CN108184032B - A service method and device for a customer service system - Google Patents

Abstract

The invention discloses a service method and device for a customer service system, comprising: receiving a speech synthesis instruction; determining a speech text to be synthesized according to the received speech synthesis instruction; The voice parameter model library established by the timbre of the customer service personnel, synthesizes the voice with the characteristics of the timbre of the customer service staff in the voice text to be synthesized; receives the instructions of the customer service staff, and plays the sentences composed of the synthesized voice and/or the artificial voice of the customer service staff according to the instructions. Since what is played to the user is the synthesized voice with the timbre characteristics of the customer service staff and/or the sentences composed of the artificial voice of the customer service staff, the amount of speech of the customer service staff during the manual service process is greatly reduced, and the fatigue pressure of the customer service staff is reduced. , and users will tacitly assume that the customer service staff has been communicating with them verbally, thereby improving the service quality of the customer service system and enhancing the user experience.

Description

A service method and device for a customer service system

technical field

The present invention relates to the field of communication technologies, and in particular, to a service method and device of a customer service system.

Background technique

At present, the customer service systems of the three major communication companies of China Mobile, China Unicom and China Telecom are usually composed of machine customer service and human customer service. In the process of telephone service, when a conversation message from a user is received, the machine customer service will first perform the service. When the user thinks that the machine customer service cannot solve the problem raised by him, he manually selects the human customer service and consults the human customer service.

In the current customer service system, the voice of the machine customer service is relatively dull and does not sound as vivid as natural language. Moreover, the machine customer service does not have the ability to respond on the spot, and the problems that can be solved are limited. Therefore, in the process of telephone service, manual Customer service plays an important role. However, the manual customer service needs to work continuously for more than 6 hours per shift, and needs to give a large number of answers and explanations according to the user's unused questions and situations during the telephone service process, which is very easy to feel fatigued. Fatigue can cause inaccurate pronunciation or mispronunciation of human customer service personnel, thereby reducing the quality of customer service and affecting user experience.

Therefore, how to improve the service quality of the customer service system, thereby improving the user experience, is a technical problem that needs to be solved urgently.

SUMMARY OF THE INVENTION

Embodiments of the present invention provide a service method and device for a customer service system, which are used to solve the problem in the prior art of how to improve the service quality of the customer service system, thereby improving user experience.

The embodiment of the present invention provides a service method of a customer service system, including:

receive voice synthesis instructions;

According to the received speech synthesis instruction, determine the speech text to be synthesized;

According to the determined voice text to be synthesized, and the voice parameter model library established in advance according to the timbre of the customer service staff currently answering the call, the voice of the voice text to be synthesized with the timbre characteristics of the customer service staff is synthesized;

Receive an instruction from the customer service staff, and play a sentence composed of the synthesized voice and/or the artificial voice of the customer service staff according to the instruction.

In a possible implementation manner, in the above service method provided by the embodiment of the present invention, the determining of the speech text to be synthesized according to the received speech synthesis instruction specifically includes:

Determine whether the speech text to be synthesized corresponding to the received speech synthesis instruction is a standard vocabulary sentence;

If yes, then determine the standard speech sentence corresponding to the speech synthesis instruction as the speech text to be synthesized;

If not, the fill-in-the-blank vocabulary phrase after filling in the text carried by the speech synthesis instruction is used as the speech text to be synthesized.

In a possible implementation manner, in the above-mentioned service method provided by the embodiment of the present invention, the voice parameter model library established according to the voice text to be synthesized determined and the voice parameter model library established in advance according to the voice of the customer service staff currently answering the call , synthesizing the voice with the timbre characteristic of the customer service personnel of the voice text to be synthesized, specifically including:

Using a text analyzer to perform word segmentation on the determined voice text to be synthesized, to obtain a word annotation file corresponding to the voice text to be synthesized;

According to the word annotation file and the voice parameter model library established in advance according to the timbre of the customer service staff currently answering the call, determine the voice feature parameters corresponding to the voice text to be synthesized;

According to the speech feature parameters, the speech with the timbre characteristic of the customer service personnel of the speech text to be synthesized is synthesized.

