Method, device, equipment and storage medium for identifying same person based on voice
By acquiring and processing speech data, using the combination of vector machine model and deep convolutional neural network, the tone feature vectors are extracted for comparison, which solves the problem of low speech recognition accuracy in the prior art and improves the accuracy of fan recognition. Suitable for scenarios such as intelligent diagnosis and treatment and remote consultation.
Patent Information
- Application Number
- CN202110836229.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-07-23
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2041-07-23
AI Technical Summary
The accuracy of voice-based fan recognition in the prior art is low, and it is impossible to effectively characterize the personality characteristics of the speaker, resulting in inaccurate recognition results.
By obtaining the voice to be recognized by the target user, extracting the flag parameter information, format conversion and feature parameter extraction, using the preset vector machine model and deep convolutional neural network, the output tone feature vectors are compared to judge the target user and the registered user Is it a fan?
It improves the accuracy of speech recognition and the accuracy of fan recognition, and is suitable for scenarios such as intelligent diagnosis and treatment and remote consultation.
Smart Images

Figure CN113555022B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of artificial intelligence, and in particular to a method, device, equipment and storage medium for identifying same person based on voice. Background Art
[0002] With the continuous development of artificial intelligence, voice has been widely used in many fields, such as human-computer interaction. For example, voice can be used to control devices or conduct intelligent voice conversations through robots. Voice can also be used to support disease auxiliary diagnosis, health management, remote consultation, etc. Therefore, a large number of human-computer interaction products need to distinguish the speaker itself, that is, to identify and distinguish the speaker's identity through voice.
[0003] In the prior art, when the identity of a speaker is identified and distinguished based on speech recognition, only a limited length of speech features in the target user's speech data is extracted for recognition. The result cannot accurately represent the personality characteristics of a speaker, and the recognition result is based on probability calculation, which makes it difficult to achieve a very high resolution, resulting in low accuracy of same-person recognition. Summary of the invention
[0004] The main purpose of the present invention is to solve the technical problem of low accuracy of same-person recognition based on voice in the prior art.
[0005] The first aspect of the present invention provides a method for identifying same-person characters based on speech, and the method comprises: obtaining the speech to be identified of a target user, and extracting the marker parameter information of the speech to be identified; performing parameter analysis on the marker parameter information to determine the format type and attribute information of the speech to be identified; performing format conversion on the speech to be identified according to the format type and attribute information, and extracting the feature parameters of the speech to be identified after the format conversion; performing age recognition on the speech to be identified based on a preset vector machine model and the feature parameters to determine the age group of the target user, and extracting speech data corresponding to the age group from a preset speech database of registered users; respectively inputting the speech data and the speech to be identified into a preset deep convolutional neural network, outputting the corresponding timbre feature vector, and comparing the speech data with the timbre feature vector of the speech to be identified to determine whether the target user and the registered user are same-person characters.
[0006] Optionally, in a first implementation method of the first aspect of the present invention, the format conversion of the speech to be recognized according to the format type and attribute information, and extracting characteristic parameters of the speech to be recognized after the format conversion include: extracting the sampling rate, bit rate and channel from the attribute information of the speech to be recognized according to the format type; judging whether the sampling rate and bit rate meet preset requirements; if they do not meet the preset requirements, converting the sampling rate and bit rate based on preset conversion rules, and judging whether the channel of the speech to be recognized is mono; if it is not mono, converting the channel to mono according to preset channel conversion rules; extracting the characteristic parameters of the speech to be recognized after the format conversion, wherein the characteristic parameters include time domain characteristic parameters and frequency domain characteristic parameters.
[0007] Optionally, in a second implementation method of the first aspect of the present invention, the age recognition of the speech to be recognized based on the preset vector machine model and the feature parameters, the age group of the target user is determined, and the speech data corresponding to the age group is extracted from the preset voice database of registered users, including: performing dimensionality reduction and aggregation processing on the feature parameters to obtain age feature parameters; based on the preset vector machine model and the age feature parameters, performing age recognition on the speech to be recognized to obtain a recognition result; comparing the recognition result with the recognition rate in the vector machine model, and calculating the confidence of the recognition result; determining the age group of the target user according to the confidence; and extracting speech data corresponding to the age group from the preset voice database of registered users.
[0008] Optionally, in a third implementation method of the first aspect of the present invention, the voice data and the voice to be recognized are respectively input into a preset deep convolutional neural network, a timbre feature vector is output, and the timbre feature vector is compared to determine whether the target user and the registered user are the same person, including: based on a preset deep convolutional neural network, respectively extracting the voiceprint features of the voice data and the voice to be recognized; clustering the voiceprint features to obtain a timbre feature vector; calculating a similarity value of the timbre feature vector, and determining whether the similarity value is not less than a preset timbre similarity threshold; if so, determining that the target user and the registered user are the same person.
[0009] Optionally, in a fourth implementation method of the first aspect of the present invention, clustering the voiceprint features to obtain a timbre feature vector includes: calculating the chromaticity eigenvalues of the voiceprint features, and generating a voiceprint matrix based on the chromaticity eigenvalues; inputting the voiceprint features into the deep convolutional neural network, and outputting a timbre feature representation; mapping the timbre feature representation to a preset feature space, and quantitatively characterizing the timbre feature representation based on the feature space to obtain a timbre feature vector.
[0010] Optionally, in the fifth implementation method of the first aspect of the present invention, before respectively extracting the voiceprint features of the voice data and the voice to be recognized in the preset deep convolutional neural network, it also includes: respectively performing frame processing on the voice data and the voice data to be recognized to obtain audio frames; extracting the short-time energy of the audio frame, and judging whether the short-time energy is less than a preset energy threshold, wherein the short-time energy is the strength of the audio frame at different moments; if so, discarding the corresponding audio frame.
[0011] Optionally, in a sixth implementation manner of the first aspect of the present invention, after respectively inputting the voice data and the voice to be recognized into a preset deep convolutional neural network, outputting a timbre feature vector, and comparing the timbre feature vector to determine whether the target user and the registered user are the same person, it also includes: extracting the frame voiceprint features of the voice to be recognized; calculating the posterior probability of the frame voiceprint features based on a preset time-delay neural network; calculating the hot unique value of the posterior probability; classifying the frame voiceprint features according to the hot unique value, and identifying the frame voiceprint features according to the classification result; registering the target user's voice according to the identification, and storing the voice to be recognized in the voice database of the registered user.
