Voiceprint engine arbitration method and system and storage medium

Abstract

The invention provides a voiceprint engine arbitration method and system and a storage medium, and the method comprises the steps: carrying out the text conversion of a to-be-calibrated audio, carrying out the word segmentation of an audio text, obtaining a text segmented word, and obtaining a formant value of the text segmented word; performing vocabulary combination on the text segmented words in different audios to be calibrated to obtain a segmented word combination, and calculating a formant difference value in the segmented word combination according to the formant value; determining an optimal segmented word according to the formant difference value, and determining a calibration audio according to the optimal segmented word; and inputting the calibration audio into the voiceprint engine to be arbitrated for audio evaluation to obtain an audio score, and determining a target voiceprint engine in the voiceprint engine to be arbitrated according to the audio score. According to the embodiment of the invention, the target voiceprint engine in the voiceprint engine to be arbitrated can be effectively selected based on the audio score, manual arbitration selection of the voiceprint engine to be arbitrated is not needed, and the voiceprint engine arbitration accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of speech recognition technology, and in particular to a voiceprint engine arbitration method, system and storage medium. Background Art

[0002] A voiceprint engine is a technical system that can extract voiceprint features from voice signals and perform speaker recognition or verification. The voiceprint engine uses signal processing technology and machine learning algorithms to extract features, perform pattern matching, and perform identity authentication on voice signals. During the use of the voiceprint engine, multiple different voiceprint engines will be deployed at the same time to adapt to different application scenarios. In order to improve the working efficiency of the voiceprint engine, the issue of voiceprint engine arbitration selection has received increasing attention.

[0003] In the existing voiceprint engine arbitration process, the voiceprint engine is generally manually selected by human experience, which leads to low accuracy of voiceprint engine arbitration. Summary of the invention

[0004] The purpose of the embodiments of the present invention is to provide a voiceprint engine arbitration method, system and storage medium to solve the problem of low accuracy of voiceprint engine arbitration in the prior art.

[0005] The embodiment of the present invention is implemented as follows: a voiceprint engine arbitration method, the method comprising:

[0006] Acquire the audio to be calibrated, and convert the audio to be calibrated into text to obtain an audio text;

[0007] Segmenting the audio text to obtain text segmentation, and acquiring formant values ​​of the text segmentation;

[0008] Combining the text segmentations in different audios to be calibrated to obtain a segmentation combination, and calculating the formant difference between the text segmentations in the segmentation combination according to the formant values;

[0009] Determining a preferred participle in the participle combination according to the formant difference, and determining a calibration audio according to the preferred participle;

[0010] The calibrated audio is input into the voiceprint engine to be arbitrated for audio evaluation to obtain an audio score, and a target voiceprint engine in the voiceprint engine to be arbitrated is determined according to the audio score.

[0011] Preferably, obtaining the resonance peak value of the text segmentation includes:

[0012] Extracting features from the text segmentation to obtain phoneme features, and performing phoneme matching on the phoneme features to obtain a vowel phoneme sequence;

[0013] Segmenting the word segmentation audio corresponding to the text word segmentation according to the timestamp of each vowel phoneme in the vowel phoneme sequence to obtain vowel audio, and performing spectrum analysis on the vowel audio to obtain a frequency domain signal;

[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 value.

[0016] Preferably, the text segmentations in different audios to be calibrated are combined into vocabulary to obtain a segmentation combination, including:

[0017] Performing voiceprint recognition on the segmented audio corresponding to the text segmentation to obtain voiceprint features, and combining the text segmentations in different audios to be calibrated to obtain candidate combinations;

[0018] Calculating the voiceprint similarity between the text segmentations in the candidate combination according to the voiceprint features, and acquiring the text semantics of the text segmentations;

[0019] According to the text semantics, the semantic similarity between the text segment words in the candidate combination is calculated, and the segment word combination in the candidate combination is determined according to the semantic similarity and the voiceprint similarity.