In a possible implementation manner, in the above service method provided by the embodiment of the present invention, the voice parameter model library established in advance according to the word annotation file and the voice tone of the customer service staff currently answering the call is determined to be the same as the voice parameter model library. The speech feature parameters corresponding to the speech text to be synthesized, specifically including:

Searching for the voice parameter model corresponding to each word in the word annotation file in the voice parameter model library established in advance according to the voice tone of the customer service staff currently answering the call;

According to the speech parameter model corresponding to each word, the parameter generation algorithm is used to determine the LF0 obtained by converting the fundamental frequency information corresponding to the speech text to be synthesized in the log domain, the average value BAP of the aperiodic component spectral information in different frequency bands, and the vocal tract. Spectral information is extracted within the frame of the 18-dimensional line spectrum pair parameter LSP.

In a possible implementation manner, in the above-mentioned service method provided by the embodiment of the present invention, the synthesizing the voice with the timbre feature of the customer service staff in the voice text to be synthesized according to the voice feature parameter specifically includes: :

Using the determined LFO and the BAP to form a mixed excitation source corresponding to the to-be-synthesized speech text;

The determined mixed excitation source is input into a filter, and the determined LSP is used to control the filter to synthesize the voice of the to-be-synthesized voice text with the timbre characteristic of the customer service staff.

In a possible implementation manner, the above-mentioned service method provided by the embodiment of the present invention further includes: establishing a voice parameter model library with the timbre of the customer service personnel in the following manner:

Decomposing the original voice waveform file contained in the voice database of the customer service personnel to obtain fundamental frequency information, aperiodic component spectrum information and channel spectrum information of each syllable in the original voice waveform file;

Convert the fundamental frequency information of each syllable to log domain to obtain LF0;

Taking the average value of the aperiodic component spectrum information of each syllable in each preset frequency band to obtain BAP;

Extracting the 18-dimensional line spectrum pair parameter LSP within the frame of the vocal tract spectrum information of each syllable;

According to the word annotation file corresponding to the original speech waveform file, the LF0, BAP and LSP determined for each syllable are established according to the Hidden Markov Model speech parameter model;

After performing model clustering and model training on each established voice parameter model, a voice parameter model library with the voice of the customer service personnel is obtained.

The embodiment of the present invention also provides a service device of a customer service system, including:

a receiving unit for receiving a speech synthesis instruction;

a determining unit, configured to determine the speech text to be synthesized according to the received speech synthesis instruction;

The synthesis unit is used for synthesizing the voice of the voice text to be synthesized with the timbre feature of the customer service staff according to the determined voice text to be synthesized and the voice parameter model library established in advance according to the timbre of the customer service staff currently answering the call ;

A playing unit, configured to receive an instruction from the customer service staff, and play a sentence composed of the synthesized voice and/or the artificial voice of the customer service staff according to the instruction.

In a possible implementation manner, in the above-mentioned service device provided by the embodiment of the present invention, the determining unit is specifically configured to determine whether the speech text to be synthesized corresponding to the received speech synthesis instruction is a standard speech sentence If it is, then the standard vocabulary sentence corresponding to the speech synthesis instruction is determined as the speech text to be synthesized; if not, then the fill-in-the-blank type vocabulary sentence after the text carried by the speech synthesis instruction is used as the described speech synthesis Speech text to be synthesized.

In a possible implementation manner, in the above service apparatus provided by the embodiment of the present invention, the combining unit includes:

a first synthesis subunit, configured to use a text analyzer to perform word segmentation on the determined speech text to be synthesized, and obtain a word annotation file corresponding to the speech text to be synthesized;

The second synthesis subunit is used to determine the speech feature parameters corresponding to the speech text to be synthesized according to the word annotation file and the speech parameter model library established in advance according to the timbre of the customer service staff currently answering the call;

The third synthesizing subunit is configured to synthesize the voice of the to-be-synthesized voice text with the timbre feature of the customer service personnel according to the voice feature parameter.

In a possible implementation manner, in the above-mentioned service device provided by the embodiment of the present invention, the second synthesis subunit is specifically used in the voice parameter model library established in advance according to the voice of the customer service staff currently answering the call, Find the speech parameter model corresponding to each word in the word labeling file; according to the speech parameter model corresponding to each word, determine the LF0 obtained by converting the fundamental frequency information corresponding to the speech text to be synthesized into the log domain by a parameter generation algorithm, and not The average value BAP of the periodic component spectral information in different frequency bands, and the 18-dimensional line spectral pair parameter LSP extracted from the channel spectral information in the frame.

