Human voice similarity detection method and system and storage medium

Abstract

The invention provides a human voice similarity detection method and system and a storage medium, and the method comprises the steps: obtaining a to-be-detected audio, and carrying out the preprocessing of the to-be-detected audio; performing feature extraction on the preprocessed to-be-detected audio to obtain audio features, and inputting the audio features into a pre-trained acoustic model for phoneme matching to obtain a vowel phoneme sequence; determining formant data of the audio to be detected according to the timestamp of each vowel phoneme in the vowel phoneme sequence, and determining a formant value of each vowel phoneme according to the formant data; and according to the formant value, calculating a formant difference value between the same vowel phonemes in the different to-be-detected audios, and according to the formant difference value, determining the human voice similarity between the different to-be-detected audios. According to the embodiment of the invention, the human voice similarity between different to-be-detected audios can be effectively determined based on the formant difference value, and the accuracy of human voice similarity detection is improved.

Description

Technical Field

[0001] The present invention relates to the field of speech recognition technology, and in particular to a human voice similarity detection method, system and storage medium. Background Art

[0002] Speech recognition technology refers to the technology that converts human speech into text or instructions. With the continuous optimization of speech recognition algorithms and the improvement of computing power, speech recognition technology has been able to achieve high-precision speech-to-text functions. As a branch of speech recognition technology, human voice similarity detection focuses on similarity analysis of two or more speech segments to determine whether the audio is from the same speaker or has similar sound characteristics.

[0003] In the existing human voice similarity detection process, human voice similarity detection is generally performed by manually selecting appropriate audio segments in the audio, which leads to low accuracy of human voice similarity detection. Summary of the invention

[0004] The purpose of the embodiments of the present invention is to provide a method, system and storage medium for detecting human voice similarity, so as to solve the problem of low accuracy of human voice similarity detection in the prior art.

[0005] The embodiment of the present invention is implemented as follows: a method for detecting human voice similarity, the method comprising:

[0006] Acquire the audio to be detected, and preprocess the audio to be detected;

[0007] Extracting features from the preprocessed audio to be detected to obtain audio features, and inputting the audio features into a pre-trained acoustic model for phoneme matching to obtain a vowel phoneme sequence;

[0008] Determine the formant data of the audio to be detected according to the timestamp of each vowel phoneme in the vowel phoneme sequence, and determine the formant value of each vowel phoneme according to the formant data;

[0009] According to the formant values, the formant differences between the same vowel phonemes in different audios to be detected are calculated, and the human voice similarities between the different audios to be detected are determined according to the formant differences.

[0010] Preferably, determining the formant data of the audio to be detected according to the timestamp of each vowel phoneme in the vowel phoneme sequence includes:

[0011] Segment the audio to be detected according to the timestamp of each vowel phoneme in the vowel phoneme sequence to obtain vowel audio, and perform spectrum analysis on the vowel audio to obtain a frequency domain signal;

[0012] Formant extraction is performed on the frequency domain signal to obtain formant information, and in the same audio to be detected, the formant information is classified according to the phoneme type of each vowel phoneme to obtain the formant data.

[0013] Preferably, performing resonance peak extraction on the frequency domain signal to obtain resonance peak information includes:

[0014] Calculating a power spectrum according to the frequency domain signal, and performing a logarithmic operation on the power spectrum to obtain a logarithmic power spectrum;

[0015] The logarithmic power spectrum is subjected to inverse Fourier transformation to obtain a cepstrum domain signal, and the cepstrum domain signal is subjected to low-pass filtering to obtain the resonance peak information.

[0016] Preferably, determining the human voice similarity between different audios to be detected according to the formant difference includes:

[0017] Respectively obtain the similarity weight of each vowel phoneme, and calculate the similarity score between different audios to be detected according to the similarity weight and the formant difference;

[0018] The similarity score is matched with a similarity query table to obtain the voice similarity.