[0020] Preferably, determining the word segmentation combination in the candidate combination according to the semantic similarity and the voiceprint similarity includes:

[0021] Determining a first matching score of the candidate combination according to the voiceprint similarity, and determining a second matching score of the candidate combination according to the semantic similarity;

[0022] Performing a weighted operation on the first matching score and the second matching score to obtain a total matching score;

[0023] If any of the total matching scores is greater than the score threshold, the candidate combination corresponding to the total matching score is determined as the word segmentation combination.

[0024] Preferably, determining the preferred participle in the participle combination according to the resonance peak difference includes:

[0025] If any of the resonance peak differences is less than the difference threshold, the word combination corresponding to the resonance peak difference is determined as a candidate word combination, and the candidate word combination is checked for duplicates;

[0026] Screening the candidate participles according to the duplicate detection result, and obtaining homophone information of the screened candidate participles;

[0027] If the number of homophones in any of the homophone information is greater than the number threshold, the candidate participles are deleted, and the remaining candidate participles are determined as the preferred participles.

[0028] Preferably, the calibrated audio is input into the voiceprint engine to be arbitrated for audio evaluation to obtain an audio score, including:

[0029] Calculating the signal power and noise power of the calibration audio according to the voiceprint engine to be arbitrated, and calculating the signal-to-noise ratio according to the signal power and the noise power;

[0030] The signal-to-noise ratio is matched with a score lookup table to obtain the audio score.

[0031] Preferably, determining the calibration audio according to the preferred word segmentation includes:

[0032] Acquire the segmented audio corresponding to the preferred segmented word to obtain the preferred audio, and perform audio combination on the preferred audio to obtain combined audio;

[0033] The combined audio is audio-filled according to a preset audio duration to obtain the calibrated audio.

[0034] Another object of an embodiment of the present invention is to provide a voiceprint engine arbitration system, the system comprising:

[0035] A text conversion module is used to obtain the audio to be calibrated, and perform text conversion on the audio to be calibrated to obtain an audio text;

[0036] A word segmentation module is used to segment the audio text to obtain text segmentations and obtain formant values ​​of the text segmentations;

[0037] A difference calculation module, used for performing vocabulary combination on the text segmentations in different audios to be calibrated to obtain a segmentation combination, and calculating the formant difference between the text segmentations in the segmentation combination according to the formant values;

[0038] A calibration generation module, used to determine a preferred participle in the participle combination according to the formant difference, and determine a calibration audio according to the preferred participle;

[0039] The engine arbitration module is used to input the calibration audio into the voiceprint engine to be arbitrated for audio evaluation to obtain an audio score, and determine a target voiceprint engine among the voiceprint engines to be arbitrated according to the audio score.

[0040] Preferably, the word segmentation module is also used for:

[0041] Extracting features from the text segmentation to obtain phoneme features, and performing phoneme matching on the phoneme features to obtain a vowel phoneme sequence;

[0042] Segmenting the word segmentation audio corresponding to the text word segmentation according to the timestamp of each vowel phoneme in the vowel phoneme sequence to obtain vowel audio, and performing spectrum analysis on the vowel audio to obtain a frequency domain signal;

[0043] 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;

[0044] 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 value.

[0045] The embodiment of the present invention converts the audio to be calibrated into text to obtain the audio text, and performs word segmentation on the audio text, which effectively facilitates the acquisition of the resonance peak value. By combining the text word segments in different audios to be calibrated, the calculation of the resonance peak difference between the text word segments is effectively facilitated. Based on the resonance peak difference, the preferred word segmentation in the word segmentation combination can be effectively determined. Based on the preferred word segmentation, the calibration audio can be effectively determined. By inputting the calibration audio into the voiceprint engine to be arbitrated for audio evaluation, the audio score of the calibration audio evaluated by the voiceprint engine to be arbitrated can be effectively obtained. Based on the audio score, the target voiceprint engine in the voiceprint engine to be arbitrated can be effectively selected. There is no need to manually arbitrate and select the voiceprint engine to be arbitrated in an artificial manner, thereby improving the accuracy of voiceprint engine arbitration. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a flow chart of a voiceprint engine arbitration method provided by a first embodiment of the present invention;

[0047] Figure 2 is a schematic diagram of calibration audio generation provided by the first embodiment of the present invention;

[0048] Figure 3 is a schematic diagram of arbitration of a voiceprint engine to be arbitrated provided by the first embodiment of the present invention;

[0049] Figure 4 is a structural diagram of a voiceprint engine arbitration system provided by a second embodiment of the present invention;

[0050] Figure 5 It is a schematic diagram of the structure of a terminal device provided in the third embodiment of the present invention. DETAILED DESCRIPTION

[0051] 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.

