Calling method and device, electronic equipment and storage medium
By extracting the Mel cepstrum coefficients of audio data and combining the Hidden Markov model for speech recognition, the problem of inaccurate identification of existing emergency call devices in complex environments is solved, the recognition accuracy is improved, and the care cost is reduced.
Patent Information
- Application Number
- CN202411727579.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-28
- Publication Date
- 2025-03-07
AI Technical Summary
The existing emergency call devices cannot adapt to the complex surrounding environment and cannot accurately identify various sounds made by the protected person, resulting in low accuracy of voice recognition and inability to ensure the personal safety of the protected person, which in turn increases the cost of personnel care.
By extracting the Mel cepspectral coefficients corresponding to the audio data, we judge whether they match the preset spectrum characteristics, determine the target word sequence based on the Mel cepspectral coefficients and the hidden Markov model, perform speech recognition, and finally make a call based on the recognition results.
It improves the accuracy of voice recognition, can more accurately identify the sounds made by the protected person, reduces the cost of personnel care, and enhances the personal safety guarantee of the protected person.
Smart Images
Figure CN119580728A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of speech recognition, and in particular, relates to a calling method, a device, an electronic device and a storage medium. Background Art
[0002] In the related art, common emergency call devices mainly have functions such as one-button calling and telephone alarm, but this type of emergency call function cannot adapt to all protected persons. In real life, there are many protected persons who are unable to move, so a voice call function is needed in an emergency.
[0003] However, existing calling devices can only simply recognize fixed sentences, cannot adapt to complex surrounding environments, and cannot accurately recognize various sounds made by the protected person. The voice recognition accuracy is low and the personal safety of the protected person cannot be guaranteed, which in turn leads to increased personnel care costs. Summary of the invention
[0004] The disclosed embodiments provide a solution to solve the problem in the related art that some calling devices can only simply recognize fixed sentences, cannot adapt to complex surrounding environments, cannot accurately recognize various sounds made by protected persons, have low voice recognition accuracy, cannot guarantee the personal safety of protected persons, and thus lead to increased personnel care costs.
[0005] In a first aspect, the present disclosure provides a calling method, the method comprising:
[0006] Extracting Mel-cepstral coefficients corresponding to the first audio data;
[0007] Determining whether the Mel-frequency cepstral coefficient matches a preset frequency spectrum feature;
[0008] If yes, determining a target word sequence corresponding to the Mel-cepstral coefficients based on the Mel-cepstral coefficients and the hidden Markov model;
[0009] Recognize the target word sequence to obtain a speech recognition result;
[0010] A call is made according to the voice recognition result.
[0011] In a second aspect, the present disclosure provides a calling device, the device comprising:
[0012] An extraction unit, used to extract Mel-cepstral coefficients corresponding to the first audio data;
[0013] A judging unit, used for judging whether the Mel-frequency cepstral coefficient matches a preset frequency spectrum feature;
[0014] a determining unit, configured to determine, if yes, a target word sequence corresponding to the Mel-cepstral coefficients based on the Mel-cepstral coefficients and a hidden Markov model;
[0015] A recognition unit, used to recognize the target word sequence and obtain a speech recognition result;
[0016] A calling unit is used to make a call according to the speech recognition result.
[0017] In a third aspect, the present disclosure provides an electronic device, including:
[0018] Processor; and
[0019] A memory, configured to store executable instructions of the processor;
[0020] The processor is configured to execute any method in the first aspect or any possible implementation of the first aspect by executing the executable instructions.
[0021] In a fourth aspect, an embodiment of the present disclosure provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the computer program implements any method in the first aspect or any possible implementation of the first aspect.