[0012] The second aspect of the present invention proposes a voice-based same-person identification device, which includes: an acquisition module, which is used to acquire the voice to be identified of the target user and extract the marker parameter information of the voice to be identified; an analysis module, which is used to perform parameter analysis on the marker parameter information to determine the format type and attribute information of the voice to be identified; a conversion module, which is used to perform format conversion on the voice to be identified according to the format type and attribute information, and extract the feature parameters of the voice to be identified after the format conversion; a recognition module, which is used to perform age recognition on the voice to be identified based on a preset vector machine model and the feature parameters, determine the age group of the target user, and extract voice data corresponding to the age group from a preset voice database of registered users; a comparison module, which is used to input the voice data and the voice to be identified into a preset deep convolutional neural network respectively, output the corresponding timbre feature vector, and compare the voice data with the timbre feature vector of the voice to be identified to determine whether the target user and the registered user are same-person.
[0013] Optionally, in a first implementation method of the second aspect of the present invention, the conversion module includes: a first extraction unit, used to extract the sampling rate, bit rate and channel from the attribute information of the speech to be recognized according to the format type; a judgment unit, used to judge whether the sampling rate and bit rate meet preset requirements; a first conversion unit, used to convert the sampling rate and bit rate based on a preset conversion rule if the sampling rate and bit rate do not meet the preset requirements, and judge whether the channel of the speech to be recognized is mono; a second conversion unit, used to convert the channel to mono according to a preset channel conversion rule if the channel is not mono; a second extraction unit, used to extract feature parameters of the speech to be recognized after the format conversion, wherein the feature parameters include time domain feature parameters and frequency domain feature parameters.
[0014] Optionally, in a second implementation method of the second aspect of the present invention, the recognition module is specifically used to: perform dimensionality reduction and aggregation processing on the feature parameters to obtain age feature parameters; perform age recognition on the speech to be recognized based on a preset vector machine model and the age feature parameters to obtain a recognition result; compare the recognition result with the recognition rate in the vector machine model, and calculate the confidence of the recognition result; determine the age group of the target user based on the confidence; and extract voice data corresponding to the age group from a preset registered user's voice database.
[0015] Optionally, in a third implementation method of the second aspect of the present invention, the comparison module includes: a third extraction unit, used to extract the voiceprint features of the voice data and the voice to be recognized, respectively, based on a preset deep convolutional neural network; a clustering unit, used to cluster the voiceprint features to obtain a timbre feature vector; a calculation unit, used to calculate the similarity value of the timbre feature vector, and determine whether the similarity value is not less than a preset timbre similarity threshold; a determination unit, used to determine that the target user and the registered user are the same person if the similarity value is not less than a preset timbre similarity threshold.
[0016] Optionally, in a fourth implementation method of the second aspect of the present invention, the clustering unit is specifically used to: calculate the chromaticity eigenvalue of the voiceprint feature, and generate a voiceprint matrix based on the chromaticity eigenvalue; input the voiceprint feature into the deep convolutional neural network, and output a timbre feature representation; map the timbre feature representation to a preset feature space, and quantitatively characterize the timbre feature representation based on the feature space to obtain a timbre feature vector.
[0017] Optionally, in a fifth implementation of the second aspect of the present invention, the speech-based same-person recognition device also includes a elimination module, which is specifically used to: respectively perform frame processing on the speech data and the speech data to be recognized to obtain audio frames; extract the short-time energy of the audio frames, and determine whether the short-time energy is less than a preset energy threshold; if so, eliminate the corresponding audio frames.
[0018] Optionally, in a sixth implementation manner of the second aspect of the present invention, the voice-based same-person recognition device also includes a registration module, which is specifically used to: extract frame voiceprint features of the voice to be recognized; calculate the posterior probability of the frame voiceprint features based on a preset time-delay neural network; calculate the hot unique value of the posterior probability; classify the frame voiceprint features according to the hot unique value, and identify the frame voiceprint features according to the classification result; perform voice registration on the target user according to the identification, and store the voice to be recognized in the voice database of the registered user.
[0019] The third aspect of the present invention provides a voice-based fan recognition device, which includes: a memory and at least one processor, wherein instructions are stored in the memory, and the memory and the at least one processor are interconnected via a line; the at least one processor calls the instructions in the memory so that the voice-based fan recognition device executes the steps of the above-mentioned voice-based fan recognition method.
[0020] A fourth aspect of the present invention provides a computer-readable storage medium having instructions stored thereon, which, when executed on a computer, enables the computer to execute the steps of the above-mentioned method for same-person identification based on speech.
[0021] In the technical solution provided by the present invention, by obtaining the landmark parameter information in the speech to be recognized of the target user, performing parameter analysis on it, extracting the characteristic parameters of the speech to be recognized, determining the age group of the target user based on the preset vector machine model and characteristic parameters, and extracting the speech data corresponding to the age group from the preset voice database of registered users, and selectively selecting the speech data of registered users who need to perform same-person recognition, thereby improving the recognition rate of speech, inputting the speech data and the speech to be recognized into a preset deep convolutional neural network respectively, outputting the timbre feature vector, and judging whether the target user and the registered user are the same person. The present invention can be applied to intelligent diagnosis and treatment and remote consultation, and by extracting the timbre feature vector from the speech for vector comparison, the accuracy of same-person recognition is improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] Figure 1 Schematic diagram of a first embodiment of a method for identifying same person based on speech in an embodiment of the present invention;
[0023] Figure 2 Schematic diagram of a second embodiment of the method for identifying same person based on speech in an embodiment of the present invention;
[0024] Figure 3 Schematic diagram of a third embodiment of the method for identifying same person based on speech in an embodiment of the present invention;
[0025] Figure 4 Schematic diagram of a fourth embodiment of a method for identifying same person based on speech in an embodiment of the present invention;
[0026] Figure 5 Schematic diagram of an embodiment of a voice-based same-person identification device in an embodiment of the present invention;
[0027] Figure 6 Schematic diagram of another embodiment of a device for identifying same person based on speech in an embodiment of the present invention;
[0028] Figure 7 The figure is a schematic diagram of an embodiment of a voice-based fan recognition device in an embodiment of the present invention. DETAILED DESCRIPTION
[0029] The embodiment of the present invention provides a method, device, equipment and storage medium for same-person identification based on speech. By acquiring the landmark parameter information in the speech to be identified of the target user, performing parameter analysis on it, extracting the characteristic parameters of the speech to be identified, determining the age group of the target user based on a preset vector machine model and characteristic parameters, and extracting the speech data corresponding to the age group from a preset voice database of registered users, directionally selecting the speech data of the registered user who needs to be identified as the same-person, thereby improving the recognition rate of speech. The speech data and the speech to be identified are respectively input into a preset deep convolutional neural network, and a timbre feature vector is output to determine whether the target user and the registered user are the same-person. The embodiment of the present invention can be applied to intelligent diagnosis and treatment and remote consultation. By extracting the timbre feature vector from the speech for vector comparison, the accuracy of same-person identification is improved.