In a possible implementation manner, in the above-mentioned service apparatus provided by the embodiment of the present invention, the third synthesis subunit is specifically configured to use the determined LFO and the BAP to form the speech to be synthesized with the speech to be synthesized The mixed excitation source corresponding to the text; input the determined mixed excitation source into a filter, and control the filter through the determined LSP, and synthesize the to-be-synthesized voice text with the timbre of the customer service staff characteristic voice.

In a possible implementation manner, the above-mentioned service device provided in the embodiment of the present invention further includes: a modeling unit, configured to decompose the original voice waveform file contained in the voice database of the customer service personnel to obtain the original voice waveform The fundamental frequency information, aperiodic component spectral information and vocal tract spectral information of each syllable in the file; the fundamental frequency information of each syllable is converted to the log domain to obtain LF0; the aperiodic component spectral information of each syllable is Take the average value of each preset frequency band to obtain the BAP; extract the 18-dimensional line spectrum pair parameter LSP in the frame of the channel spectrum information of each syllable; according to the word annotation file corresponding to the original speech waveform file, The LF0, BAP and LSP determined for each syllable are established according to the Hidden Markov Model to establish a speech parameter model; after model clustering and model training are performed on each established speech parameter model, the speech parameters with the timbre of the customer service personnel are obtained. Model library.

The beneficial effects of the present invention are as follows:

The service method and device for a customer service system provided by the embodiments of the present invention include: receiving a speech synthesis instruction; determining the speech text to be synthesized according to the received speech synthesis instruction; The voice parameter model library established by the timbre of the customer service personnel, synthesizes the voice with the characteristics of the timbre of the customer service staff in the voice text to be synthesized; receives the instructions of the customer service staff, and plays the sentences composed of the synthesized voice and/or the artificial voice of the customer service staff according to the instructions. Due to the voice parameter model library established in advance according to the timbre of the customer service staff currently answering the call, the voice with the timbre characteristics of the customer service staff of the voice text to be synthesized is obtained, and the synthesized voice and/or the artificial voice of the customer service staff can be synthesized according to the instructions of the customer service staff. The sentences composed of voice are played to the user, therefore, it can reduce the amount of words spoken by the customer service staff during the manual service process, reduce the fatigue pressure of the customer service staff, thereby improving the service quality of the customer service system and enhancing the user experience. In addition, the voice with the timbre of the customer service staff is played to the user, which sounds vivid and vivid, so that the user does not perceive that there are many machines participating in the interaction process. The default is that the customer service staff has been communicating with them. It improves the service quality of the customer service system and enhances the user experience.

Description of drawings

1 is a flowchart of a service method of a customer service system provided by an embodiment of the present invention;

Fig. 2 is the flow chart of synthesizing the voice with the timbre characteristic of the customer service personnel of the voice text to be synthesized in the embodiment of the present invention;

Fig. 3 is the flow chart of establishing the speech parameter model library with the timbre of customer service personnel in the embodiment of the present invention;

4 is a schematic structural diagram of a service device of a customer service system provided by an embodiment of the present invention;

5 is a parametric speech synthesis system framework based on a hidden Markov model provided by an embodiment of the present invention;

FIG. 6 is a schematic diagram of assisting customer service personnel to provide services through a service device of a customer service system according to an embodiment of the present invention.

Detailed ways

Specific implementations of the service method and device of the customer service system provided by the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

A service method of a customer service system provided by an embodiment of the present invention, as shown in FIG. 1 , specifically includes the following steps:

S101. Receive a voice synthesis instruction;

S102, according to the received speech synthesis instruction, determine the speech text to be synthesized;

S103, according to the determined voice text to be synthesized, and the voice parameter model library established in advance according to the timbre of the customer service staff currently answering the call, synthesize the voice with the timbre feature of the customer service staff of the voice text to be synthesized;

S104. Receive an instruction from the customer service personnel, and play a sentence composed of synthesized speech and/or an artificial voice of the customer service personnel according to the instruction.