[0022] The technical solution provided by the present disclosure extracts the Mel-cepstral coefficients corresponding to the first audio data; determines whether the Mel-cepstral coefficients match the preset spectral features; if so, determines the target word sequence corresponding to the Mel-cepstral coefficients based on the Mel-cepstral coefficients and the hidden Markov model, identifies the target word sequence, and obtains the speech recognition result; and makes a call based on the speech recognition result. The technical solution provided by each embodiment of the present disclosure can determine whether the extracted Mel-cepstral coefficients corresponding to the first audio data match the preset spectral features. If they match, the target word sequence corresponding to the Mel-cepstral coefficients can be determined. Finally, a call can be made based on the target word sequence. There is no need to make a call directly when the audio data of the protected person is collected. Instead, the audio data will be recognized and processed to obtain the sentence actually called by the protected person, thereby realizing the call, improving the accuracy of speech recognition, and greatly reducing the cost of personnel care. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:
[0024] Figure 1A flowchart of a calling method provided by an embodiment of the present disclosure;
[0025] Figure 2 A schematic diagram of the structure of a calling device provided by an embodiment of the present disclosure;
[0026] Figure 3 A schematic diagram of the structure of an electronic device provided by an embodiment of the present disclosure. DETAILED DESCRIPTION
[0027] Embodiments of the present disclosure are described in detail below, and examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present disclosure, but should not be understood as limiting the present disclosure.
[0028] The terms "first" and "second" and the like in the specification, claims, and drawings of the disclosed embodiments 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 interchangeable where appropriate, so that the embodiments of the disclosed embodiments described herein can, for example, be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or devices.
[0029] The calling method provided in the embodiments of the present disclosure may be run on a terminal device or a server. The terminal device may be a local terminal device. The server may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, 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, CDN, and big data and artificial intelligence platforms.
[0030] In the related technology, common emergency call devices mainly have functions such as one-button calling and telephone alarm, but such emergency call functions cannot adapt to all protected persons. In real life, there are many protected persons who are unable to move, so voice calling functions are needed in emergency situations. However, existing call devices can only simply recognize fixed sentences, cannot adapt to complex surrounding environments, and cannot accurately recognize various sounds made by protected persons. The voice recognition accuracy is low, and the personal safety of protected persons cannot be guaranteed, which in turn leads to increased personnel care costs.
[0031] The following specific embodiments are used to describe in detail the technical solution of the present invention and how the technical solution of the present invention solves the above-mentioned technical problems. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present invention will be described below in conjunction with the accompanying drawings.
[0032] Figure 1 A flowchart of a calling method provided by an exemplary embodiment of the present disclosure is provided. The method can be applied to various devices with data processing functions. Taking the method applied to a user terminal as an example, the solution at least includes the following steps S101-S105:
[0033] S101, extracting Mel-cepstral coefficients corresponding to first audio data.
[0034] In some embodiments, the Mel-Frequency Cepstral Coefficient is also called Mel-Frequency Cepstral Coefficient, MFCC (Mel-Frequency Cepstral Coefficients) sequence, which is a speech feature parameter commonly used in speech recognition and speaker recognition.
[0035] In some embodiments, the method for obtaining the first audio data further includes steps S11-S12:
[0036] S11, obtaining original audio data.
[0037] In some embodiments, the original audio data refers to the voice data of the protected person collected by the user terminal. Specifically, the user terminal can collect surrounding voice data in real time.
[0038] S12: pre-process the original audio data to obtain the first audio data.
[0039] The preprocessing includes audio signal amplification, analog-to-digital conversion and environmental noise filtering.
[0040] In some embodiments, extracting Mel-cepstral coefficients of the first audio data includes steps S21-S24:
[0041] S21: Perform bandpass filtering on the first audio data to obtain second audio data.
[0042] The bandpass filter is mainly used to process the signal of a specific frequency band in the first audio data. Specifically, since the human voice frequency is mostly between 85 and 1100 Hz, the bandpass filter is used on the collected first audio data to increase the gain of the human voice frequency band.
[0043] S22: Perform short-time Fourier transform on the second audio data to obtain an energy value of a Mel spectrum corresponding to the second audio data.
[0044] Among them, short-time Fourier transform (STFT, short-time Fourier transform, or short-term Fourier transform) is a mathematical transformation related to Fourier transform, which is used to determine the frequency and phase of the sine wave in the local area of the time-varying signal.