[0030] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention 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 described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0031] For ease of understanding, the specific contents of the embodiments of the present invention are described below. Figure 1 The first embodiment of the method for identifying same person based on speech in the embodiment of the present invention includes:
[0032] 101, obtaining the speech to be recognized of the target user, and extracting the marker parameter information of the speech to be recognized;
[0033] In an embodiment of the present invention, the voice-based same-person recognition method can be applied to intelligent diagnosis and treatment and remote consultation. The medical platform collects the target user's voice to be recognized in real time during the digital consultation process based on medical big data, and uploads the collected voice to be recognized to the server for storage. When storing the voice to be recognized, the voice to be recognized is preprocessed, and the marker parameter information is extracted from the preprocessed voice to be recognized. Among them, the preprocessing includes framing, windowing and pre-emphasis. Framing is to truncate the speech signal according to short-term stability. The frame length is generally 20ms, and the frame shift is generally 10ms. Windowing generally uses a Hamming window or a Hanning window, because the main lobe width corresponds to the frequency resolution. The wider the main lobe width, the lower the corresponding frequency resolution. Therefore, when selecting the window function, the energy should be concentrated on the main lobe as much as possible, or the relative amplitude of the maximum side lobe height should be as small as possible. The Hamming window has a large side lobe attenuation in the amplitude-frequency characteristic and can reduce the Gibbs effect. Therefore, the Hamming window is generally selected for the windowing of speech signals. Since the speech signal is easily affected by glottal excitation and oral and nasal radiation, the frequency components above 800Hz will be attenuated by 6dB / octave. Therefore, it is necessary to use the pre-emphasis method to increase the energy of the high-frequency part and take the opportunity to compensate for the high-frequency loss. Generally, a first-order high-pass filter is used to achieve pre-emphasis. In addition, pre-processing can also include anti-aliasing filtering.
[0034] In addition, the embodiments of the present invention can acquire and process relevant voice data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. Basic artificial intelligence technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. Artificial intelligence software technology mainly includes computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.
[0035] The server in the embodiment of the present invention can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.
[0036] 102, performing parameter analysis on the flag parameter information to determine the format type and attribute information of the speech to be recognized;
[0037] Perform parameter analysis on each parameter in the flag parameter information. Since the flag parameter information of different format types is not consistent, the format type and attribute information of the voice to be recognized are determined according to the type of each parameter contained in the flag parameter information. Among them, the format types of voice include WAV, MP3, and OGG. These voice formats have industry-unified standards. If the recording device uses a certain format, the player must use this format for decoding, otherwise the voice cannot be played normally. Specifically, the flag parameter information is input into the voice format recognition unit of the server, and the file header, standard parameters, etc. of the WAV format are checked. If it meets the WAV standard, the voice is judged to be in WAV format. Otherwise, the flag parameter information is subjected to the decoder / parameter check of the MP3 format standard. If it meets the MP3 standard, the voice is judged to be in MP3 format. Otherwise, the flag parameter information is subjected to the decoder / parameter check of the OGG format standard. If it meets the OGG standard, the voice is judged to be in OGG format, and the format is returned to the server. If the format of the voice still cannot be recognized, an information prompt is issued that the voice in this format is not supported.
[0038] In the present embodiment, WAV is a sound file format developed by Microsoft, which complies with the RIFF (Resource Interchange File Format) file specification and is used to save audio information resources on the Windows platform. The WAV file header is a piece of data at the beginning of the file that undertakes a certain task, usually a description of the main data. The WAV file header consists of four parts: RIFF chunk (RIFF-Chunk), format chunk (Format-Chunk), additional chunk (Fact-Chunk), and data chunk (Data-Chunk). The file header contains 44 bytes of flag parameter information, including: 4 bytes of RIFF flag, 4 bytes of file length, 4 bytes of "WAVE" type chunk identifier, etc. By judging these flag parameter information, it can be determined whether the voice file is WAV.
[0039] MP3 stands for MPEG Audio Layer 3. It is an efficient computer audio encoding scheme that converts audio files into smaller files with a .mp3 extension at a higher compression ratio, basically maintaining the sound quality of the source file. MP3 files are generally divided into three parts: ID3V2, audio data, and ID3V1. The audio data records parameter information such as sampling rate and bit rate. By judging these parameter information, you can determine whether the voice file is MP3.
[0040] OGG stands for OGG Vorbis, an audio compression format similar to music formats such as MP3. After decoding, an OGG file forms a bitstream. The first three headers of the bitstream are: identification header, comment header, and setup header in the order in the file. The identification header sets the version and simple audio characteristics of the stream (such as sampling rate and number of channels). By judging these audio characteristics, it can be determined whether the speech to be recognized is in OGG format.