Specifically, in the above-mentioned service method provided by the embodiment of the present invention, the voice with the timbre characteristic of the customer service staff of the voice text to be synthesized is obtained according to the voice parameter model library established in advance according to the timbre of the customer service staff currently answering the call, and can be According to the instructions of the customer service staff, the sentence composed of the synthesized voice and/or the artificial voice of the customer service staff is played to the user. Therefore, the amount of speech of the customer service staff during the manual service process can be reduced, the fatigue pressure of the customer service staff can be reduced, and the customer service system can be improved. quality of service and enhanced user experience. In addition, the voice with the timbre of the customer service staff is played to the user, which sounds vivid and vivid, so that the user does not perceive that there are many machines participating in the interaction process. The default is that the customer service staff has been communicating with them. It improves the service quality of the customer service system and enhances the user experience.

During specific implementation, in the above-mentioned service method provided by the embodiment of the present invention, step S102 determines the speech text to be synthesized according to the received speech synthesis instruction, which can be specifically implemented in the following ways:

Determine whether the speech text to be synthesized corresponding to the received speech synthesis instruction is a standard vocabulary sentence;

If so, the standard speech sentence corresponding to the speech synthesis instruction is determined as the speech text to be synthesized;

If not, the fill-in-the-blank vocabulary phrase after filling in the text carried by the speech synthesis instruction is used as the speech text to be synthesized.

Specifically, in the above-mentioned service method provided by the embodiment of the present invention, the standard vocabulary sentences in the specific implementation of step S102 are some basic communication sentences used by the customer service personnel when serving the user's telephone, for example: "I am very happy At your service", "Please enter your ID number". In addition, in the process of playing the standard vocabulary sentence to the user, if either the user or the customer service staff speaks, the voice playback can be stopped at any time, so as to ensure good interaction between the customer service staff and the user and improve the user experience.

Specifically, in the above service method provided by the embodiment of the present invention, the fill-in-the-blank vocabulary sentence in the specific implementation of step S102 is a sentence that needs to be grouped according to the actual consumption situation or traffic situation of the user. For example: "Your current call charge balance is XX yuan", where XX is the data in the billing system, which needs to be filled in a fixed sentence pattern, and then the online synthesized voice output is performed through the personalized speech synthesis technology. Of course, there are other ways to implement the fill-in-the-blank vocabulary sentence. For example, still taking "Your current call balance is XX yuan" as an example, you can only output "Your current call balance is yuan" for speech synthesis, and The call balance "XX" can be stated by the customer service staff, which is not limited here.

In specific implementation, in the above service method provided by the embodiment of the present invention, step S103 synthesizes the speech text to be synthesized according to the determined speech text to be synthesized and the speech parameter model library established in advance according to the tone color of the customer service staff currently answering the call The voice with the timbre characteristics of the customer service staff, as shown in Figure 2, can specifically include the following steps:

S201, using a text analyzer to perform word segmentation on the determined voice text to be synthesized, to obtain a word annotation file corresponding to the voice text to be synthesized;

S202, according to the word annotation file and the voice parameter model library established in advance according to the timbre of the customer service staff currently answering the call, determine the voice feature parameters corresponding to the voice text to be synthesized;

S203. According to the speech feature parameters, synthesizing the speech with the timbre characteristic of the customer service personnel of the speech text to be synthesized.

Specifically, in the above-mentioned service method provided by the embodiment of the present invention, for example, taking "I'm glad to serve you" as the speech text to be synthesized as an example, using a text analyzer can get "very", "high", "happy" and "wei" "You", "Service", "Service" and their corresponding annotation files; then combined with the annotation files, in the voice parameter model library pre-established according to the voice of the customer service staff currently answering the call, you can find the words "very" and "high". "Xing", "Wei", "You", "Service" and "Service" correspond to the speech feature parameters respectively; finally, according to the corresponding speech feature parameters found, the voice "I'm glad for you" can be synthesized with the voice characteristics of the customer service staff. service" voice.

During specific implementation, in the above service method provided by the embodiment of the present invention, step S202 determines the speech feature corresponding to the speech text to be synthesized according to the word annotation file and the speech parameter model library established in advance according to the timbre of the customer service staff currently answering the call parameters, which can be implemented in the following ways:

Search for the voice parameter model corresponding to each word in the word annotation file in the voice parameter model library established in advance according to the voice of the customer service staff currently answering the call;

According to the speech parameter model corresponding to each word, the parameter generation algorithm is used to determine the LF0 obtained by converting the fundamental frequency information corresponding to the speech text to be synthesized in the log domain, the average value BAP of the aperiodic component spectrum information in different frequency bands, and the channel spectrum information. The 18-dimensional line spectrum extracted within the frame is paired with the parameter LSP.