[0045] The energy value of the Mel spectrum is calculated by converting the spectrum of the audio signal into the Mel scale.
[0046] In some embodiments, performing a short-time Fourier transform on the second audio data to obtain an energy value of a Mel spectrum corresponding to the second audio data includes steps S221-S223:
[0047] S221: Pre-emphasize the second audio data to obtain third audio data.
[0048] Among them, pre-emphasis is a signal processing technology, which is mainly used to compensate for the loss of high-frequency components of the signal during transmission.
[0049] By pre-emphasizing the second audio data, the attenuation of the second audio data during the transmission process can be compensated, thereby improving the signal-to-noise ratio and transmission quality of the signal, which is beneficial to subsequent processing.
[0050] S222, divide the third audio data into frames to obtain a plurality of third sub-audio data.
[0051] In some embodiments, dividing the third audio data into frames to obtain a plurality of third sub-audio data includes: dividing the third audio data into frames according to a preset length to obtain a plurality of third sub-audio data.
[0052] The preset length is preferably 20-30 milliseconds, and may also be other lengths, which can be set according to actual conditions.
[0053] Specifically, taking the preset length of 20 milliseconds as an example, the third audio data is divided into a plurality of short time frames, that is, a plurality of third sub-audio data, wherein the length of each frame is 20 milliseconds.
[0054] S223: Window each of the third sub audio data, and perform Fourier transform on each of the windowed third sub audio data to obtain an energy value of a Mel spectrum corresponding to the second audio data.
[0055] The windowing refers to applying a Hamming window or a Hanning window to each frame signal to reduce the discontinuity between frames.
[0056] In some embodiments, Fourier transform is performed on each third sub-audio data after windowing to obtain an energy value of the Mel spectrum corresponding to the second audio data, including: Fourier transform is performed on each third sub-audio data after windowing, and each third sub-audio data is converted from the time domain to the frequency domain to obtain a Mel spectrum; a set of triangular filters based on Mel frequency are constructed, and the Mel spectrum is filtered using the triangular filters to obtain the energy value of the Mel spectrum corresponding to the second audio data.
[0057] S23, taking the logarithm of the energy value to obtain the FBank feature vector.
[0058] Among them, FBank (FilterBank) is a front-end processing algorithm used for speech signal processing. FBank feature vectors are mainly used in speech recognition, speech analysis, and speech synthesis.
[0059] S24, performing discrete cosine transform on the FBank feature vector to obtain the Mel-cepstral coefficient.
[0060] Among them, the FBank feature vector is different from the MFCC feature vector. The MFCC feature vector is the Hz spectrum feature calculated using the nonlinear correspondence between the Mel frequency and the Hz frequency; the FBank feature vector is a feature obtained by frequency domain analysis and is often used in speech recognition and audio analysis.
[0061] In some embodiments, the Mel-frequency cepstral coefficient corresponding to the extracted first audio data, that is, the MFCC sequence is shown as follows:
[0062] O=o1,o2,…,o n
[0063] Wherein, O is the MFCC sequence corresponding to the first audio data, and each o in the above formula i , i=1,2…n represents a set of MFCC feature vectors.
[0064] S102, determining whether the Mel-frequency cepstral coefficient matches a preset frequency spectrum feature.
[0065] In some embodiments, the preset frequency spectrum feature is a frequency spectrum feature corresponding to the collected voice data of the protected person.
[0066] In some embodiments, the process of determining whether the Mel-frequency cepstral coefficient matches the preset frequency spectrum feature is as follows: Steps S31-S32:
[0067] S31, obtaining a plurality of first probability values based on the Mel-cepstral coefficients, the Gaussian mixture model and the general background model.