[0041] 103, performing format conversion on the speech to be recognized according to the format type and attribute information, and extracting feature parameters of the speech to be recognized after the format conversion;
[0042] According to the format type, the sampling rate, bit rate and channel in the attribute information of the speech to be recognized are extracted, and the sampling rate and bit rate are compared to see whether they are the same as the preset requirements, that is, whether the sampling rate and bit rate meet the preset requirements. If they do not meet the preset requirements, format conversion is required. If they meet the preset requirements, format conversion is not required. The preset requirements are that the sampling rate is 8k and the bit rate is 16bit. The process of format conversion of the speech to be recognized is to convert the sampling rate and bit rate of the speech to be recognized into the sampling rate and bit rate that meet the preset requirements according to the preset conversion rules.
[0043] Determine whether the channel of the speech to be recognized is mono. If the speech to be recognized is not mono, that is, the speech to be recognized is dual-channel, it is converted to mono according to the preset channel conversion rule. Specifically, the server provides a jni interface for channel separation based on the C++ program. The interface defines parameters (left channel or right channel). The Java program calls the interface and passes the left channel to the interface. After the C++ program is processed, the dual-channel speech is converted to mono and transmitted to the Java program.
[0044] The feature parameters contained in the converted speech to be recognized are extracted, wherein the feature parameters include time domain feature parameters and frequency domain feature parameters. The time domain feature parameters include short-time zero-crossing rate, short-time energy spectrum and fundamental frequency period, and the frequency domain feature parameters include linear prediction cepstral frequency (LPCC) and Mel frequency cepstral coefficient (MFCC).
[0045] 104, based on the preset vector machine model and feature parameters, perform age recognition on the speech to be recognized, determine the age group of the user, and extract speech data corresponding to the age group from a preset voice database of registered users;
[0046] The feature parameters are input into the preset vector machine model, and the vector machine model performs parameter analysis on the feature parameters, extracts the age feature parameters therefrom, and performs age recognition on the speech to be recognized based on the age feature parameters to obtain a recognition result, that is, to determine the age group corresponding to the speech to be recognized. After the age group corresponding to the speech to be recognized is determined, the speech data of the same age group is extracted from the preset voice database of registered users. The preset voice database of registered users contains the voice data of all registered users registered through the medical platform in the server.
[0047] After determining the age group of the target user corresponding to the voice to be recognized, extract the voice data corresponding to the age group from the preset voice database of registered users, that is, extract the voice data of registered users of the same age group from the voice database. Perform frame processing on the voice data and the voice data to be recognized respectively to obtain audio frames; extract the short-time energy of the audio frame, and determine whether the short-time energy is less than the preset energy threshold, where the short-time energy is the strength of the audio frame at different times; if the short-time energy is less than the preset energy threshold, remove the corresponding audio frame, that is, filter the audio frame of the voice data and the voice to be recognized, thereby improving the accuracy of the subsequent comparison of the timbre feature vector.
[0048] 105, input the voice data and the voice to be recognized into a preset deep convolutional neural network respectively, output the corresponding timbre feature vector, and compare the voice data with the timbre feature vector of the voice to be recognized to determine whether the target user and the registered user are the same person.
[0049] The speech data and the speech to be recognized are respectively input into the deep convolutional neural network preset by the server, and the speech is analyzed by the deep convolutional neural network, and a three-dimensional timbre feature representation is output, and the timbre feature representation is mapped to the preset space, and the timbre feature is quantitatively characterized by the feature space to obtain a timbre feature vector. The timbre feature vector of the registered user and the timbre feature vector corresponding to the speech to be recognized are vector-compared in the feature space. When the two timbre feature vectors are consistent, it indicates that the target user corresponding to the speech to be recognized is the same person as the registered user. When the two timbre feature vectors are inconsistent, it indicates that the target user corresponding to the speech to be recognized is not the same person as the registered user, wherein the preset deep convolutional neural network is trained in advance. The training process of the deep convolutional neural network is a prior art and will not be described in detail here.
[0050] In this embodiment, the voice data of two target users may be obtained at random, and the voice data of the two targets may be format converted, age recognized, and timbre feature vectors compared to determine whether the two target users are the same person. That is, in this embodiment, the target user to be identified and the registered user may be identified as the same person, or the two target users to be identified may be identified as the same person.
[0051] In the embodiment of the present invention, the voice to be recognized is subjected to format conversion and age recognition processing, and then, based on the processing result, the voice data of registered users with the same age group as the voice to be recognized is selectively selected for same-person recognition, thereby improving the recognition rate of the voice. Then, the timbre feature vector is extracted from the voice for vector comparison, thereby improving the accuracy of same-person recognition.
[0052] See also Figure 2 The second embodiment of the method for identifying same person based on speech in the embodiment of the present invention includes:
[0053] 201, obtaining the speech to be recognized of the target user, and extracting the marker parameter information of the speech to be recognized;
[0054] 202, performing parameter analysis on the flag parameter information to determine the format type and attribute information of the speech to be recognized;
[0055] 203, performing format conversion on the speech to be recognized according to the format type and attribute information, and extracting feature parameters of the speech to be recognized after the format conversion;
[0056] 204, performing dimension reduction and aggregation processing on the feature parameters to obtain age feature parameters;
[0057] The characteristic parameters are subjected to dimensionality reduction and aggregation processing. In this embodiment, the dimensionality reduction processing is to use the principal component analysis algorithm (PCA algorithm) to perform data dimensionality reduction of the characteristic parameters, and the aggregation is to use the K-means clustering algorithm (k-means algorithm) to aggregate the data of the characteristic parameters. After the dimensionality reduction and aggregation processing of the characteristic parameters are performed, the age characteristic parameters are obtained. Among them, the dimensionality reduction and aggregation processing of the data belong to the prior art and will not be described in detail here.
[0058] 205, based on the preset vector machine model and age feature parameters, perform age recognition on the speech to be recognized to obtain a recognition result;
[0059] A vector machine model is set up on the server. According to the support vector machine (SVM) method of the vector machine model, the age feature parameters are mapped to a high-dimensional or even infinite-dimensional feature space (Hilbert space) through a nonlinear mapping p. The expansion theorem of the kernel function is applied. Under the premise of almost no increase in computational complexity, the age group to which the speech to be recognized belongs is identified based on the vector machine model and the age feature parameters to obtain the recognition result.