Specifically, in the above service method provided by the embodiment of the present invention, in order to improve the quality of the synthesized speech, in the specific implementation of step S202, the average value BAP of the aperiodic component spectrum information in different frequency bands may be the aperiodic component spectrum Ap The BAP is obtained by averaging 5 frequency bands, wherein the 5 frequency bands can be 0~1000Hz, 1000~2000Hz, 2000~4000Hz, 4000~6000Hz, 6000~8000Hz, which are not limited here.

In specific implementation, in the above-mentioned service method provided by the embodiment of the present invention, step S203, according to the speech feature parameters, synthesizes the voice with the timbre feature of the customer service personnel to be synthesized into the voice text, which can be specifically implemented in the following ways:

Using the determined LF0 and BAP to form a mixed excitation source corresponding to the speech text to be synthesized;

The determined mixed excitation source is input into the filter, and the filter is controlled by the determined LSP to synthesize the voice with the timbre characteristic of the customer service personnel of the voice text to be synthesized.

During specific implementation, in the above-mentioned service method provided by the embodiment of the present invention, it may further include: establishing a voice parameter model library with the timbre of the customer service personnel in the following manner, as shown in FIG. 3 :

S301, decompose the original voice waveform file contained in the voice database of the customer service personnel, and obtain fundamental frequency information, aperiodic component spectrum information and channel spectrum information of each syllable in the original voice waveform file;

S302, convert the fundamental frequency information of each syllable to log domain to obtain LF0;

S303 , taking the average value of the aperiodic component spectrum information of each syllable in each preset frequency band to obtain a BAP; wherein, each preset frequency band can be 0~1000Hz, 1000~2000Hz, 2000~4000Hz, 4000~ 6000HZ, 6000~8000Hz, not limited here;

S304, extract the 18-dimensional line spectrum pair parameter LSP within the frame of the channel spectrum information of each syllable;

S305, according to the word annotation file corresponding to the original speech waveform file, establish a speech parameter model according to the hidden Markov model for the LF0, BAP and LSP determined for each syllable;

S306 , after performing model clustering and model training on each established speech parameter model, a speech parameter model library having the timbre of the customer service personnel is obtained.

It should be noted that, the order of steps S302-S304 in the above service method provided by the embodiment of the present invention may be interchanged, and is not limited to the order described above.

Based on the same inventive concept, an embodiment of the present invention also provides a service device for a customer service system. Since the principle of the service device for solving problems is similar to the above-mentioned service method, the implementation of the service device can refer to the implementation of the above-mentioned service method. The repetition will not be repeated.

The service device of the customer service system provided by the embodiment of the present invention, as shown in FIG. 4 , may include:

a receiving unit 401 for receiving a speech synthesis instruction;

Determining unit 402, configured to determine the speech text to be synthesized according to the received speech synthesis instruction;

The synthesis unit 403 is used to synthesize the voice of the voice text to be synthesized with the timbre feature of the customer service staff according to the determined voice text to be synthesized and the voice parameter model library established in advance according to the timbre of the customer service staff currently answering the call;

The playing unit 404 is configured to receive an instruction from the customer service staff, and play a sentence composed of synthesized speech and/or artificial voice of the customer service staff according to the instruction.

During specific implementation, in the above-mentioned service device provided by the embodiment of the present invention, the determining unit 402 may be specifically configured to determine whether the speech text to be synthesized corresponding to the received speech synthesis instruction is a standard speech sentence; The standard speech sentence corresponding to the synthesis instruction is determined as the speech text to be synthesized; if not, the fill-in-the-blank speech speech sentence after filling in the text carried by the speech synthesis instruction is used as the speech text to be synthesized.

During specific implementation, in the above service apparatus provided by the embodiment of the present invention, the combining unit 403 may include:

The first synthesizing subunit 4031 is used to perform word segmentation on the determined speech text to be synthesized by using a text analyzer, and obtain a word annotation file corresponding to the speech text to be synthesized;

The second synthesis subunit 4032 is used to determine the speech feature parameters corresponding to the speech text to be synthesized according to the word annotation file and the speech parameter model library established in advance according to the timbre of the customer service personnel currently answering the call;

The third synthesizing subunit 4033 is configured to synthesize, according to the speech feature parameters, the voice with the timbre characteristic of the customer service staff of the voice text to be synthesized.