[0019] Preferably, after calculating the similarity scores between different audios to be detected according to the similarity weights and the formant differences, the method further includes:

[0020] Performing differential processing on the audio features to obtain feature differential results, and performing binarization processing on the audio features according to the feature differential results to obtain feature vectors;

[0021] Calculating the similarity of the feature vectors between different audios to be detected to obtain vector similarity, and performing a weighted operation on the vector similarity and the similarity score to obtain a similarity weighted value;

[0022] The similarity weighted value is matched with the similarity query table to obtain the human voice similarity.

[0023] Preferably, after inputting the audio features into the pre-trained acoustic model for phoneme matching to obtain a vowel phoneme sequence, the method further includes:

[0024] Matching the vowel phoneme sequence with a preset vowel group to obtain a target vowel string, and determining a target audio segment in the audio to be detected according to the vowel positions of the target vowel string;

[0025] Perform audio replacement on the target audio segment corresponding to the audio to be detected.

[0026] Preferably, before the audio features are input into the pre-trained acoustic model for phoneme matching to obtain the vowel phoneme sequence, the method further comprises:

[0027] Acquire an audio sample, and input the audio sample into the acoustic model for feature extraction to obtain sample features;

[0028] Performing phoneme prediction on the sample features to obtain a predicted vowel phoneme string, and calculating a model loss based on the predicted vowel phoneme string;

[0029] The parameters of the acoustic model are updated according to the model loss until the acoustic model converges, thereby obtaining the pre-trained acoustic model.

[0030] Another object of an embodiment of the present invention is to provide a voice similarity detection system, the system comprising:

[0031] A preprocessing module, used to obtain the audio to be detected and preprocess the audio to be detected;

[0032] A phoneme matching module is used to extract features from the pre-processed audio to be detected to obtain audio features, and input the audio features into a pre-trained acoustic model for phoneme matching to obtain a vowel phoneme sequence;

[0033] A formant determination module, used to determine the formant data of the audio to be detected according to the timestamp of each vowel phoneme in the vowel phoneme sequence, and determine the formant value of each vowel phoneme according to the formant data;

[0034] The similarity calculation module is used to calculate the formant difference between the same vowel phonemes in different audios to be detected according to the formant value, and determine the human voice similarity between the different audios to be detected according to the formant difference.

[0035] Preferably, the resonance peak determination module is further used for:

[0036] Segment the audio to be detected according to the timestamp of each vowel phoneme in the vowel phoneme sequence to obtain vowel audio, and perform spectrum analysis on the vowel audio to obtain a frequency domain signal;

[0037] Formant extraction is performed on the frequency domain signal to obtain formant information, and in the same audio to be detected, the formant information is classified according to the phoneme type of each vowel phoneme to obtain the formant data.

[0038] The embodiment of the present invention can automatically locate vowel phonemes in the audio to be detected by inputting audio features into a pre-trained acoustic model for phoneme matching, and obtain a vowel phoneme sequence. Based on the timestamp of each vowel phoneme in the vowel phoneme sequence, the resonance peak data of the audio to be detected can be effectively determined. Based on the resonance peak data, the resonance peak value of each vowel phoneme can be automatically determined. Based on the resonance peak value, the resonance peak difference between the same vowel phonemes in different audios to be detected can be effectively calculated. Based on the resonance peak difference, the human voice similarity between different audios to be detected can be effectively determined, thereby improving the accuracy of human voice similarity detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is a flow chart of a method for detecting voice similarity provided by a first embodiment of the present invention;

[0040] Figure 2 is a structural diagram of a voice similarity detection system provided by a second embodiment of the present invention;

[0041] Figure 3 is a schematic diagram of specific implementation steps of a voice similarity detection system provided by a second embodiment of the present invention;

[0042] Figure 4 It is a schematic diagram of the structure of a terminal device provided in the third embodiment of the present invention. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0044] In order to illustrate the technical solution of the present invention, a specific embodiment is provided below for illustration.

