsampling on the second voice data, and simulating telephone channel transmission on the downsampled data to obtain third voice data, and obtaining the third voice data based on y=F3(x2,v), wherein x2 represents non-telephone channel voice data, and v represents data encoding and decoding mode and parameters; Based on the third voice data, training a preset voiceprint recognition model; After the training is completed, the method further includes: extracting telephone voice data; inputting the telephone voice data into the trained voiceprint recognition model to determine the voiceprint feature vector in the telephone voice data, wherein after extracting the telephone voice data, the method further performs echo elimination on the telephone voice data, including: weakening the data with a frequency lower than a preset cutoff frequency in the telephone voice data based on a high-pass filter, and transmitting the telephone voice signal with a frequency higher than or equal to the preset cutoff frequency; based on a preset echo elimination rule, adaptively filtering the signal with a frequency lower than the preset cutoff frequency in the telephone voice signal; suppressing the echo signal of a preset frequency band in the telephone voice signal after adaptive filtering to obtain a signal after echo elimination; The adaptive filtering of the signal with a frequency lower than the preset cutoff frequency in the telephone voice signal based on the preset echo cancellation rule includes: dynamically adjusting the tap coefficient of the high-pass filter with the goal of minimizing the mean square value of the difference between the telephone voice signal output by the high-pass filter and the expected signal; In the adaptive process, the adaptive algorithm dynamically adjusts the filter tap coefficient h^ according to the error signal e(n) through the minimum mean square error criterion, and finally makes the filter output signal y^(n) infinitely close to y(n). According to the minimum mean square error criterion, the objective function is defined as: Where R x =E[x(n)x T (n)] is the autocorrelation function of the original audio signal, and R dx =E[d(n)x T (n)] is the cross-correlation function between the input signal and the reference signal. The optimal weight coefficient h^ is selected to minimize the objective function ε(n). When the following formula is equal to 0, the optimal weight vector of the filter obtained is also the Wiener optimal solution. When the weight coefficient of the adaptive filter satisfies the above formula, the mean square error reaches the minimum value.

2. The information processing method for telephone speech recognition according to claim 1, characterized in that: The telephone channel transmission simulation of the downsampled data comprises: Based on different compression methods, compressing the downsampled data to obtain data in different compression methods; The down-sampled data and the data in the different compression modes are used as the third voice data.

3. The information processing method for telephone speech recognition according to claim 1, characterized in that: Downsampling the second voice data, and simulating telephone channel transmission on the downsampled data to obtain the third voice data includes: Performing clipping simulation on the second voice data to obtain fourth voice data; The fourth voice data is audio-downsampled, and the sampled data is subjected to telephone channel transmission simulation to obtain third voice data.

4. The information processing method for telephone speech recognition according to claim 1 or 3, characterized in that: The step of performing telephone channel transmission simulation on the downsampled data to obtain the third voice data comprises: Performing telephone channel coding on the downsampled data to obtain encoded audio data; Performing a network packet loss simulation on the encoded audio data to obtain fifth voice data; Performing packet repetition simulation on the encoded audio data to obtain sixth voice data; Performing packet damage simulation on the encoded audio data to obtain seventh voice data; Performing packet disorder simulation on the encoded audio data to obtain eighth voice data; The fifth voice data, the sixth voice data, the seventh voice data, and the eighth voice data are decoded, and the decoded data are used as the third voice data.

5. The information processing method for telephone speech recognition according to claim 3, characterized in that: The second voice data is subjected to clipping simulation to obtain fourth voice data: Determining data in a specific time interval in the second voice data based on the random mask; Perform plosive sound simulation on the data in the specific time interval to obtain fourth voice data.

6. An information processing device for telephone speech recognition, characterized in that: include: A first processing unit is configured to obtain non-telephone channel voice data and perform random speed disturbance on the non-telephone channel voice data to obtain first voice data; A second processing unit is configured to add noise and reverberation data to the first voice data to obtain second voice data; A third processing unit is configured to perform audio downsampling on the second voice data, and perform telephone channel transmission simulation on the downsampled data to obtain third voice data, and obtain the third voice data based on y=F3(x2,v), wherein x2 represents non-telephone channel voice data, and v represents data encoding and decoding mode and parameters; A training unit is configured to train a preset voiceprint recognition model based on the third voice data, so as to train the preset voiceprint recognition model using the third voice data as a training sample; After the training is completed, the device also extracts telephone voice data; inputs the telephone voice data into the trained voiceprint recognition model to determine the voiceprint feature vector in the telephone voice data, wherein after the telephone voice data is extracted, the method also performs echo elimination on the telephone voice data, including: based on a high-pass filter, weakening the data with a frequency lower than a preset cutoff frequency in the telephone voice data, and transmitting the telephone voice signal with a frequency higher than or equal to the preset cutoff frequency; based on a preset echo elimination rule, adaptively filtering the signal with a frequency lower than the preset cutoff frequency in the telephone voice signal; suppressing the echo signal of a preset frequency band in the telephone voice signal after adaptive filtering to obtain a signal after echo elimination; The adaptive filtering of the signal with a frequency lower than the preset cutoff frequency in the telephone voice signal based on the preset echo cancellation rule includes: dynamically adjusting the tap coefficient of the high-pass filter with the goal of minimizing the mean square value of the difference between the telephone voice signal output by the high-pass filter and the expected signal; In the adaptive process, the adaptive algorithm dynamically adjusts the filter tap coefficient h^ according to the error signal e(n) through the minimum mean square error criterion, and finally makes the filter output signal y^(n) infinitely close to y(n). According to the minimum mean square error criterion, the objective function is defined as: Where R x =E[x(n)x T (n)] is the autocorrelation function of the original audio signal, and R dx =E[d(n)x T (n)] is the cross-correlation function between the input signal and the reference signal. The optimal weight coefficient h^ is selected to minimize the objective function ε(n). When the following formula is equal to 0, the optimal weight vector of the filter obtained is also the Wiener optimal solution. When the weight coefficient of the adaptive filter satisfies the above formula, the mean square error reaches the minimum value.

7. An electronic device, characterized in that: include: a memory for storing a computer program product; A processor is used to execute the computer program product stored in the memory, and when the computer program product is executed, it implements the method described in any one of claims 1 to 5.