In specific implementation, in the above-mentioned service device provided by the embodiment of the present invention, the second synthesis subunit 4032 can be specifically used for searching and labeling words in the voice parameter model library established in advance according to the voice of the customer service staff currently answering the call The speech parameter model corresponding to each word in the file; according to the speech parameter model corresponding to each word, the LF0 obtained by converting the fundamental frequency information corresponding to the speech text to be synthesized by converting the log domain is determined by the parameter generation algorithm, and the aperiodic component spectral information is in different frequency bands. The average value of BAP, and the 18-dimensional line spectrum pair parameter LSP extracted from the channel spectrum information in the frame.

During specific implementation, in the above-mentioned service device provided by the embodiment of the present invention, the third synthesis subunit, 4033 may be specifically configured to use the determined LF0 and BAP to form a mixed excitation source corresponding to the speech text to be synthesized; The mixed excitation source is input into the filter, and the filter is controlled by the determined LSP to synthesize the voice with the timbre characteristic of the customer service personnel of the voice text to be synthesized.

During specific implementation, the above-mentioned service apparatus provided in the embodiment of the present invention may further include: a modeling unit 405, configured to decompose the original voice waveform files contained in the voice database of the customer service personnel, and obtain each of the original voice waveform files. The fundamental frequency information, aperiodic component spectral information and vocal tract spectral information of the syllable; the fundamental frequency information of each syllable is converted to the log domain to obtain LF0; the aperiodic component spectral information of each syllable is divided into preset frequency bands. Take the average value to obtain BAP; extract the 18-dimensional line spectrum pair parameter LSP from the channel spectrum information of each syllable in the frame; according to the word annotation file corresponding to the original speech waveform file, determine the LF0, BAP and LSP for each syllable The speech parameter model is established according to the hidden Markov model; after the model clustering and model training are performed on each established speech parameter model, the speech parameter model library with the timbre of the customer service personnel is obtained.

In order to better understand the technical solutions of the present invention, the present invention provides specific embodiments of establishing a voice parameter model library with the timbre of the customer service personnel and synthesizing the voice text with the timbre characteristics of the customer service personnel in the above-mentioned service method. The parametric speech synthesis system framework of the hidden Markov model is shown in Figure 5:

Part A in FIG. 5 shows a specific embodiment of establishing a speech parameter model library with the timbre of the customer service personnel. The voice database of the target customer service personnel contains the original voice waveform file in wav format and the corresponding label file. The original speech waveform file is effectively decomposed into source information and channel information through adaptive weighted general interpolation technology, namely STRAIGHT analysis technology. The source information includes fundamental frequency F0 and aperiodic component spectrum AP, and channel information is channel information. Spectrum SP. Then, further processing converts the fundamental frequency F0 to the log domain to obtain LF0; the aperiodic component spectrum Ap is averaged according to 5 frequency bands to obtain BAP, wherein the 5 frequency bands are 0~1000Hz, 1000~2000Hz, 2000~4000Hz, 4000~6000Hz, 6000~8000Hz; extract the 18-dimensional line spectrum pair parameter LSP from the channel spectrum sp in the frame. Finally, combined with the parameter combination of LF0, BAP and LSP in the label file, the hidden Markov model is used to establish a speech parameter model, and then the model clustering and model training are performed on the established speech parameter models, and the target is obtained in about 3 cycles. Voice parameter model for customer service personnel.

Part B in FIG. 5 shows a specific embodiment of the voice having the timbre characteristic of the customer service staff to synthesize the voice text to be synthesized. The text of the speech to be synthesized is obtained through the text analyzer to obtain the label form of the annotation file required for synthesis, and then, combined with the speech parameter model library of the target customer service personnel obtained in Part A in Figure 5, the speech feature parameters corresponding to the speech text to be synthesized are found, LF0, BAP and LSP. Finally, use LF0 and BAP to form a mixed excitation source corresponding to the speech text to be synthesized; input the determined mixed excitation source into the filter, and control the filter through the determined LSP to synthesize the speech text to be synthesized. timbre characteristic of the voice.