[0052] In order to illustrate the technical solution of the present invention, a specific embodiment is provided below for illustration.

[0053] Embodiment 1

[0054] See also Figure 1 , is a flow chart of a voiceprint engine arbitration method provided by a first embodiment of the present invention. The voiceprint engine arbitration method can be applied to any device or system. The voiceprint engine arbitration method includes the following steps:

[0055] Step S10, obtaining the audio to be calibrated, and converting the audio to be calibrated into text to obtain an audio text;

[0056] Wherein, the number of audios to be calibrated is greater than or equal to two, and by performing text conversion on the audios to be calibrated, the audios to be calibrated can be effectively converted into audio texts. In this step, before the audios to be calibrated are converted into texts, the audios to be calibrated are preprocessed, and the preprocessing includes noise reduction, gain adjustment, sampling rate conversion, and pre-emphasis. By performing noise reduction on the audios to be calibrated, 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 using 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 is intended to increase the energy of high-frequency components to compensate for the attenuation of high-frequency components of the speech signal during transmission. The audio quality of the audio to be calibrated is effectively improved by preprocessing the audios to be calibrated.

[0057] Step S20, segmenting the audio text to obtain text segmentation, and obtaining the formant value of the text segmentation;

[0058] Wherein, by segmenting the audio text, the acquisition of the resonance peak value is effectively facilitated. Optionally, the resonance peak value of the text segmentation is obtained, including:

[0059] Performing feature extraction on the text segmentation to obtain phoneme features, and performing phoneme matching on the phoneme features to obtain a vowel phoneme sequence; wherein, by performing feature extraction on the text segmentation, the phoneme features of the text segmentation can be effectively obtained, the phoneme features are similarity calculated with a preset phoneme vector to obtain vector similarity, the vowel phonemes corresponding to the phoneme features are determined based on the vector similarity, and the determined vowel phonemes are combined to obtain a vowel phoneme sequence;

[0060] According to the timestamp of each vowel phoneme in the vowel phoneme sequence, the word segmentation audio corresponding to the text segmentation 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 calibrated, 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;

[0061] A power spectrum is calculated according to the frequency domain signal, a logarithmic operation is performed on the power spectrum to obtain a logarithmic power spectrum, an inverse Fourier transform is performed on the logarithmic power spectrum to obtain a cepstrum domain signal, and the cepstrum domain signal is low-pass filtered to obtain the resonance peak value; wherein, by performing a logarithmic operation on the power spectrum, the power spectrum can be effectively converted into a logarithmic power spectrum, and by performing an inverse Fourier transform on the logarithmic power spectrum, it can be effectively converted into a cepstrum domain signal.

[0062] Step S30, combining the text segmentations in different audios to be calibrated to obtain a segmentation combination, and calculating the formant difference between the text segmentations in the segmentation combination according to the formant values;

[0063] Among them, by combining the text segmentations in different audios to be calibrated, the calculation of the formant difference between the text segmentations is effectively facilitated. Optionally, combining the text segmentations in different audios to be calibrated to obtain a segmentation combination includes:

[0064] Performing voiceprint recognition on the segmented audio corresponding to the text segmentation to obtain voiceprint features, and combining the text segmentations in different audios to be calibrated to obtain candidate combinations;

[0065] According to the voiceprint features, the voiceprint similarity between the text segmentations in the candidate combination is calculated, and the text semantics of the text segmentations are obtained; wherein, the voiceprint features are vector mapped to obtain voiceprint vectors, and the distance between the voiceprint vectors between the text segmentations in the candidate combination is calculated according to the Euclidean distance formula to obtain the voiceprint similarity, and the text semantics are obtained by matching the text segmentations with the semantic query table, and the corresponding relationship between different text segmentations and corresponding text semantics is stored in the semantic query table;

[0066] According to the text semantics, the semantic similarity between the text segmentations in the candidate combination is calculated, and the segmentation combination in the candidate combination is determined according to the semantic similarity and the voiceprint similarity; wherein the text semantics are combined to obtain a semantic combination, the semantic combination is matched with a similarity query table to obtain a semantic similarity, and the similarity query table stores a correspondence between different semantic combinations and corresponding semantic similarities.

