A call center long speech recognition model training method

Abstract

The present invention discloses a call center long speech recognition model training method, comprising the steps of: obtaining to-be-recognized speech data of the same user in the call center, the to-be-recognized speech data including discontinuous short speech and continuous long speech, and preprocessing the speech data; adopting a speech splicing algorithm to splice the preprocessed discontinuous short speech to obtain a spliced ​​long speech A; adopting a speech cutting and splicing method to process the preprocessed continuous long speech to obtain a spliced ​​long speech B; extracting features of the continuous long speech and the spliced ​​long speech A and the spliced ​​long speech B respectively to obtain a semantic feature vector matrix; inputting the semantic feature vector matrix into a pre-constructed long speech recognition model for training, and obtaining a trained long speech recognition model. The present invention provides a basis for long speech discrimination by combining discontinuous short speech and continuous long speech, and has the advantages of high accuracy and fast recognition speed.

Description

Technical Field

[0001] The invention relates to the field of speech recognition, and in particular to a call center chief speech recognition model training method. Background Art

[0002] With the development of communication technology, more and more large enterprises provide services to customers by establishing call center systems. In large enterprises, they usually establish their own call centers according to their respective division of labor to provide different customer services. Take the call center system established by operators as an example. Usually, the business services provided by operators are classified in detail, so there are many call centers within operators, and different call centers are independent of each other.

[0003] At present, in the customer service of the call center, it is generally divided into menu bar access dialogue and customer service dialogue. The menu bar is relatively complex, and it is difficult for users to quickly locate the menu they want to access. In recent years, voice technology has developed rapidly. Through voice recognition technology, a quick access method for voice menus has been created to improve the shortcut for users to enter the menu; and the customer service dialogue uses artificial intelligence customer service to recognize the customer's voice and then give suggestions or judgments. However, both the menu bar access dialogue and the customer service dialogue rely on the voice recognition system.

[0004] Most existing speech recognition systems / models are designed for the recognition of short speech. Since some users speak at a relatively continuous speed, that is, long speech, the speech recognition system / model is limited in recognition during the recognition of such long speech, and often has low accuracy or recognition errors, and cannot accurately understand the specific semantics of the long speech. Based on this, the present invention proposes a call center long speech recognition model training method. Summary of the invention

[0005] The purpose of the present invention is to provide a call center chief speech recognition model training method to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a call center chief speech recognition model training method, comprising the steps of: S1: Acquire the voice data to be recognized of the same user in the call center, the voice data to be recognized includes discontinuous short voice and continuous long voice, and pre-process the voice data; S2: The pre-processed discontinuous short speech is processed by a speech splicing algorithm to obtain a spliced ​​long speech A; the pre-processed continuous long speech is processed by a speech cutting and splicing method to obtain a spliced ​​long speech B; S3: extracting features of the continuous long speech in step S1 and the concatenated long speech A and concatenated long speech B in step S2 to obtain a semantic feature vector matrix; S4: Input the semantic feature vector matrix into the pre-built long speech recognition model for training to obtain a trained long speech recognition model.

[0007] Preferably, the preprocessing process in step S1 includes: removing noise in the continuous short speech and the continuous long speech by using spectrum subtraction and acoustic analysis processing, and outputting the preprocessed continuous short speech and the continuous long speech after data enhancement and repair in sequence.

[0008] Preferably, the specific process of spectrum subtraction includes: Perform Fourier transform on continuous short speech and continuous long speech to obtain the corresponding amplitude and Phase ; Take a low-frequency signal from the sound signal before transformation, the corresponding frequency width is NIS, and find the average spectrum value of the low-frequency signal segment ; Will be obtained and Substitute it into the spectrum subtraction formula to get the value of the spectrum subtraction output, which is the quadratic value of the signal amplitude after spectrum subtraction. , wherein the formula of the spectral subtraction is: ; Where: , is a constant, is called the oversubtraction factor, It is called the gain compensation factor; Based on the obtained and , the preprocessed sound signal can be obtained through inverse fast Fourier transform.

[0009] Preferably, the speech splicing algorithm in step S2 is based on the semantic splicing logic of high and low frequencies, and its specific splicing process includes: S201a: Determine high-frequency components and low-frequency components of multiple discontinuous short speech segments; S202a: extracting a first cepstrum feature and a second cepstrum feature corresponding to the discontinuous short speech according to the high-frequency component and the low-frequency component; S203a: splicing the first cepstrum feature and the second cepstrum feature in each frame of the discontinuous short speech in units of frames to obtain a parameter sequence; S204a: performing semantic extraction processing on the multiple discontinuous short speech segments to obtain a semantic feature sequence; S205a: Perform speech concatenation based on the parameter sequence and the semantic feature sequence to obtain a concatenated long speech A Preferably, in step S2, the specific process of obtaining the spliced ​​long speech B after the continuous long speech is processed by the speech cutting and splicing method includes: S201b: Segment the continuous long speech to obtain at least two speech segments, wherein there are partially overlapping speech segments between every two adjacent speech segments; S202b: inputting the above-mentioned speech segment into a pre-built speech recognition model for speech recognition processing to obtain speech posterior probability features; S203b: Inputting the speech posterior probability feature and the target timbre identifier into a speech conversion model to perform speech conversion processing to obtain a speech conversion result; S204b: performing voice concatenation processing on at least two voice segments and the voice conversion result to obtain a voice output result; , In the formula, showing the voice output result; Indicates a speech segment; Indicates the result of voice conversion.