In addition, the present invention also provides a specific embodiment in which the customer service personnel realize the voice service through the above-mentioned service method and service device, as shown in FIG. 6 :

After the customer service staff accesses the user's phone, they can wait to synthesize voice text with standard vocabulary sentences and fill-in-the-blank vocabulary sentences, and synthesize a voice with the customer service staff's timbre through the above-mentioned service device, and play them to the user. For example, "Hello, I'm glad to serve you" is a standard phraseology sentence, and the voice with the voice of the customer service staff is synthesized by the above-mentioned service device and played to the user. For another example, when the user needs to handle or change the business, the above-mentioned service device can generate a voice with the voice of the customer service staff in the standard phrase "please input your ID number", and play it to the user. And in order to ensure a better user experience, the voice playback can be stopped at any time. When the user inquires about the balance of the call charge, it is necessary to fill in the data XX of the call charge balance corresponding to the user currently inquiring in the billing system into the fixed sentence "Your current call charge balance is Yuan", and then the sentence "Your current call balance is XX Yuan" filled in the call balance XX is synthesized and output by the above service device. It can be seen that the customer service personnel only need to adjust the answer content at any time according to the user's communication method. In the above two cases, the voice with its own timbre characteristics can be played to the customer, and the user can perceive The reason is that the customer service staff is communicating with it, and the experience is better.

The service method and device for the above-mentioned customer service system provided by the embodiments of the present invention include: receiving a speech synthesis instruction; determining the speech text to be synthesized according to the received speech synthesis instruction; The voice parameter model library established by the voice of the customer service staff of the voice, synthesizes the voice with the voice characteristics of the customer service staff to be synthesized into the voice text; receives the instructions of the customer service staff, and plays the sentences composed of the synthesized voice and/or the artificial voice of the customer service staff according to the instructions. . Due to the voice parameter model library established in advance according to the timbre of the customer service staff currently answering the call, the voice with the timbre characteristics of the customer service staff of the voice text to be synthesized is obtained, and the synthesized voice and/or the artificial voice of the customer service staff can be synthesized according to the instructions of the customer service staff. The sentences composed of voice are played to the user, therefore, it can reduce the amount of words spoken by the customer service staff during the manual service process, reduce the fatigue pressure of the customer service staff, thereby improving the service quality of the customer service system and enhancing the user experience. In addition, the voice with the timbre of the customer service staff is played to the user, which sounds vivid and vivid, so that the user does not perceive that there are many machines participating in the interaction process. The default is that the customer service staff has been communicating with them. It improves the service quality of the customer service system and enhances the user experience.

In addition, personalized speech synthesis technology is a technology for synthesizing the voice of the target speaker by establishing a speech feature model of the target speaker. This technology first collects recording materials with certain phoneme coverage, then extracts the speech features of the speaker's characteristics, establishes the feature model of the target speaker, and then for any sentence text, the speech parameter features of the text can be generated through the model, and finally through the The vocoder synthesizes the voice of the text with the characteristics of the target speaker. The current speech synthesis technologies are mainly waveform splicing speech synthesis technology and parametric speech synthesis technology.

However, at present, the speech synthesis technology is only used for voice broadcast in the field of customer service, and is not widely used in other applications in the field of customer service. However, in the service method and device of the customer service system provided by the embodiments of the present invention, a new application scenario of the speech synthesis technology in the customer service field is created, and the personalized speech synthesis technology is used in the process of incoming and outgoing calls of the customer service, extremely It greatly reduces the workload of customer service personnel, thereby improving the service quality and user experience of the customer service system, and has broad application prospects.

It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit and scope of the invention. Thus, provided that these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims (10)

1. A service method of a customer service system, comprising:

receiving a voice synthesis instruction;

determining a voice text to be synthesized according to the received voice synthesis instruction;

synthesizing the voice with the tone characteristics of the customer service personnel of the voice text to be synthesized according to the determined voice text to be synthesized and a voice parameter model library established in advance according to the tone of the customer service personnel currently receiving the call;

receiving an instruction of the customer service staff, and playing a sentence consisting of synthesized voice and/or artificial voice of the customer service staff according to the instruction;

the determining a speech text to be synthesized according to the received speech synthesis instruction specifically includes:

determining whether the received voice text to be synthesized corresponding to the voice synthesis instruction is a standard dialect sentence;

if so, determining a standard dialect sentence corresponding to the voice synthesis instruction as the voice text to be synthesized;

and if not, taking the filling-in type speech term sentence filled with the text carried by the speech synthesis instruction as the speech text to be synthesized.