[0067] Further, determining the word segmentation combination in the candidate combination according to the semantic similarity and the voiceprint similarity includes:

[0068] Determining a first matching score of the candidate combination according to the voiceprint similarity, and determining a second matching score of the candidate combination according to the semantic similarity; wherein the voiceprint similarity and the semantic similarity are matched with a score query table to obtain a first matching score and a second matching score, and the score query table stores a correspondence between different voiceprint similarities, different semantic similarities and corresponding matching scores;

[0069] Performing a weighted operation on the first matching score and the second matching score to obtain a total matching score; wherein, during the weighted operation, the weight coefficients of the first matching score and the second matching score can be set as required;

[0070] If any of the total matching scores is greater than the score threshold, the candidate combination corresponding to the total matching score is determined as the word segmentation combination; wherein the score threshold can be set according to demand, and if any of the total matching scores is greater than the score threshold, it is determined that the two word segmentations in the candidate combination corresponding to the total matching score are similar word segmentations, and therefore, the candidate combination is determined as the word segmentation combination.

[0071] Step S40, determining a preferred word segment in the word segment combination according to the formant difference, and determining a calibration audio according to the preferred word segment;

[0072] Among them, when the resonance peak difference is smaller, the two participles in the participle combination are more similar. Therefore, the preferred participle in the participle combination can be effectively determined based on the resonance peak difference. The preferred participle represents the same or similar participles in different audios to be calibrated.

[0073] Optionally, determining a preferred participle in the participle combination according to the resonance peak difference includes:

[0074] If any of the resonance peak differences is less than the difference threshold, the word combination corresponding to the resonance peak difference is determined as a candidate word, and the candidate word is checked for duplicates; wherein the difference threshold can be set according to demand, and the candidate word is checked for duplicates to detect the duplication between different candidate word segments;

[0075] The candidate segmented words are screened according to the duplicate detection results, and homonym information of the screened candidate segmented words is obtained; wherein the candidate segmented words are screened according to the duplicate detection results, which effectively improves the quality of the candidate segmented words, and the screened candidate segmented words are matched with the word information database to obtain homonym information, and the word information database stores the corresponding relationship between different segmented words and corresponding homonym information;

[0076] If the number of homophones in any of the homophone information is greater than the quantity threshold, the candidate participle is deleted, and the remaining candidate participle is determined as the preferred participle; wherein the quantity threshold can be set according to demand, and if the number of homophones in any of the homophone information is greater than the quantity threshold, it is determined that there are many homophones in the candidate participle, and therefore, the candidate participle is deleted, thereby improving the quality of subsequent calibrated audio.

[0077] Further, determining the calibration audio according to the preferred word segmentation includes:

[0078] Acquire the segmented audio corresponding to the preferred segmented word to obtain the preferred audio, and perform audio combination on the preferred audio to obtain combined audio; wherein the audio is acquired according to the timestamp of the preferred segmented word to obtain the preferred audio, and the preferred audio is audio spliced ​​to obtain the combined audio;

[0079] The combined audio is audio-filled according to the preset audio duration to obtain the calibrated audio; wherein, the preset audio duration can be set according to demand, and by audio-filling the combined audio, the audio duration of the calibrated audio is effectively guaranteed, which facilitates the subsequent evaluation of the calibrated audio by the voiceprint engine to be arbitrated.