[0068] Since the user terminal needs to implement an alarm function for the emergency situation of a specific protected person, in order to prevent false alarms, it is necessary to match the Mel-frequency cepstral coefficients corresponding to the first audio data with the preset spectrum features. Therefore, the present invention uses the EM algorithm (expected maximum algorithm) to establish a GMM-UBM model based on MFCC to determine multiple first probability values, as shown in the following formula:
[0069]
[0070] in, The GMM model parameters after combining with the UBM parameters need to be iteratively updated until convergence before combining, which is used to adjust the impact of the GMM model parameters on the final prediction model; and p(O,μ n ,σ n 2 ) is the probability density function of GMM distribution; is the probability that the input MFCC feature vector o belongs to the jth GMM model, that is, the probability that the Mel cepstral coefficient matches the preset spectrum feature.
[0071] S32: If at least one of the multiple first probability values is greater than a preset probability value, the Mel-frequency cepstral coefficient matches a preset feature vector.
[0072] In some embodiments, the method further includes: if a maximum first probability value among the plurality of first probability values is greater than a preset probability value, matching the Mel-cepstral coefficient with a preset feature vector.
[0073] S103: If yes, determine the target word sequence corresponding to the Mel-cepstral coefficients based on the Mel-cepstral coefficients and the hidden Markov model.
[0074] In some embodiments, the target word sequence refers to a keyword sequence corresponding to the first audio data.
[0075] In some implementations, based on the Mel-cepstral coefficients and the hidden Markov model, determining the target word sequence corresponding to the Mel-cepstral coefficients includes steps S41-S42:
[0076] S41, based on the hidden Markov model and the Mel-cepstral coefficients, determining a state sequence with the maximum probability for a given observation value sequence.
[0077] In some embodiments, the hidden Markov model is a probability model about time series, which describes the process of randomly generating a random sequence of unobservable states by a hidden Markov chain, and then generating an observation from each state to produce a random sequence of observations.
[0078] State sequence: a randomly generated state sequence of a hidden Markov chain;
[0079] Observation sequence: Each state generates one observation, and the resulting random sequence of observations.
[0080] In some embodiments, after step S32, it is necessary to match the Mel-frequency cepstral coefficients to the corresponding target word sequence. Specifically, the MFCC sequence uses the Viterbi algorithm to search for the HMM state sequence with the maximum probability of generating a given observation value sequence on the HMM (Hidden Markov Model) model.
[0081] S42: Based on the state sequence, obtain a target word sequence corresponding to the Mel-cepstral coefficients.
[0082] In some embodiments, the step of obtaining a target word sequence corresponding to the Mel-frequency cepstral coefficients based on the state sequence includes steps S421-S423:
[0083] S421, determining a first word sequence corresponding to the state sequence.
[0084] In some embodiments, the first word sequence is obtained from a word library stored locally.
[0085] Specifically, the first word sequence is shown as follows:
[0086] W=w1,w1,…,w m
[0087] Where W is the first word sequence that matches the MFCC sequence, and its length is m.
[0088] S422: Based on the first audio data and the first word sequence, determine a posterior probability that any word sequence in the first word sequence matches the first audio data under the first audio data.
[0089] The target word sequence is as follows:
[0090]
[0091] In the above formula, O is the MFCC sequence, W is any word sequence, that is, the first word sequence, P(W|O) is the posterior probability that any word sequence in the first word sequence matches the first audio data under the first audio data, is the most likely word sequence that matches the MFCC sequence, i.e., the target word sequence.
[0092] Among them, the calculation of P(W|O) is the decoding process of the decoder, which can be calculated according to the Bayesian formula, and the formula is as follows:
[0093]
[0094] In the above formula, the likelihood probability P(O|W) is an acoustic model based on GMM-HMM (Gaussian mixture model-hidden Markov model); the prior probability P(W) is an n-gram model (n-gram language model), which is used to predict the joint probability distribution between a series of word characters.
[0095] In some embodiments, the n language model finds the target word sequence by calculating the probabilities of various recognition results, specifically the joint probability distribution of the target word sequence, as shown in the following formula:
[0096]
[0097] The word sequence w in the above formula i =w1,w1,…,w m Any w in i Only with the adjacent last observation w i-1 It can be regarded as a first-order Markov chain. Therefore, the above formula can be further simplified as:
[0098]
[0099] In some embodiments, the n-gram language model can be viewed as an n-order Markov chain, so the above formula can be used to calculate the probability of the i-th word appearing after the first word sequence is input in sequence, thereby finding the most likely word sequence matching result.