[0060] 206, comparing the recognition result with the recognition rate in the vector machine model, and calculating the confidence of the recognition result;
[0061] 207, determining the age group of the target users based on the confidence level;
[0062] The recognition rate of each age group is set in the vector machine model. The recognition result is compared with the recognition rate, and the confidence of the recognition result is calculated, that is, the confidence of the recognition result is analyzed according to the recognition rate. When the value of the confidence is not less than the preset confidence threshold, it means that the recognition result is relatively accurate, and the age group of the target user corresponding to the voice to be recognized can be determined.
[0063] 208, extracting voice data corresponding to the age group from a preset voice database of registered users;
[0064] 209, input the voice data and the voice to be recognized into a preset deep convolutional neural network respectively, output the corresponding timbre feature vector, and compare the voice data and the timbre feature vector of the voice to be recognized to determine whether the target user and the registered user are the same person.
[0065] After determining the age group of the target user corresponding to the voice to be recognized, extract the voice data corresponding to the age group from the preset voice database of registered users, that is, extract the voice data of registered users of the same age group from the voice database. The voice data and the voice to be recognized are respectively input into the deep convolutional neural network preset by the server, and the voice is analyzed by the deep convolutional neural network, and a three-dimensional timbre feature representation is output, and the timbre feature representation is mapped to the preset space, and the timbre feature is quantitatively characterized by the feature space to obtain the timbre feature vector. The timbre feature vector of the registered user and the timbre feature vector corresponding to the voice to be recognized are vector-compared in the feature space. When the two timbre feature vectors are consistent, it indicates that the target user corresponding to the voice to be recognized is the same person as the registered user. When the two timbre feature vectors are inconsistent, it indicates that the target user corresponding to the voice to be recognized is not the same person as the registered user, wherein the preset deep convolutional neural network is trained in advance. The training process of the deep convolutional neural network is a prior art and will not be described in detail here.
[0066] In the embodiment of the present invention, steps 201-203 are consistent with steps 101-103 in the first embodiment of the above-mentioned method for identifying a person based on speech, and are not described in detail here.
[0067] In an embodiment of the present invention, the format of the speech to be recognized is converted, feature parameters are extracted, and the feature parameters are subjected to dimensionality reduction and aggregation processing to obtain age feature parameters. Age recognition is performed based on the age feature parameters, and then the confidence of the recognition result is calculated to determine the age group of the speech to be recognized, thereby selectively selecting the voice data of registered users with the same age group as the speech to be recognized for same-person recognition, thereby improving the recognition rate of the speech.
[0068] See also Figure 3The third embodiment of the method for identifying same person based on speech in the embodiment of the present invention includes:
[0069] 301, obtaining the speech to be recognized of the target user, and extracting the marker parameter information of the speech to be recognized;
[0070] 302, performing parameter analysis on the flag parameter information to determine the format type and attribute information of the speech to be recognized;
[0071] 303, performing format conversion on the speech to be recognized according to the format type and attribute information, and extracting feature parameters of the speech to be recognized after the format conversion;
[0072] 304, based on the preset vector machine model and feature parameters, perform age recognition on the speech to be recognized, determine the age group of the user, and extract speech data corresponding to the age group from a preset voice database of registered users;
[0073] 305, based on a preset deep convolutional neural network, respectively extracting voice data and voiceprint features of the voice to be recognized;
[0074] In this embodiment, the voice print feature is a feature representing the voice characteristics of the user, ie, the speaker. The voice print feature can be extracted by a preset neural network model, wherein the neural network model is a pre-trained sequential deep convolutional neural network.
[0075] Specifically, according to the voice endpoint detection method (VAD detection) in the deep convolutional neural network, endpoint detection is performed on the voice data and the voice to be recognized, so that the voice data and the voice to be recognized are respectively divided into multiple audio data of the speaker, such as audio data corresponding to each time period of 0-3 seconds, 4-7 seconds and 7-10 seconds, and features (embedding) are extracted according to the audio data, that is, the voiceprint features corresponding to each audio data segment are extracted. The voiceprint feature can be regarded as a vector, and the dimension can be set as needed, such as 128 dimensions or 512 dimensions. The unique characteristics of the speaker can be characterized by the voiceprint feature. Among them, the audio data of different lengths can be extracted to obtain a vector of fixed dimension. For example, the matrix corresponding to each audio data can be input into the deep convolutional neural network. The voiceprint can be a frequency, and the matrix is composed in time order, that is, the matrix is a two-dimensional array of time-frequency, and the deep convolutional neural network outputs the corresponding vectors of fixed dimension.
[0076] 306, clustering the voiceprint features to obtain a timbre feature vector;
[0077] In this embodiment, clustering processing of voiceprint features may be performed by using K-means clustering algorithm (K-means) clustering or spectral clustering, where K represents the number of categories, and K may be determined according to the number of speakers in the target voice data.
[0078] After the server extracts the corresponding voiceprint features, a matrix can be formed by the voiceprint features. Each row of the matrix can represent the voiceprint features corresponding to a segment of audio data in the speech. The voiceprint features are fixed-dimensional vectors, and the duration of the audio data corresponding to each row can be different. For example, the first row of the matrix can represent a vector of 0-3 seconds, the second row can represent a vector of 4-7 seconds, the third row can represent a vector of 7-10 seconds, and so on. The matrix of voiceprint features is clustered to obtain the clustering results of the voiceprint features corresponding to each segment of audio data. After clustering, the voiceprint features are vectorized to obtain a timbre feature vector.
[0079] 307, calculating the similarity value of the timbre feature vector, and determining whether the similarity value is not less than a preset timbre similarity threshold;
[0080] 308 , if the similarity value is not less than a preset timbre similarity threshold, it is determined that the target user and the registered user are the same person.
[0081] The timbre feature vector of the target user corresponding to the voice to be recognized and the timbre feature vector of the registered user are similarly calculated, that is, the similarity value of the two timbre feature vectors is calculated, and whether the target user corresponding to the voice to be recognized and the registered user are the same person is determined according to the similarity value. When the similarity value of the two timbre feature vectors is not less than a preset timbre similarity threshold, it is determined that the target user and the registered user are the same person, and when the similarity value is less than the timbre similarity threshold, the target user and the registered user are not the same person.