[0010] Preferably, the semantic feature vector matrix in step S3 is obtained by the following steps: S301: extracting speech feature sequences corresponding to the continuous long speech, the concatenated long speech A, and the concatenated long speech B through a preset filter group; S302: Obtaining acoustic representations corresponding to speech features in the speech feature sequence through a pre-built encoder; S303: Mapping the latent vector of the acoustic representation to the source language vocabulary through a pre-constructed word embedding matrix to obtain the semantic feature vector matrix.

[0011] Compared with the prior art, the present invention has the following beneficial effects: The present invention can realize long speech recognition, and its recognition logic is based on splicing long speech A and splicing long speech B, wherein the splicing long speech A is obtained by splicing discontinuous short speech using a speech splicing algorithm, and the splicing long speech B is obtained by processing continuous long speech using a speech cutting and splicing method. The basis for long speech discrimination is provided by comprehensively combining discontinuous short speech and continuous long speech, and the invention has the advantages of high accuracy and fast recognition speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a flowchart of the system of the present invention. DETAILED DESCRIPTION

[0013] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0014] See also Figure 1 The present invention provides a technical solution: a call center long speech recognition model training method, comprising the steps of: S1: Acquire the voice data to be recognized of the same user in the call center, the voice data to be recognized includes discontinuous short voice and continuous long voice, and pre-process the voice data; S2: The pre-processed discontinuous short speech is processed by a speech splicing algorithm to obtain a spliced ​​long speech A; the pre-processed continuous long speech is processed by a speech cutting and splicing method to obtain a spliced ​​long speech B; S3: extracting features of the continuous long speech in step S1 and the concatenated long speech A and concatenated long speech B in step S2 to obtain a semantic feature vector matrix; S4: Input the semantic feature vector matrix into the pre-built long speech recognition model for training to obtain a trained long speech recognition model.

[0015] In this embodiment, the preprocessing process in step S1 includes: removing noise in the continuous short speech and the continuous long speech by using spectrum subtraction and acoustic analysis processing, and outputting the preprocessed continuous short speech and the continuous long speech after data enhancement and repair in turn.

[0016] In this embodiment, the specific process of spectrum subtraction includes: Perform Fourier transform on continuous short speech and continuous long speech to obtain the corresponding amplitude and Phase ; Take a low-frequency signal from the sound signal before transformation, the corresponding frequency width is NIS, and find the average spectrum value of the low-frequency signal segment ; Will be obtained and Substitute it into the spectrum subtraction formula to get the value of the spectrum subtraction output, which is the quadratic value of the signal amplitude after spectrum subtraction. , wherein the formula of the spectral subtraction is: ; Where: , is a constant, is called the oversubtraction factor, It is called the gain compensation factor; Based on the obtained and , the preprocessed sound signal can be obtained through inverse fast Fourier transform.

[0017] In this embodiment, the speech splicing algorithm in step S2 is based on the semantic splicing logic of high and low frequencies, and its specific splicing process includes: S201a: Determine high-frequency components and low-frequency components of multiple discontinuous short speech segments; S202a: extracting a first cepstrum feature and a second cepstrum feature corresponding to the discontinuous short speech according to the high-frequency component and the low-frequency component; S203a: splicing the first cepstrum feature and the second cepstrum feature in each frame of the discontinuous short speech in units of frames to obtain a parameter sequence; S204a: performing semantic extraction processing on the multiple discontinuous short speech segments to obtain a semantic feature sequence; S205a: Perform speech concatenation based on the parameter sequence and the semantic feature sequence to obtain a concatenated long speech A In this embodiment, the specific process of obtaining the spliced ​​long speech B after the continuous long speech is processed by the speech cutting and splicing method in step S2 includes: S201b: Segment the continuous long speech to obtain at least two speech segments, wherein there are partially overlapping speech segments between every two adjacent speech segments; S202b: inputting the above-mentioned speech segment into a pre-built speech recognition model for speech recognition processing to obtain speech posterior probability features; S203b: Inputting the speech posterior probability feature and the target timbre identifier into a speech conversion model to perform speech conversion processing to obtain a speech conversion result; S204b: performing voice concatenation processing on at least two voice segments and the voice conversion result to obtain a voice output result; , In the formula, showing the voice output result; Indicates a speech segment; Indicates the result of voice conversion.