[0045] Embodiment 1

[0046] See also Figure 1 , is a flow chart of a method for detecting human voice similarity provided by a first embodiment of the present invention. The method for detecting human voice similarity can be applied to any device or system. The method for detecting human voice similarity comprises the steps of:

[0047] Step S10, obtaining the audio to be detected and preprocessing the audio to be detected;

[0048] Among them, the number of audios to be detected is greater than or equal to two, and the preprocessing of the audio to be detected includes noise reduction, gain adjustment, sampling rate conversion and pre-emphasis, etc. By performing noise reduction processing on the audio to be detected, the noise components in the audio signal can be effectively reduced or eliminated to improve the signal-to-noise ratio of the signal. Noise reduction methods include the use of filters (such as low-pass, high-pass, band-pass or band-stop filters), spectral subtraction, Wiener filtering, etc. Gain adjustment is used to change the amplitude of the audio signal to achieve the desired volume level. Sampling rate conversion (also known as resampling) involves changing the sampling rate of the audio signal to a different value. Pre-emphasis aims to increase the energy of high-frequency components to compensate for the attenuation of high-frequency components of the voice signal during transmission. The audio quality of the audio to be detected is effectively improved by preprocessing the audio to be detected.

[0049] Step S20, extracting features from the preprocessed audio to be detected to obtain audio features, and inputting the audio features into a pre-trained acoustic model for phoneme matching to obtain a vowel phoneme sequence;

[0050] Among them, audio features include time domain features (such as energy, zero-crossing rate, etc.), frequency domain features (such as spectrum, Mel-frequency cepstral coefficients, etc.) or time-frequency domain features. By inputting audio features into the pre-trained acoustic model for phoneme matching, the effective audio in the audio to be detected can be automatically located and the phonemes converted, without the need to manually select the appropriate audio segment in the audio, thereby improving the accuracy of voice similarity detection.

[0051] Optionally, after inputting the audio features into a pre-trained acoustic model for phoneme matching to obtain a vowel phoneme sequence, the method further includes:

[0052] Matching the vowel phoneme sequence with a preset vowel group to obtain a target vowel string, determining a target audio segment in the audio to be detected according to the vowel positions of the target vowel string, and performing audio replacement on the target audio segment corresponding to the audio to be detected;

[0053] Among them, the preset vowel group can be set according to needs, and the preset vowel group corresponds to a preset word or a preset character. The vowel phoneme sequence is matched with the preset vowel group to determine whether there is a preset word or a preset character corresponding to the preset vowel group in the vowel phoneme sequence, and the vowel phonemes in the vowel phoneme sequence that match the preset vowel group are determined as the target vowel string, and the target timestamp is determined according to the vowel position of the target vowel string, and audio acquisition is performed on the audio to be detected based on the target timestamp to obtain the target audio segment.

[0054] In this step, by performing audio replacement on the target audio segment corresponding to the audio to be detected, the target audio segment corresponding to the preset word or preset character can be effectively set as the audio to be detected for voice similarity according to user needs.

[0055] Furthermore, before the audio features are input into the pre-trained acoustic model for phoneme matching to obtain a vowel phoneme sequence, the method further includes:

[0056] Acquire an audio sample, and input the audio sample into the acoustic model for feature extraction to obtain sample features; wherein the audio sample is input into an encoder in the acoustic model for feature encoding to obtain sample features;

[0057] Performing phoneme prediction on the sample features to obtain a predicted vowel phoneme string, and calculating a model loss based on the predicted vowel phoneme string; wherein, performing feature decoding on the sample features according to a decoder in the acoustic model to obtain a predicted vowel phoneme string, and performing loss calculation on the predicted vowel phoneme string and the sample label of the audio sample according to a preset loss function to obtain a model loss;

[0058] The parameters of the acoustic model are updated according to the model loss until the acoustic model converges to obtain the pre-trained acoustic model; wherein the model loss is gradient propagated in the acoustic model by back propagation to achieve the effect of updating the acoustic model parameters.

[0059] Step S30, determining the formant data of the audio to be detected according to the timestamp of each vowel phoneme in the vowel phoneme sequence, and determining the formant value of each vowel phoneme according to the formant data;

[0060] Among them, when outputting a vowel phoneme sequence, each vowel phoneme in the vowel phoneme sequence is timestamped, and based on the timestamp of each vowel phoneme, the resonance peak data of the audio to be detected can be effectively determined, and based on the resonance peak data of the audio to be detected, the resonance peak value of each vowel phoneme can be effectively determined.