[0082] In the embodiment of the present invention, steps 301-304 are consistent with steps 101-104 in the first embodiment of the above-mentioned method for identifying same person based on speech, and are not described in detail here.
[0083] In an embodiment of the present invention, voiceprint features are extracted from the speech of the target user and the registered user, and the voiceprint features are clustered to obtain timbre feature vectors, and the similarity values of the timbre feature vectors are calculated. Based on the comparison of their similarity values with a similarity threshold, it is determined whether the target user and the registered user are the same person, thereby improving the accuracy of same-person identification.
[0084] See also Figure 4 The fourth embodiment of the method for identifying same person based on speech in the embodiment of the present invention includes:
[0085] 401, obtaining the speech to be recognized of the target user, and extracting the marker parameter information of the speech to be recognized;
[0086] 402, performing parameter analysis on the flag parameter information to determine the format type and attribute information of the speech to be recognized;
[0087] 403, performing format conversion on the speech to be recognized according to the format type and attribute information, and extracting feature parameters of the speech to be recognized after the format conversion;
[0088] 404, based on the preset vector machine model and feature parameters, perform age recognition on the speech to be recognized, determine the age group of the user, and extract speech data corresponding to the age group from a preset voice database of registered users;
[0089] 405, inputting the voice data and the voice to be recognized into a preset deep convolutional neural network respectively, outputting a timbre feature vector, and comparing the timbre feature vector to determine whether the user and the registered user are the same person;
[0090] 406, extracting frame voiceprint features of the speech to be recognized;
[0091] When the target user and the registered user are not the same person, the target user can be voice registered to become a registered user, or when the target user and the registered user are the same person, the voice can also be registered and added to the voice database of the registered user to enrich the voice data of the registered user. In this embodiment, the voiceprint feature can be a Mel Frequency Cepstrum Coefficient (MFCC). Specifically, the characteristic parameter MFCC of the speech to be recognized is directly extracted. Since the MFCC is automatically framed during the extraction process, and the MFCC corresponding to each frame is processed, the frame voiceprint feature corresponding to each frame in the speech to be recognized is obtained. Alternatively, the speech to be recognized can be sliced in frames, and the MFCC features of each slice can be extracted separately to obtain the MFCC features corresponding to each frame, that is, the frame voiceprint feature.
[0092] 407, calculating the posterior probability of the frame voiceprint feature based on a preset time-delay neural network;
[0093] In this embodiment, a universal background model (UBM) implemented by a time delay neural network (TDNN), namely a TDNN-UBM model, is used to calculate the posterior probability. Specifically, the voiceprint features of each frame are used as data input, and the posterior probability corresponding to each frame of the speech to be recognized is obtained based on the TDNN-UBM model (time delay neural network, TDNN).
[0094] 408, calculating the hot unique value of the posterior probability;
[0095] 409, classifying the frame voiceprint features according to the hot unique value, and marking the frame voiceprint features according to the classification result;
[0096] Classify each frame voiceprint feature according to the obtained posterior probability. Calculate the one-hot value of each posterior probability, and then classify the frame voiceprint features corresponding to the same one-hot value into the same category, and record the one-hot value as the type identifier of the corresponding category, that is, identify the frame voiceprint features according to the one-hot value.
[0097] In this embodiment, the collected speech to be recognized is classified, and the corresponding type identification is recorded for search and matching in the subsequent voiceprint recognition process. By using the TDNN-UBM model, the posterior probability corresponding to each frame segment is obtained, and the frame segments are classified based on the posterior probability, so as to complete the convergence of the speech to be recognized, thereby extracting the key features of the speech to be recognized; then the frame segments of the same type are classified into the same category to obtain clearer recognition features, which can provide more comprehensive recognition verification for the subsequent recognition process to improve recognition accuracy. In addition, when classifying frame voiceprint features based on the posterior probability, the classification standard can be generated by calculating the hot unique value of the posterior probability, thereby improving the classification accuracy.
[0098] 410 , register the target user's voice according to the identifier, and store the voice to be recognized in a voice database of registered users.
[0099] According to the type identification, the user corresponding to the voice to be recognized is registered, and the 1:1 registration interface provided by the medical platform set up on the server is called, wherein the registration interface defines parameters (the identification number of the registered user, the path of the voice), and the registration is realized by calling this registration interface. When the registration is completed, the voice to be recognized for voice registration is stored in the voice database of the registered user. In addition, the voice during voice registration is not the original voice to be recognized, but the voice in 8k, 16bit, wav format obtained after the above steps (same person recognition processing, classification and identification).
[0100] In the embodiment of the present invention, steps 401-405 are consistent with steps 101-105 in the first embodiment of the above-mentioned voice-based same person identification method, and are not described in detail here.
[0101] In an embodiment of the present invention, after the same person recognition is completed, the frame voiceprint features of the target user's voice are extracted, the frame voiceprint features are classified and identified, so as to register the voice of the target user and store it in the language database to expand the registered user and voice data.
[0102] The above describes the method for identifying a same person based on voice in the embodiment of the present invention. The following describes the device for identifying a same person based on voice in the embodiment of the present invention. Figure 5 An embodiment of the voice-based peer recognition device in the embodiment of the present invention includes:
[0103] The acquisition module 501 is used to acquire the speech to be recognized of the target user and extract the marker parameter information of the speech to be recognized;
[0104] An analysis module 502 is used to perform parameter analysis on the flag parameter information to determine the format type and attribute information of the speech to be recognized;
[0105] The conversion module 503 is used to perform format conversion on the speech to be recognized according to the format type and attribute information, and extract feature parameters of the speech to be recognized after format conversion;
[0106] The recognition module 504 is used to perform age recognition on the speech to be recognized based on a preset vector machine model and the characteristic parameters, determine the age group of the target user, and extract speech data corresponding to the age group from a preset voice database of registered users;
[0107] The comparison module 505 is used to input the voice data and the voice to be recognized into a preset deep convolutional neural network respectively, output the corresponding timbre feature vector, and compare the voice data with the timbre feature vector of the voice to be recognized to determine whether the target user and the registered user are the same person.