[0100] In other embodiments, a word sequence may have multiple pronunciations, so the likelihood probability P(O|W) can also be calculated using the total probability formula under multiple pronunciation sequences, as shown in the following formula:
[0101]
[0102] Where Q = q1, q2, …, q m represents a candidate pronunciation sequence, whose length is the same as the first word sequence W, and corresponds to the MFCC sequence O and the first word sequence W respectively.
[0103] P(O|W) in the above formula, that is, the pronunciation sequence likelihood probability corresponding to the first word sequence, can be calculated by the following formula:
[0104]
[0105] Based on the above formula, P(O|Q) can be further decomposed into:
[0106]
[0107] Wherein, A is the state transition parameter corresponding to the MFCC sequence O.
[0108] In some embodiments, the matching between sequences of all possible likelihood probabilities P(O|Q) is very large and complex and cannot be calculated by enumeration. Therefore, the Viterbi algorithm is preferably used to find the best path corresponding to the likelihood function calculation, that is, to find the optimal word sequence, that is, the target word sequence, under given audio input data.
[0109] Among them, the Viterbi algorithm is a dynamic programming method that uses the idea of space-for-time. This algorithm can find the optimal state transfer parameters when a pronunciation sequence Q and an MFCC sequence O are given. As shown below:
[0110]
[0111] In the above formula, is the likelihood probability under the optimal hidden path.
[0112] If we assume here that we enter state a at time t t,i The probability of P(o1,o2,…,o t ,a t,i |Q), for the Viterbi algorithm, at time t, it enters state a t,i The shortest path is:
[0113]
[0114] In the formula, O t-1 are all subpaths from the initial state to time t-1.
[0115] According to the recursive calculation, if δ is known j (t-1) is the shortest path at time t-1. The shortest path at time t can be calculated from the shortest path at the previous time, which can be expressed as:
[0116]
[0117] From the above formula, we can find that the Viterbi algorithm only needs to consider the current and previous states at each moment in the calculation process. At the same time, in the actual calculation process, the continuous product of probabilities will cause floating point underflow, so it is necessary to use a logarithmic function to convert it into an addition form, as shown in the following formula:
[0118]
[0119] According to the above formula, the likelihood probability The calculation can be achieved with lower computational complexity, which can greatly shorten the calculation process.
[0120] S423: Determine a target word sequence corresponding to the Mel-cepstral coefficients based on the posterior probability and the first word sequence.
[0121] In some embodiments, based on the posterior probability of each word sequence in the first word sequence, the word sequence whose posterior probability is greater than the preset probability is used as the target word sequence. In this way, the accuracy of the recognized word sequence can be greatly improved.
[0122] S104, identifying the target word sequence to obtain a speech recognition result.
[0123] In some embodiments, after the target word sequence is obtained according to step S104, the keywords spoken by the protected person are determined, and the speech recognition results can be obtained according to the keywords.
[0124] Specifically, if the target word sequence obtained is: stomachache, pain, uncomfortable, the speech recognition result obtained is: the protected person has a stomachache.
[0125] S105: Making a call according to the voice recognition result.
[0126] Specifically, if the target word sequence obtained is: stomachache, pain, uncomfortable, the speech recognition result obtained is: the protected person has a stomachache, so the user terminal can make a call.
[0127] Among them, the calling method can be automatic telephone alarm, emergency call, etc., which can be set according to actual conditions.
[0128] In actual operation, a 4G or 5G module is set in the user terminal, and the call information can be uploaded to the cloud through the base station near the user terminal, and then the call information is sent to other users' smart terminals through the cloud. The relevant information of the protected person can be transmitted in time.