[0018] In this embodiment, the semantic feature vector matrix in step S3 is obtained by the following steps: S301: extracting speech feature sequences corresponding to the continuous long speech, the concatenated long speech A, and the concatenated long speech B through a preset filter group; S302: Obtaining acoustic representations corresponding to speech features in the speech feature sequence through a pre-built encoder; S303: Mapping the latent vector of the acoustic representation to the source language vocabulary through a pre-constructed word embedding matrix to obtain the semantic feature vector matrix.

[0019] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments 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 or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application 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 device. 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 one website, computer, server or data center to another website, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0020] Those of ordinary skill in the art will appreciate that the units 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 application.

[0021] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0022] In the several embodiments provided in the present application, 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 units is only one, and there may be other division methods in actual implementation, such as multiple units 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 units, which can be electrical, mechanical or other forms.

[0023] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0024] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0025] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0026] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0027] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

[0028] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A call center voice recognition model training method, characterized in that: Includes steps: S1: Acquire the voice data to be recognized of the same user in the call center, the voice data to be recognized includes discontinuous short voice and continuous long voice, and pre-process the voice data; S2: The pre-processed discontinuous short speech is processed by a speech splicing algorithm to obtain a spliced ​​long speech A; the pre-processed continuous long speech is processed by a speech cutting and splicing method to obtain a spliced ​​long speech B; S3: extracting features of the continuous long speech in step S1 and the concatenated long speech A and concatenated long speech B in step S2 to obtain a semantic feature vector matrix; S4: Input the semantic feature vector matrix into the pre-built long speech recognition model for training to obtain a trained long speech recognition model.

2. A call center manager speech recognition model training method according to claim 1, characterized in that: The preprocessing process in step S1 includes: removing noise in the continuous short speech and the continuous long speech by using spectrum subtraction and acoustic analysis processing, and outputting the preprocessed continuous short speech and the continuous long speech after data enhancement and repair in turn.

3. A call center manager speech recognition model training method according to claim 2, characterized in that: The specific process of the spectrum subtraction includes: Perform Fourier transform on continuous short speech and continuous long speech to obtain the corresponding amplitude and Phase ; Take a low-frequency signal from the sound signal before transformation, the corresponding frequency width is NIS, and find the average spectrum value of the low-frequency signal segment ; Will be obtained and Substitute it into the spectrum subtraction formula to get the value of the spectrum subtraction output, which is the quadratic value of the signal amplitude after spectrum subtraction. , wherein the formula of the spectral subtraction is: ； Where: , is a constant, is called the oversubtraction factor, It is called the gain compensation factor; Based on the obtained and , the preprocessed sound signal can be obtained through inverse fast Fourier transform.

4. The call center manager speech recognition model training method according to claim 1, characterized in that: The speech splicing algorithm in step S2 is based on the semantic splicing logic of high and low frequencies, and its specific splicing process includes: S201a: Determine high-frequency components and low-frequency components of multiple discontinuous short speech segments; S202a: extracting a first cepstrum feature and a second cepstrum feature corresponding to the discontinuous short speech according to the high-frequency component and the low-frequency component; S203a: splicing the first cepstrum feature and the second cepstrum feature in each frame of the discontinuous short speech in units of frames to obtain a parameter sequence; S204a: performing semantic extraction processing on the multiple discontinuous short speech segments to obtain a semantic feature sequence; S205a: Perform speech concatenation based on the parameter sequence and the semantic feature sequence to obtain a concatenated long speech A.

5. The call center manager speech recognition model training method according to claim 1, characterized in that: The specific process of obtaining the spliced ​​long speech B after the continuous long speech is processed by the speech cutting and splicing method in step S2 includes: S201b: Segment the continuous long speech to obtain at least two speech segments, wherein there are partially overlapping speech segments between every two adjacent speech segments; S202b: inputting the above-mentioned speech segment into a pre-built speech recognition model for speech recognition processing to obtain speech posterior probability features; S203b: Inputting the speech posterior probability feature and the target timbre identifier into a speech conversion model to perform speech conversion processing to obtain a speech conversion result; S204b: performing voice concatenation processing on at least two voice segments and the voice conversion result to obtain a voice output result; ， In the formula, showing the voice output result; Indicates a speech segment; Indicates the result of voice conversion.

6. The call center manager speech recognition model training method according to claim 1, characterized in that: The semantic feature vector matrix in step S3 is obtained by the following steps: S301: extracting speech feature sequences corresponding to the continuous long speech, the concatenated long speech A, and the concatenated long speech B through a preset filter group; S302: Obtaining acoustic representations corresponding to speech features in the speech feature sequence through a pre-built encoder; S303: Mapping the latent vector of the acoustic representation to the source language vocabulary through a pre-constructed word embedding matrix to obtain the semantic feature vector matrix.

Images