[0061] Optionally, determining the formant data of the audio to be detected according to the timestamp of each vowel phoneme in the vowel phoneme sequence includes:

[0062] According to the timestamp of each vowel phoneme in the vowel phoneme sequence, the audio to be detected is segmented to obtain vowel audio, and the vowel audio is subjected to spectrum analysis to obtain a frequency domain signal; wherein the timestamp of each vowel phoneme is aligned with the audio time in the audio to be detected, the detection audio is segmented according to the alignment result to obtain vowel audio, and by performing spectrum analysis on the vowel audio, the vowel audio can be effectively converted from a time domain signal to a frequency domain signal;

[0063] Resonant peaks are extracted from the frequency domain signal to obtain resonance peak information, and in the same audio to be detected, the resonance peak information is classified according to the phoneme type of each vowel phoneme to obtain the resonance peak data; wherein, the resonance peak coordinates in the frequency domain signal are obtained, and the frequency domain signal is divided into regions according to the resonance peak coordinates to obtain the resonance peak information, and in the same audio to be detected, each vowel phoneme is classified and stored to obtain the resonance peak data.

[0064] Furthermore, resonance peaks are extracted from the frequency domain signal to obtain resonance peak information, including: calculating a power spectrum based on the frequency domain signal, and performing a logarithmic operation on the power spectrum to obtain a logarithmic power spectrum; performing an inverse Fourier transform on the logarithmic power spectrum to obtain a cepstrum domain signal, and performing a low-pass filtering on the cepstrum domain signal to obtain the resonance peak information.

[0065] Step S40, calculating the formant differences between the same vowel phonemes in different audios to be detected according to the formant values, and determining the human voice similarity between the different audios to be detected according to the formant differences;

[0066] Among them, in different audios to be detected, the larger the difference in resonance peaks between the same vowel phonemes, the lower the similarity between the corresponding different audios to be detected, and the smaller the difference in resonance peaks between the same vowel phonemes, the greater the similarity between the corresponding different audios to be detected.

[0067] Optionally, determining the human voice similarity between different audios to be detected according to the formant difference includes:

[0068] Respectively obtain the similarity weight of each vowel phoneme, calculate the similarity score between different audios to be detected according to the similarity weight and the formant difference, and match the similarity score with the similarity query table to obtain the human voice similarity;

[0069] Among them, each vowel phoneme is matched with a weight query table to obtain a similarity weight. The weight query table stores the correspondence between different vowel phonemes and corresponding similarity weights. A weighted operation is performed based on the similarity weight of each vowel phoneme and the resonance peak difference to obtain a similarity score between different audios to be detected, and the similarity score is matched with the similarity query table to obtain the human voice similarity. The similarity query table stores the correspondence between different similarity scores and corresponding human voice similarities.

[0070] Further, after calculating the similarity scores between different audios to be detected according to the similarity weights and the formant differences, the method further includes:

[0071] The audio features are differentially processed to obtain feature differential results, and the audio features are binarized according to the feature differential results to obtain feature vectors; wherein, in the same audio feature, adjacent features of different dimensions are differentially processed to obtain feature differential results, the feature differential results are binarized to obtain binary characters, and the binary characters are vectorized to obtain feature vectors;

[0072] Calculate the similarity of the feature vectors between the different audios to be detected to obtain the vector similarity, perform weighted operation on the vector similarity and the similarity score to obtain a similarity weighted value, and match the similarity weighted value with the similarity query table to obtain the human voice similarity; wherein, according to the Euclidean distance formula, calculate the distance between the feature vectors of the different audios to be detected to obtain the vector similarity, perform weighted operation on the vector similarity and the similarity score to obtain a similarity weighted value, and during the weighted operation, the weighted values ​​of the vector similarity and the similarity score can be set according to needs.