[0129] The technical solution provided by the present disclosure extracts the Mel-cepstral coefficients corresponding to the first audio data; determines whether the Mel-cepstral coefficients match the preset spectral features; if so, determines the target word sequence corresponding to the Mel-cepstral coefficients based on the Mel-cepstral coefficients and the hidden Markov model, identifies the target word sequence, and obtains the speech recognition result; and makes a call based on the speech recognition result. The technical solution provided by each embodiment of the present disclosure can determine whether the extracted Mel-cepstral coefficients corresponding to the first audio data match the preset spectral features. If they match, the target word sequence corresponding to the Mel-cepstral coefficients can be determined, and finally, the call can be made based on the target word sequence. There is no need to directly make a call when facing the audio data of the protected person. Instead, the audio data will be recognized and processed to obtain the sentence actually called by the protected person, thereby realizing the call, improving the accuracy of speech recognition, and greatly reducing the cost of personnel care.
[0130] Figure 2 A schematic diagram of the structure of a calling device provided by an exemplary embodiment of the present disclosure;
[0131] The device includes: an extracting unit 201, a judging unit 202, a determining unit 203, an identifying unit 204 and a calling unit 205;
[0132] An extraction unit 201 is used to extract Mel-cepstral coefficients corresponding to the first audio data;
[0133] A judging unit 202 is used to judge whether the Mel-frequency cepstral coefficient matches a preset frequency spectrum feature;
[0134] A determination unit 203 is configured to determine, if yes, a target word sequence corresponding to the Mel-cepstral coefficients based on the Mel-cepstral coefficients and a hidden Markov model;
[0135] The recognition unit 204 is used to recognize the target word sequence and obtain a speech recognition result;
[0136] The calling unit 205 is used to make a call according to the speech recognition result.
[0137] In some embodiments, the device is further used to:
[0138] Get the original audio data;
[0139] Preprocessing the original audio data to obtain the first audio data;
[0140] The preprocessing includes audio signal amplification, analog-to-digital conversion and environmental noise filtering.
[0141] In some embodiments, the apparatus is used to extract Mel-cepstral coefficients of first audio data, and the apparatus is specifically used to:
[0142] Performing bandpass filtering on the first audio data to obtain second audio data;
[0143] Performing short-time Fourier transform on the second audio data to obtain an energy value of a Mel spectrum corresponding to the second audio data;
[0144] Taking the logarithm of the energy value, obtaining the FBank feature vector;
[0145] Perform discrete cosine transform on the FBank feature vector to obtain the Mel-frequency cepstral coefficient.
[0146] In some embodiments, the device is used to perform short-time Fourier transform on the second audio data to obtain an energy value of the Mel spectrum corresponding to the second audio data, and the device is specifically used to:
[0147] Pre-emphasize the second audio data to obtain third audio data;
[0148] Divide the third audio data into frames to obtain a plurality of third sub-audio data;
[0149] Each of the third sub-audio data is windowed, and each of the windowed third sub-audio data is Fourier transformed to obtain an energy value of a Mel spectrum corresponding to the second audio data.
[0150] In some embodiments, the device is used to determine the target word sequence corresponding to the Mel-cepstral coefficients based on the Mel-cepstral coefficients and the hidden Markov model, and the device is specifically used to:
[0151] Determine the state sequence with the maximum probability for a given observation value sequence based on the hidden Markov model and the Mel-cepstral coefficients;
[0152] Based on the state sequence, a target word sequence corresponding to the Mel-frequency cepstral coefficients is obtained.
[0153] In some embodiments, the device is further used to:
[0154] Based on the Mel-cepstral coefficients, the Gaussian mixture model and the general background model, a plurality of first probability values are obtained;
[0155] If at least one of the multiple first probability values is greater than a preset probability value, the Mel-cepstral coefficient matches a preset feature vector.
[0156] In some embodiments, the device is used to obtain the target word sequence corresponding to the Mel-frequency cepstral coefficient based on the state sequence, and the device is specifically used to:
[0157] Determine a first word sequence corresponding to the state sequence;
[0158] Based on the first audio data and the first word sequence, determining a posterior probability that any word sequence in the first word sequence matches the first audio data under the first audio data;
[0159] Based on the posterior probability and the first word sequence, a target word sequence corresponding to the Mel-cepstral coefficients is determined.