[0108] In the embodiment of the present invention, the voice to be recognized is subjected to format conversion and age recognition processing by a voice-based same person recognition device, and then, based on the processing result, the voice data of registered users with the same age group as the voice to be recognized is selectively selected for same person recognition, thereby improving the recognition rate of the voice, and then the timbre feature vector is extracted from the voice for vector comparison, thereby improving the accuracy of same person recognition.
[0109] See also Figure 6 Another embodiment of the voice-based peer recognition device in the embodiment of the present invention includes:
[0110] The acquisition module 501 is used to acquire the speech to be recognized of the target user and extract the marker parameter information of the speech to be recognized;
[0111] An analysis module 502 is used to perform parameter analysis on the flag parameter information to determine the format type and attribute information of the speech to be recognized;
[0112] The conversion module 503 is used to perform format conversion on the speech to be recognized according to the format type and attribute information, and extract feature parameters of the speech to be recognized after format conversion;
[0113] The recognition module 504 is used to perform age recognition on the speech to be recognized based on a preset vector machine model and the characteristic parameters, determine the age group of the target user, and extract speech data corresponding to the age group from a preset voice database of registered users;
[0114] The comparison module 505 is used to input the voice data and the voice to be recognized into a preset deep convolutional neural network respectively, output the corresponding timbre feature vector, and compare the voice data and the timbre feature vector of the voice to be recognized to determine whether the target user and the registered user are the same person.
[0115] Wherein, the conversion module 503 includes:
[0116] The first extraction unit 5031 is used to extract the sampling rate, bit rate and channel from the attribute information of the speech to be recognized according to the format type;
[0117] The judging unit 5032 is used to judge whether the sampling rate and the bit rate meet the preset requirements;
[0118] The first conversion unit 5033 is used to convert the sampling rate and the bit rate based on a preset conversion rule if the sampling rate and the bit rate do not meet the preset requirements, and determine whether the channel is mono;
[0119] A second conversion unit 5034 is configured to convert the channel of the to-be-recognized speech into a mono channel according to a preset channel conversion rule if the channel of the to-be-recognized speech is not a mono channel;
[0120] The second extraction unit 5035 is used to extract feature parameters of the speech to be recognized after the format conversion, wherein the feature parameters include time domain feature parameters and frequency domain feature parameters.
[0121] The identification module 504 is specifically used for:
[0122] Performing dimension reduction and aggregation processing on the characteristic parameters to obtain age characteristic parameters;
[0123] Based on a preset vector machine model and the age feature parameters, age recognition is performed on the speech to be recognized to obtain a recognition result;
[0124] Comparing the recognition result with the recognition rate in the vector machine model, and calculating the confidence of the recognition result;
[0125] Determining the age group of the target user according to the confidence level;
[0126] The voice data corresponding to the age group is extracted from a preset voice database of registered users.
[0127] Wherein, the comparison module 505 includes:
[0128] The third extraction unit 5051 is used to extract the voiceprint features of the voice data and the voice to be recognized respectively based on a preset deep convolutional neural network;
[0129] A clustering unit 5052, configured to perform clustering processing on the voiceprint features to obtain a timbre feature vector;
[0130] The calculation unit 5053 is used to calculate the similarity value of the timbre feature vector and determine whether the similarity value is not less than a preset timbre similarity threshold;
[0131] The determination unit 5054 is configured to determine that the target user and the registered user are the same person if the similarity value is not less than a preset timbre similarity threshold.
[0132] The clustering unit 5052 is specifically used for:
[0133] Calculating the chromaticity eigenvalue of the voiceprint feature, and generating a voiceprint matrix according to the chromaticity eigenvalue;
[0134] Inputting the voiceprint feature into the deep convolutional neural network and outputting a timbre feature representation;
[0135] The timbre feature representation is mapped to a preset feature space, and the timbre feature representation is quantitatively characterized according to the feature space to obtain a timbre feature vector.
[0136] The speech-based homoerotic identification device further includes a rejection module 506, which is specifically used for:
[0137] Performing frame processing on the voice data and the voice data to be recognized respectively to obtain audio frames;
[0138] Extracting the short-time energy of the audio frame, and determining whether the short-time energy is less than a preset energy threshold;
[0139] If yes, the corresponding audio frame is discarded.
[0140] The voice-based same person identification device further includes a registration module 507, which is specifically used for:
[0141] Extracting the frame voiceprint features of the speech to be recognized;
[0142] Calculating the posterior probability of the frame voiceprint feature based on a preset time-delay neural network;
[0143] Calculate the hot unique value of the posterior probability;
[0144] Classifying the frame voiceprint features according to the hot unique values, and marking the frame voiceprint features according to the classification results;
[0145] According to the identifier, the target user's voice is registered, and the voice to be recognized is stored in the voice database of the registered user.
[0146] In the embodiment of the present invention, a voice-based fan identification device performs format conversion and age recognition on the voice, and then extracts the voice of registered users in the same age group as the target user for fan comparison, thereby improving the recognition rate of the voice and the accuracy of fan identification.
[0147] See also Figure 7 , the following describes in detail an embodiment of a voice-based fan recognition device in an embodiment of the present invention from the perspective of hardware processing.
[0148] Figure 7 7 is a schematic diagram of the structure of a voice-based fan recognition device provided by an embodiment of the present invention. The voice-based fan recognition device 700 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 710 (for example, one or more processors) and a memory 720, and one or more storage media 730 (for example, one or more mass storage devices) storing application programs 733 or data 732. Among them, the memory 720 and the storage medium 730 may be temporary storage or permanent storage. The program stored in the storage medium 730 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the voice-based fan recognition device 700. Furthermore, the processor 710 may be configured to communicate with the storage medium 730 to execute a series of instruction operations in the storage medium 730 on the voice-based fan recognition device 700.
[0149] The voice-based peer recognition device 700 may also include one or more power supplies 740, one or more wired or wireless network interfaces 750, one or more input and output interfaces 760, and or, one or more operating systems 731, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will appreciate that Figure 7The illustrated structure of the voice-based fan recognition device does not constitute a limitation of the voice-based fan recognition device, and may include more or less components than those illustrated, or may combine certain components, or may have different arrangements of components.
[0150] The blockchain referred to in this invention is a new application mode of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.
[0151] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the speech-based same-person identification method.