[0073] In this embodiment, by inputting audio features into a pre-trained acoustic model for phoneme matching, the vowel phonemes in the audio to be detected can be automatically located to obtain a vowel phoneme sequence. Based on the timestamp of each vowel phoneme in the vowel phoneme sequence, the resonance peak data of the audio to be detected can be effectively determined. Based on the resonance peak data, the resonance peak value of each vowel phoneme can be automatically determined. Based on the resonance peak value, the resonance peak difference between the same vowel phonemes in different audios to be detected can be effectively calculated. Based on the resonance peak difference, the human voice similarity between different audios to be detected can be effectively determined, thereby improving the accuracy of human voice similarity detection.

[0074] Embodiment 2

[0075] See also Figure 2 , is a schematic diagram of the structure of a voice similarity detection system 100 provided in a second embodiment of the present invention, comprising:

[0076] The preprocessing module 10 is used to obtain the audio to be detected and preprocess the audio to be detected.

[0077] The phoneme matching module 11 is used to extract features from the pre-processed audio to be detected to obtain audio features, and input the audio features into the pre-trained acoustic model for phoneme matching to obtain a vowel phoneme sequence.

[0078] Optionally, the phoneme matching module 11 is further used to: match the vowel phoneme sequence with a preset vowel group to obtain a target vowel string, and determine a target audio segment in the audio to be detected according to the vowel positions of the target vowel string;

[0079] Perform audio replacement on the target audio segment corresponding to the audio to be detected.

[0080] Furthermore, the phoneme matching module 11 is also used to: obtain an audio sample, and input the audio sample into the acoustic model for feature extraction to obtain sample features;

[0081] Performing phoneme prediction on the sample features to obtain a predicted vowel phoneme string, and calculating a model loss based on the predicted vowel phoneme string;

[0082] The parameters of the acoustic model are updated according to the model loss until the acoustic model converges, thereby obtaining the pre-trained acoustic model.

[0083] The formant determination module 12 is used to determine the formant data of the audio to be detected according to the timestamp of each vowel phoneme in the vowel phoneme sequence, and determine the formant value of each vowel phoneme according to the formant data.

[0084] Optionally, the formant determination module 12 is further used to: segment the audio to be detected according to the timestamp of each vowel phoneme in the vowel phoneme sequence to obtain vowel audio, and perform spectrum analysis on the vowel audio to obtain a frequency domain signal;

[0085] Formant extraction is performed on the frequency domain signal to obtain formant information, and in the same audio to be detected, the formant information is classified according to the phoneme type of each vowel phoneme to obtain the formant data.

[0086] Furthermore, the resonance peak determination module 12 is further used to: calculate a power spectrum according to the frequency domain signal, and perform a logarithmic operation on the power spectrum to obtain a logarithmic power spectrum;

[0087] The logarithmic power spectrum is subjected to inverse Fourier transformation to obtain a cepstrum domain signal, and the cepstrum domain signal is subjected to low-pass filtering to obtain the resonance peak information.

[0088] The similarity calculation module 13 is used to calculate the formant difference between the same vowel phonemes in different audios to be detected according to the formant value, and determine the human voice similarity between the different audios to be detected according to the formant difference.

[0089] Optionally, the similarity calculation module 13 is further used to: respectively obtain the similarity weight of each vowel phoneme, and calculate the similarity score between different audios to be detected according to the similarity weight and the formant difference;

[0090] The similarity score is matched with a similarity query table to obtain the voice similarity.

[0091] Furthermore, the similarity calculation module 13 is also used to: perform differential processing on the audio features to obtain feature differential results, and perform binarization processing on the audio features according to the feature differential results to obtain feature vectors;

[0092] Calculating the similarity of the feature vectors between different audios to be detected to obtain vector similarity, and performing a weighted operation on the vector similarity and the similarity score to obtain a similarity weighted value;

[0093] The similarity weighted value is matched with the similarity query table to obtain the human voice similarity.