[0160] The technical solution provided by the present disclosure extracts the Mel-cepstral coefficients corresponding to the first audio data; determines whether the Mel-cepstral coefficients match the preset spectral features; if so, determines the target word sequence corresponding to the Mel-cepstral coefficients based on the Mel-cepstral coefficients and the hidden Markov model, identifies the target word sequence, and obtains the speech recognition result; and makes a call based on the speech recognition result. The technical solution provided by each embodiment of the present disclosure can determine whether the extracted Mel-cepstral coefficients corresponding to the first audio data match the preset spectral features. If they match, the target word sequence corresponding to the Mel-cepstral coefficients can be determined, and finally, the call can be made based on the target word sequence. There is no need to directly make a call when facing the audio data of the protected person. Instead, the audio data will be recognized and processed to obtain the sentence actually called by the protected person, thereby realizing the call, improving the accuracy of speech recognition, and greatly reducing the cost of personnel care.
[0161] It should be understood that the device embodiment and the method embodiment may correspond to each other, and similar descriptions may refer to the method embodiment. To avoid repetition, no further description is given here. Specifically, the device may perform the above method embodiment, and the above and other operations and / or functions of each module in the device are the corresponding processes in each method in the above method embodiment, respectively, and no further description is given here for the sake of brevity.
[0162] The above describes the device of the embodiment of the present disclosure from the perspective of the functional module in conjunction with the accompanying drawings. It should be understood that the functional module can be implemented in hardware form, can be implemented by instructions in software form, and can also be implemented by a combination of hardware and software modules. Specifically, the steps of the method embodiment in the embodiment of the present disclosure can be completed by the hardware integrated logic circuit and / or software form instructions in the processor, and the steps of the method disclosed in conjunction with the embodiment of the present disclosure can be directly embodied as a hardware decoding processor to execute, or a combination of hardware and software modules in the decoding processor to execute. Optionally, the software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps in the above method embodiment in conjunction with its hardware.
[0163] Figure 3 is a schematic block diagram of an electronic device provided by an embodiment of the present disclosure, and the electronic device may include:
[0164] The memory 301 and the processor 302, the memory 301 is used to store the computer program and transmit the program code to the processor 302. In other words, the processor 302 can call and run the computer program from the memory 301 to implement the method in the embodiment of the present disclosure.
[0165] For example, the processor 302 may be configured to execute the above method embodiments according to instructions in the computer program.
[0166] In some embodiments of the present disclosure, the processor 302 may include but is not limited to:
[0167] General-purpose processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, etc.
[0168] In some embodiments of the present disclosure, the memory 301 includes but is not limited to:
[0169] Volatile memory and / or non-volatile memory. Among them, the non-volatile memory can be read-only memory (ROM), programmable ROM (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM) or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link DRAM (SLDRAM), and direct RAM bus random access memory (DR RAM).
[0170] In some embodiments of the present disclosure, the computer program may be divided into one or more modules, which are stored in the memory 301 and executed by the processor 302 to complete the method provided by the present disclosure. The one or more modules may be a series of computer program instruction segments capable of completing specific functions, and the instruction segments are used to describe the execution process of the computer program in the electronic device.
[0171] like Figure 3 As shown, the electronic device may also include:
[0172] The transceiver 303 may be connected to the processor 302 or the memory 301 .
[0173] The processor 302 may control the transceiver 303 to communicate with other devices, specifically, to send information or data to other devices, or to receive information or data sent by other devices. The transceiver 303 may include a transmitter and a receiver. The transceiver 303 may further include an antenna, and the number of antennas may be one or more.
[0174] It should be understood that the various components in the electronic device are connected via a bus system, wherein the bus system includes not only a data bus but also a power bus, a control bus and a status signal bus.
[0175] The present disclosure also provides a computer storage medium on which a computer program is stored, and when the computer program is executed by a computer, the computer can perform the method of the above method embodiment. In other words, the present disclosure embodiment also provides a computer program product containing instructions, and when the instructions are executed by a computer, the computer can perform the method of the above method embodiment.