[0080] Step S50, inputting the calibration audio into the voiceprint engine to be arbitrated for audio evaluation to obtain an audio score, and determining a target voiceprint engine among the voiceprint engines to be arbitrated according to the audio score;

[0081] Among them, by inputting the calibrated audio into the voiceprint engine to be arbitrated for audio evaluation, the audio score of the voiceprint engine to be arbitrated for evaluating the calibrated audio can be effectively obtained, and the voiceprint engine to be arbitrated corresponding to the maximum audio score is determined as the target voiceprint engine.

[0082] Optionally, the calibrated audio is input into the voiceprint engine to be arbitrated for audio evaluation to obtain an audio score, including: calculating the signal power and noise power of the calibrated audio according to the voiceprint engine to be arbitrated, and calculating the signal-to-noise ratio according to the signal power and the noise power; matching the signal-to-noise ratio with a score lookup table to obtain the audio score; wherein the score lookup table stores the correspondence between different signal-to-noise ratios and corresponding audio scores.

[0083] See also Figures 2 to 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, extract the resonance peak values, for example, input audio 1 and audio 2 into the ASR engine for word segmentation and resonance peak value extraction, obtain the resonance peak values ​​of the vowels of word 1, word 2...word n, cross-compare the resonance peak value differences between the same words or words with similar text in the two audios according to the resonance peak values, take the first 5 words with the smallest difference as the calibration audio, and send the calibration audio to each arbitration voiceprint engine for evaluation in turn, and the arbitration voiceprint engine with the highest score is determined as the target voiceprint audio.

[0084] In this embodiment, by converting the audio to be calibrated into text to obtain an audio text, and performing word segmentation on the audio text, the acquisition of the resonance peak value is effectively facilitated. By combining the text word segments in different audios to be calibrated into vocabulary, the calculation of the resonance peak difference between the text word segments is effectively facilitated. Based on the resonance peak difference, the preferred word segmentation in the word segmentation combination can be effectively determined. Based on the preferred word segmentation, the calibration audio can be effectively determined. By inputting the calibrated audio into the voiceprint engine to be arbitrated for audio evaluation, the audio score of the calibration audio evaluated by the voiceprint engine to be arbitrated can be effectively obtained. Based on the audio score, the target voiceprint engine in the voiceprint engine to be arbitrated can be effectively selected. There is no need to manually arbitrate and select the voiceprint engine to be arbitrated, thereby improving the accuracy of voiceprint engine arbitration.

[0085] Embodiment 2

[0086] See also Figure 4 , is a schematic diagram of the structure of a voiceprint engine arbitration system 100 provided in a second embodiment of the present invention, including:

[0087] The text conversion module 10 is used to obtain the audio to be calibrated, and perform text conversion on the audio to be calibrated to obtain an audio text.

[0088] The word segmentation module 11 is used to segment the audio text to obtain text segmentation and obtain the resonance peak value of the text segmentation.

[0089] Optionally, the word segmentation module 11 is further used to: extract features from the text word segmentation to obtain phoneme features, and perform phoneme matching on the phoneme features to obtain a vowel phoneme sequence;

[0090] Segmenting the word segmentation audio corresponding to the text word segmentation according to the timestamp of each vowel phoneme in the vowel phoneme sequence to obtain vowel audio, and performing spectrum analysis on the vowel audio to obtain a frequency domain signal;

[0091] 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;

[0092] 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 value.

[0093] The difference calculation module 12 is used to perform vocabulary combination on the text segmentations in different audios to be calibrated to obtain a segmentation combination, and calculate the resonance peak difference between the text segmentations in the segmentation combination according to the resonance peak value.

[0094] Optionally, the difference calculation module 12 is further used to: perform voiceprint recognition on the segmented audio corresponding to the text segmentation to obtain voiceprint features, and combine the text segmentations in different audios to be calibrated to obtain candidate combinations;

[0095] Calculating the voiceprint similarity between the text segmentations in the candidate combination according to the voiceprint features, and acquiring the text semantics of the text segmentations;

[0096] According to the text semantics, the semantic similarity between the text segment words in the candidate combination is calculated, and the segment word combination in the candidate combination is determined according to the semantic similarity and the voiceprint similarity.