[0094] See also Figure 3 This embodiment can automatically identify and extract all vowel phoneme parts in human voice audio based on the Automatic Speech Recognition (ASR) engine, and mark the starting timestamp. According to the characteristics of different vowel phonemes, select the more stable parts of different vowel phonemes to extract the formant values. For example, input human voice audio 1 and human voice audio 2 into the ASR engine to extract vowel phoneme time slices, and based on the extracted vowel phoneme time slices, select the formant stable points in the vowel phonemes to calculate the formant and obtain the formant values;

[0095] After extracting the formant information of all vowel phonemes in turn, the formant information of the human voice audio is obtained by storing them according to different vowel categories;

[0096] After extracting the formant information of multiple human voice audios, the formant values ​​of the same vowel phonemes of different human voice audios can be cross-compared, and the scores are obtained and sorted by cross-comparing the formant differences of the same phonemes to obtain the comparison results and scores between human voice audio 1 and human voice audio 2, that is, to obtain the human voice similarity detection results;

[0097] The calculated resonance peak difference can be used as a similarity evaluation criterion. According to the size of the resonance peak difference, the appropriate audio comparison time point can be determined, and audio screening can be further performed based on the audio comparison time point.

[0098] Optionally, in this embodiment, screening conditions such as the same character and the same word may be set according to the results of the ASR engine to further assist in the screening of the audio.

[0099] In this embodiment, by inputting audio features into a pre-trained acoustic model for phoneme matching, the vowel phonemes in the audio to be detected can be automatically located to obtain a vowel phoneme sequence. Based on the timestamp of each vowel phoneme in the vowel phoneme sequence, the resonance peak data of the audio to be detected can be effectively determined. Based on the resonance peak data, the resonance peak value of each vowel phoneme can be automatically determined. Based on the resonance peak value, the resonance peak difference between the same vowel phonemes in different audios to be detected can be effectively calculated. Based on the resonance peak difference, the human voice similarity between different audios to be detected can be effectively determined, thereby improving the accuracy of human voice similarity detection. By utilizing ASR technology, users can quickly locate the informative part of the audio and complete the screening of vowel phonemes. The comparison process is completely automatic and the results can be quickly obtained.

[0100] Embodiment 3

[0101] Figure 4 2 is a block diagram of a terminal device 2 provided in the third embodiment of the present application. Figure 4 As shown, the terminal device 2 of this embodiment includes: a processor 20, a memory 21, and a computer program 22 stored in the memory 21 and executable on the processor 20, such as a program of a human voice similarity detection method. When the processor 20 executes the computer program 22, the steps in each embodiment of the above-mentioned human voice similarity detection method are implemented.

[0102] Exemplarily, the computer program 22 may be divided into one or more modules, which are stored in the memory 21 and executed by the processor 20 to complete the present application. The one or more modules may be a series of computer program instruction segments capable of completing specific functions, which are used to describe the execution process of the computer program 22 in the terminal device 2. The terminal device may include, but is not limited to, a processor 20 and a memory 21.

[0103] The processor 20 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0104] The memory 21 may be an internal storage unit of the terminal device 2, such as a hard disk or memory of the terminal device 2. The memory 21 may also be an external storage device of the terminal device 2, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the terminal device 2. Further, the memory 21 may also include both an internal storage unit and an external storage device of the terminal device 2. The memory 21 is used to store the computer program and other programs and data required by the terminal device. The memory 21 may also be used to temporarily store data that has been output or is to be output.

[0105] In addition, each functional module in each embodiment of the present application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of software functional unit.

[0106] If the integrated module is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Among them, the computer-readable storage medium can be non-volatile or volatile. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable storage medium may include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in computer-readable storage media can be appropriately increased or decreased according to the requirements of legislation and patent practices in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practices, computer-readable storage media do not include electrical carrier signals and telecommunication signals.

[0107] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A method for detecting human voice similarity, characterized in that: The method comprises: Acquire the audio to be detected, and preprocess the audio to be detected; Extracting features from the preprocessed audio to be detected to obtain audio features, and inputting the audio features into a pre-trained acoustic model for phoneme matching to obtain a vowel phoneme sequence; Determine the formant data of the audio to be detected according to the timestamp of each vowel phoneme in the vowel phoneme sequence, and determine the formant value of each vowel phoneme according to the formant data; According to the formant values, the formant differences between the same vowel phonemes in different audios to be detected are calculated, and the human voice similarities between the different audios to be detected are determined according to the formant differences.