[0176] When software is used for implementation, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function according to the embodiment of the present disclosure is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (digital subscriber line, DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a digital video disc (digital video disc, DVD)), or a semiconductor medium (e.g., a solid state disk (solid state disk, SSD)), etc.
[0177] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this disclosure.
[0178] In the several embodiments provided in the present disclosure, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the module is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.
[0179] The modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. For example, each functional module in each embodiment of the present disclosure may be integrated into a processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.
[0180] The above are only specific embodiments of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present disclosure, which should be included in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be based on the protection scope of the claims.
Claims
1. A calling method, characterized in that: The method comprises: Extracting Mel-cepstral coefficients corresponding to the first audio data; Determining whether the Mel-frequency cepstral coefficient matches a preset frequency spectrum feature; If yes, determining a target word sequence corresponding to the Mel-cepstral coefficients based on the Mel-cepstral coefficients and the hidden Markov model; Recognize the target word sequence to obtain a speech recognition result; A call is made according to the voice recognition result.
2. The method according to claim 1, characterized in that The method further comprises: Get the original audio data; Preprocessing the original audio data to obtain the first audio data; The preprocessing includes audio signal amplification, analog-to-digital conversion and environmental noise filtering.
3. The method according to claim 1, characterized in that Extracting Mel-frequency cepstral coefficients of the first audio data, including: Performing bandpass filtering on the first audio data to obtain second audio data; Performing short-time Fourier transform on the second audio data to obtain an energy value of a Mel spectrum corresponding to the second audio data; Taking the logarithm of the energy value, obtaining the FBank feature vector; Perform discrete cosine transform on the FBank feature vector to obtain the Mel-frequency cepstral coefficient.
4. The method according to claim 3, characterized in that Performing a short-time Fourier transform on the second audio data to obtain an energy value of a Mel spectrum corresponding to the second audio data includes: Pre-emphasize the second audio data to obtain third audio data; Divide the third audio data into frames to obtain a plurality of third sub-audio data; Each of the third sub-audio data is windowed, and each of the windowed third sub-audio data is Fourier transformed to obtain an energy value of a Mel spectrum corresponding to the second audio data.
5. The method according to claim 1, characterized in that Based on the Mel-cepstral coefficients and the hidden Markov model, determining a target word sequence corresponding to the Mel-cepstral coefficients includes: Determine the state sequence with the maximum probability for a given observation value sequence based on the hidden Markov model and the Mel-cepstral coefficients; Based on the state sequence, a target word sequence corresponding to the Mel-frequency cepstral coefficients is obtained.
6. The method according to claim 1, characterized in that The method comprises: Based on the Mel-cepstral coefficients, the Gaussian mixture model and the general background model, a plurality of first probability values are obtained; If at least one of the multiple first probability values is greater than a preset probability value, the Mel-cepstral coefficient matches a preset feature vector.
7. The method according to claim 5, characterized in that The step of obtaining a target word sequence corresponding to the Mel-frequency cepstral coefficients based on the state sequence includes: Determine a first word sequence corresponding to the state sequence; Based on the first audio data and the first word sequence, determining a posterior probability that any word sequence in the first word sequence matches the first audio data under the first audio data; Based on the posterior probability and the first word sequence, a target word sequence corresponding to the Mel-cepstral coefficients is determined.
8. A calling device, applied to a service server, characterized in that: include: An extraction unit, used to extract Mel-cepstral coefficients corresponding to the first audio data; A judging unit, used for judging whether the Mel-frequency cepstral coefficient matches a preset frequency spectrum feature; a determining unit, configured to determine, if yes, a target word sequence corresponding to the Mel-cepstral coefficients based on the Mel-cepstral coefficients and a hidden Markov model; A recognition unit, used to recognize the target word sequence and obtain a speech recognition result; A calling unit is used to make a call according to the speech recognition result.
9. An electronic device, characterized in that: include: processor; as well as A memory, configured to store executable instructions of the processor; The processor is configured to perform the method of any one of claims 1 to 8 by executing the executable instructions.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.