[0152] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0153] If the integrated unit 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. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.
[0154] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those 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 thereof may be replaced by equivalents. However, these 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 invention.
Claims
1. A method for identifying homosexuals based on speech. It is characterized in that The method for identifying same person based on voice includes: Acquire the target user's speech to be recognized, and extract the marker parameter information of the speech to be recognized; Performing parameter analysis on the flag parameter information to determine the format type and attribute information of the speech to be recognized; Extracting sampling rate, bit rate and channel from attribute information of the speech to be recognized according to the format type; Determine whether the sampling rate and bit rate meet preset requirements; If the preset requirements are not met, the sampling rate and the bit rate are converted based on a preset conversion rule, and it is determined whether the channel of the speech to be recognized is monophonic; If it is not a mono channel, converting the channel into a mono channel according to a preset channel conversion rule; Extracting feature parameters of the speech to be recognized after format conversion, wherein the feature parameters include time domain feature parameters and frequency domain feature parameters; Based on a preset vector machine model and the characteristic parameters, perform age recognition on the speech to be recognized, determine the age group of the target user, and extract speech data corresponding to the age group from a preset voice database of registered users; The voice data and the voice to be recognized are respectively input into a preset deep convolutional neural network, the corresponding timbre feature vector is output, and the voice data and the timbre feature vector of the voice to be recognized are compared to determine whether the target user and the registered user are the same person.
2. The method for identifying same person based on speech according to claim 1, It is characterized in that The step of performing age recognition on the speech to be recognized based on the preset vector machine model and the characteristic parameters, determining the age group of the target user, and extracting speech data corresponding to the age group from a preset voice database of registered users includes: Performing dimension reduction and aggregation processing on the characteristic parameters to obtain age characteristic parameters; Based on a preset vector machine model and the age feature parameters, age recognition is performed on the speech to be recognized to obtain a recognition result; Comparing the recognition result with the recognition rate in the vector machine model, and calculating the confidence of the recognition result; Determining the age group of the target user according to the confidence level; The voice data corresponding to the age group is extracted from a preset voice database of registered users.
3. The method for identifying same person based on voice according to claim 2, It is characterized in that The step of inputting the voice data and the voice to be recognized into a preset deep convolutional neural network respectively, outputting a timbre feature vector, and comparing the timbre feature vector to determine whether the target user and the registered user are the same person comprises: Based on a preset deep convolutional neural network, respectively extracting the voice data and the voiceprint features of the voice to be recognized; Performing clustering processing on the voiceprint features to obtain a timbre feature vector; Calculating a similarity value of the timbre feature vector, and determining whether the similarity value is not less than a preset timbre similarity threshold; If so, it is determined that the target user and the registered user are the same person.
4. The method for identifying same person based on speech according to claim 3, It is characterized in that The clustering process of the voiceprint features to obtain the timbre feature vector comprises: Calculating the chromaticity eigenvalue of the voiceprint feature, and generating a voiceprint matrix according to the chromaticity eigenvalue; Inputting the voiceprint feature into the deep convolutional neural network and outputting a timbre feature representation; The timbre feature representation is mapped to a preset feature space, and the timbre feature representation is quantitatively characterized according to the feature space to obtain a timbre feature vector.
5. The method for identifying same person based on speech according to claim 4, It is characterized in that Before respectively extracting the voice data and the voiceprint features of the to-be-recognized voice in the preset deep convolutional neural network, the method further includes: Performing frame processing on the voice data and the voice data to be recognized respectively to obtain audio frames; Extracting the short-time energy of the audio frame, and determining whether the short-time energy is less than a preset energy threshold, wherein the short-time energy is the strength of the audio frame at different moments; If yes, the corresponding audio frame is discarded.
6. The method for identifying same person based on speech according to any one of claims 1 to 5, It is characterized in that After respectively inputting the voice data and the voice to be recognized into a preset deep convolutional neural network, outputting a timbre feature vector, and comparing the timbre feature vector to determine whether the target user and the registered user are the same person, the method further includes: Extracting the frame voiceprint features of the speech to be recognized; Based on a preset time-delay neural network, calculating the posterior probability of the frame voiceprint feature; Calculate the hot unique value of the posterior probability; Classifying the frame voiceprint features according to the hot unique values, and marking the frame voiceprint features according to the classification results; According to the identifier, the target user's voice is registered, and the voice to be recognized is stored in the voice database of the registered user.
7. A device for identifying homosexuals based on voice, It is characterized in that The voice-based homoerotic identification device comprises: An acquisition module, used to acquire the target user's speech to be recognized and extract the marker parameter information of the speech to be recognized; An analysis module, used for performing parameter analysis on the flag parameter information to determine the format type and attribute information of the speech to be recognized; A conversion module, used for extracting the sampling rate, bit rate and channel from the attribute information of the speech to be recognized according to the format type; judging whether the sampling rate and bit rate meet the preset requirements; if not, converting the sampling rate and bit rate based on the preset conversion rule, and judging whether the channel of the speech to be recognized is mono; if not, converting the channel to mono according to the preset channel conversion rule; extracting the feature parameters of the speech to be recognized after the format conversion, wherein the feature parameters include time domain feature parameters and frequency domain feature parameters; A recognition module, configured to perform age recognition on the speech to be recognized based on a preset vector machine model and the characteristic parameters, determine the age group of the target user, and extract speech data corresponding to the age group from a preset voice database of registered users; The comparison module is used to input the voice data and the voice to be recognized into a preset deep convolutional neural network respectively, output the corresponding timbre feature vector, and compare the voice data with the timbre feature vector of the voice to be recognized to determine whether the target user and the registered user are the same person.
8. A device for identifying homosexuals based on voice. It is characterized in that The voice-based peer recognition device comprises: A memory and at least one processor, wherein instructions are stored in the memory, and the memory and the at least one processor are interconnected via a line; The at least one processor calls the instructions in the memory to enable the fan identification device to perform the steps of the fan identification method based on speech according to any one of claims 1 to 6.
9. A computer-readable storage medium having instructions stored thereon, It is characterized in that When the instructions are executed by a processor, the steps of the method for same-person identification based on speech as described in any one of claims 1 to 6 are implemented.