2. The method for detecting human voice similarity according to claim 1, characterized in that: Determining the formant data of the audio to be detected according to the timestamp of each vowel phoneme in the vowel phoneme sequence includes: Segment the audio to be detected according to the timestamp of each vowel phoneme in the vowel phoneme sequence to obtain vowel audio, and perform spectrum analysis on the vowel audio to obtain a frequency domain signal; Formant extraction is performed on the frequency domain signal to obtain formant information, and in the same audio to be detected, the formant information is classified according to the phoneme type of each vowel phoneme to obtain the formant data.

3. The method for detecting human voice similarity according to claim 2, characterized in that: Extracting the resonance peak of the frequency domain signal to obtain resonance peak information includes: Calculating a power spectrum according to the frequency domain signal, and performing a logarithmic operation on the power spectrum to obtain a logarithmic power spectrum; The logarithmic power spectrum is subjected to inverse Fourier transformation to obtain a cepstrum domain signal, and the cepstrum domain signal is subjected to low-pass filtering to obtain the resonance peak information.

4. The method for detecting human voice similarity according to claim 1, wherein: Determining the human voice similarity between different audios to be detected according to the formant difference includes: Respectively obtain the similarity weight of each vowel phoneme, and calculate the similarity score between different audios to be detected according to the similarity weight and the formant difference; The similarity score is matched with a similarity query table to obtain the voice similarity.

5. The method for detecting human voice similarity according to claim 4, characterized in that: After calculating the similarity scores between different audios to be detected according to the similarity weights and the formant differences, the method further includes: Performing differential processing on the audio features to obtain feature differential results, and performing binarization processing on the audio features according to the feature differential results to obtain feature vectors; Calculating the similarity of the feature vectors between different audios to be detected to obtain vector similarity, and performing a weighted operation on the vector similarity and the similarity score to obtain a similarity weighted value; The similarity weighted value is matched with the similarity query table to obtain the human voice similarity.

6. The method for detecting human voice similarity according to claim 1, wherein: After inputting the audio features into the pre-trained acoustic model for phoneme matching to obtain a vowel phoneme sequence, the method further includes: Matching the vowel phoneme sequence with a preset vowel group to obtain a target vowel string, and determining a target audio segment in the audio to be detected according to the vowel positions of the target vowel string; Perform audio replacement on the target audio segment corresponding to the audio to be detected.

7. The method for detecting human voice similarity according to claim 1, characterized in that: The audio features are input into the pre-trained acoustic model for phoneme matching to obtain a vowel phoneme sequence, and the method further includes: Acquire an audio sample, and input the audio sample into the acoustic model for feature extraction to obtain sample features; Performing phoneme prediction on the sample features to obtain a predicted vowel phoneme string, and calculating a model loss based on the predicted vowel phoneme string; The parameters of the acoustic model are updated according to the model loss until the acoustic model converges, thereby obtaining the pre-trained acoustic model.

8. A human voice similarity detection system, characterized in that: The system comprises: A preprocessing module, used to obtain the audio to be detected and preprocess the audio to be detected; A phoneme matching module is used to extract features from the pre-processed audio to be detected to obtain audio features, and input the audio features into a pre-trained acoustic model for phoneme matching to obtain a vowel phoneme sequence; A formant determination module, used to determine the formant data of the audio to be detected according to the timestamp of each vowel phoneme in the vowel phoneme sequence, and determine the formant value of each vowel phoneme according to the formant data; The similarity calculation module is used to calculate the formant difference between the same vowel phonemes in different audios to be detected according to the formant value, and determine the human voice similarity between the different audios to be detected according to the formant difference.

9. The human voice similarity detection system according to claim 8, characterized in that: The resonance peak determination module is also used for: Segment the audio to be detected according to the timestamp of each vowel phoneme in the vowel phoneme sequence to obtain vowel audio, and perform spectrum analysis on the vowel audio to obtain a frequency domain signal; Formant extraction is performed on the frequency domain signal to obtain formant information, and in the same audio to be detected, the formant information is classified according to the phoneme type of each vowel phoneme to obtain the formant data.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images