[0097] Furthermore, the difference calculation module 12 is further used to: determine a first matching score of the candidate combination according to the voiceprint similarity, and determine a second matching score of the candidate combination according to the semantic similarity;

[0098] Performing a weighted operation on the first matching score and the second matching score to obtain a total matching score;

[0099] If any of the total matching scores is greater than the score threshold, the candidate combination corresponding to the total matching score is determined as the word segmentation combination.

[0100] The calibration generation module 13 is used to determine the preferred word segment in the word segment combination according to the resonance peak difference, and determine the calibration audio according to the preferred word segment.

[0101] Optionally, the calibration generation module 13 is further used for: if any of the resonance peak differences is less than a difference threshold, determining the word combination corresponding to the resonance peak difference as a candidate word, and performing a duplicate check on the candidate word;

[0102] Screening the candidate participles according to the duplicate detection result, and obtaining homophone information of the screened candidate participles;

[0103] If the number of homophones in any of the homophone information is greater than the number threshold, the candidate participles are deleted, and the remaining candidate participles are determined as the preferred participles.

[0104] Furthermore, the calibration generation module 13 is further used to: obtain the segmentation audio corresponding to the preferred segmentation to obtain the preferred audio, and perform audio combination on the preferred audio to obtain the combined audio;

[0105] The combined audio is audio-filled according to a preset audio duration to obtain the calibrated audio.

[0106] The engine arbitration module 14 is used to input the calibration audio into the voiceprint engine to be arbitrated for audio evaluation to obtain an audio score, and determine a target voiceprint engine among the voiceprint engines to be arbitrated according to the audio score.

[0107] Optionally, the engine arbitration module 14 is further used to: calculate the signal power and noise power of the calibration audio according to the voiceprint engine to be arbitrated, and calculate the signal-to-noise ratio according to the signal power and the noise power;

[0108] The signal-to-noise ratio is matched with a score lookup table to obtain the audio score.

[0109] In this embodiment, by converting the audio to be calibrated into text to obtain an audio text, and performing word segmentation on the audio text, the acquisition of the resonance peak value is effectively facilitated. By combining the text word segments in different audios to be calibrated into vocabulary, the calculation of the resonance peak difference between the text word segments is effectively facilitated. Based on the resonance peak difference, the preferred word segmentation in the word segmentation combination can be effectively determined. Based on the preferred word segmentation, the calibration audio can be effectively determined. By inputting the calibrated audio into the voiceprint engine to be arbitrated for audio evaluation, the audio score of the calibration audio evaluated by the voiceprint engine to be arbitrated can be effectively obtained. Based on the audio score, the target voiceprint engine in the voiceprint engine to be arbitrated can be effectively selected. There is no need to manually arbitrate and select the voiceprint engine to be arbitrated, thereby improving the accuracy of voiceprint engine arbitration.

[0110] Embodiment 3

[0111] Figure 5 2 is a block diagram of a terminal device 2 provided in the third embodiment of the present application. Figure 5 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 the voiceprint engine arbitration method. When the processor 20 executes the computer program 22, the steps in each embodiment of the above-mentioned voiceprint engine arbitration method are implemented.

[0112] 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.

[0113] 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.

[0114] 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.

[0115] 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.

[0116] 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.

[0117] 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 voiceprint engine arbitration method, characterized in that: The method comprises: Acquire the audio to be calibrated, and convert the audio to be calibrated into text to obtain an audio text; Segmenting the audio text to obtain text segmentation, and acquiring formant values ​​of the text segmentation; Combining the text segmentations in different audios to be calibrated to obtain a segmentation combination, and calculating the formant difference between the text segmentations in the segmentation combination according to the formant values; Determining a preferred participle in the participle combination according to the formant difference, and determining a calibration audio according to the preferred participle; The calibrated audio is input into the voiceprint engine to be arbitrated for audio evaluation to obtain an audio score, and a target voiceprint engine among the voiceprint engines to be arbitrated is determined according to the audio score.

2. The voiceprint engine arbitration method according to claim 1, characterized in that: Obtaining the resonance peak value of the text segmentation includes: Extracting features from the text segmentation to obtain phoneme features, and performing phoneme matching on the phoneme features to obtain a vowel phoneme sequence; Segmenting the word segmentation audio corresponding to the text word segmentation according to the timestamp of each vowel phoneme in the vowel phoneme sequence to obtain vowel audio, and performing spectrum analysis on the vowel audio to obtain a frequency domain signal; 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 value.

3. The voiceprint engine arbitration method according to claim 1, characterized in that: Combining the text segmentations in different audios to be calibrated to obtain a segmentation combination includes: Performing voiceprint recognition on the segmented audio corresponding to the text segmentation to obtain voiceprint features, and combining the text segmentations in different audios to be calibrated to obtain candidate combinations; Calculating the voiceprint similarity between the text segmentations in the candidate combination according to the voiceprint features, and acquiring the text semantics of the text segmentations; According to the text semantics, the semantic similarity between the text segment words in the candidate combination is calculated, and the segment word combination in the candidate combination is determined according to the semantic similarity and the voiceprint similarity.

4. The voiceprint engine arbitration method according to claim 3, characterized in that: Determining the word segmentation combination in the candidate combination according to the semantic similarity and the voiceprint similarity includes: Determining a first matching score of the candidate combination according to the voiceprint similarity, and determining a second matching score of the candidate combination according to the semantic similarity; Performing a weighted operation on the first matching score and the second matching score to obtain a total matching score; If any of the total matching scores is greater than the score threshold, the candidate combination corresponding to the total matching score is determined as the word segmentation combination.

5. The voiceprint engine arbitration method according to claim 1, characterized in that: Determining a preferred participle in the participle combination according to the resonance peak difference includes: If any of the resonance peak differences is less than the difference threshold, the word combination corresponding to the resonance peak difference is determined as a candidate word combination, and the candidate word combination is checked for duplicates; Screening the candidate participles according to the duplicate detection result, and obtaining homophone information of the screened candidate participles; If the number of homophones in any of the homophone information is greater than the number threshold, the candidate participles are deleted, and the remaining candidate participles are determined as the preferred participles.

6. The voiceprint engine arbitration method according to claim 1, characterized in that: The calibrated audio is input into the arbitration voiceprint engine for audio evaluation to obtain an audio score, including: Calculating the signal power and noise power of the calibration audio according to the voiceprint engine to be arbitrated, and calculating the signal-to-noise ratio according to the signal power and the noise power; The signal-to-noise ratio is matched with a score lookup table to obtain the audio score.

7. The voiceprint engine arbitration method according to claim 1, characterized in that: Determining the calibration audio according to the preferred word segmentation includes: Acquire the segmented audio corresponding to the preferred segmented word to obtain the preferred audio, and perform audio combination on the preferred audio to obtain combined audio; The combined audio is audio-filled according to a preset audio duration to obtain the calibrated audio.

8. A voiceprint engine arbitration system, characterized in that: The system comprises: A text conversion module is used to obtain the audio to be calibrated, and perform text conversion on the audio to be calibrated to obtain an audio text; A word segmentation module is used to segment the audio text to obtain text segmentations and obtain formant values ​​of the text segmentations; A difference calculation module, used for performing vocabulary combination on the text segmentations in different audios to be calibrated to obtain a segmentation combination, and calculating the formant difference between the text segmentations in the segmentation combination according to the formant values; A calibration generation module, used to determine a preferred participle in the participle combination according to the formant difference, and determine a calibration audio according to the preferred participle; The engine arbitration module is used to input the calibration audio into the voiceprint engine to be arbitrated for audio evaluation to obtain an audio score, and determine a target voiceprint engine among the voiceprint engines to be arbitrated according to the audio score.

9. The voiceprint engine arbitration system according to claim 8, characterized in that: The word segmentation module is also used for: Extracting features from the text segmentation to obtain phoneme features, and performing phoneme matching on the phoneme features to obtain a vowel phoneme sequence; Segmenting the word segmentation audio corresponding to the text word segmentation according to the timestamp of each vowel phoneme in the vowel phoneme sequence to obtain vowel audio, and performing spectrum analysis on the vowel audio to obtain a frequency domain signal; 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 value.

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