2. The service method according to claim 1, wherein the synthesizing the voice of the voice text to be synthesized with the timbre feature of the customer service person based on the determined voice text to be synthesized and a speech parameter model library established in advance according to the timbre of the customer service person currently called comprises:

adopting a text analyzer to perform word segmentation on the determined voice text to be synthesized to obtain a word annotation file corresponding to the voice text to be synthesized;

determining a voice characteristic parameter corresponding to the voice text to be synthesized according to the word annotation file and a voice parameter model library which is established in advance according to the tone of the customer service personnel receiving the call currently;

and synthesizing the voice with the timbre characteristics of the customer service staff of the voice text to be synthesized according to the voice characteristic parameters.

3. The service method according to claim 2, wherein the determining the speech feature parameters corresponding to the speech text to be synthesized according to the word annotation file and a speech parameter model library pre-established according to the tone of the customer service person currently receiving the call specifically comprises:

searching a voice parameter model corresponding to each word in the word annotation file in a voice parameter model library which is established in advance according to the tone of the customer service staff receiving the call currently;

and according to the speech parameter model corresponding to each word, determining LF0 obtained by converting the fundamental frequency information corresponding to the speech text to be synthesized into a log domain, an average value BAP of the non-periodic component spectrum information on different frequency bands and an 18-dimensional line spectrum pair parameter LSP extracted from the sound channel spectrum information in a frame through a parameter generation algorithm.

4. The service method according to claim 3, wherein the synthesizing the voice of the voice text to be synthesized with the timbre features of the customer service person according to the speech feature parameters specifically comprises:

forming a mixed excitation source corresponding to the speech text to be synthesized by using the determined LF0 and the BAP;

inputting the determined mixed excitation source into a filter, controlling the filter through the determined LSP, and synthesizing the voice of the voice text to be synthesized with the timbre characteristics of the customer service staff.

5. The service method of any one of claims 1-4, further comprising: establishing a voice parameter model library with the tone of the customer service staff by the following method:

decomposing an original voice waveform file contained in a voice database of a customer service staff to obtain fundamental frequency information, non-periodic component spectrum information and vocal tract spectrum information of each syllable in the original voice waveform file;

converting the fundamental frequency information of each syllable into a log domain to obtain LF 0;

respectively averaging the aperiodic component spectrum information of each syllable in each preset frequency band to obtain BAP;

extracting 18-dimensional line spectrum pair parameters LSP from the sound channel spectrum information of each syllable in a frame;

according to the word label file corresponding to the original voice waveform file, establishing a voice parameter model for LF0, BAP and LSP determined for each syllable according to a hidden Markov model;

and carrying out model clustering and model training on each established voice parameter model to obtain a voice parameter model library with the tone of the customer service staff.

6. A service device of a customer service system, comprising:

a receiving unit for receiving a voice synthesis instruction;

the determining unit is used for determining a voice text to be synthesized according to the received voice synthesis instruction;

the synthesis unit is used for synthesizing the voice with the tone characteristics of the customer service staff of the voice text to be synthesized according to the determined voice text to be synthesized and a voice parameter model library established in advance according to the tone of the customer service staff who receives the call currently;

the playing unit is used for receiving the instruction of the customer service staff and playing the sentence consisting of the synthesized voice and/or the artificial voice of the customer service staff according to the instruction;

the determining unit is specifically configured to determine whether a speech text to be synthesized corresponding to the received speech synthesis instruction is a standard conversational sentence; if so, determining a standard dialect sentence corresponding to the voice synthesis instruction as the voice text to be synthesized; and if not, taking the filling-in type speech term sentence filled with the text carried by the speech synthesis instruction as the speech text to be synthesized.

7. The service apparatus of claim 6, wherein the synthesis unit comprises:

the first synthesis subunit is used for performing word segmentation on the determined voice text to be synthesized by adopting a text analyzer to obtain a word annotation file corresponding to the voice text to be synthesized;

the second synthesis subunit is used for determining the voice characteristic parameters corresponding to the voice text to be synthesized according to the word annotation file and a voice parameter model library which is established in advance according to the tone of the customer service personnel receiving the call currently;

and the third synthesis subunit is used for synthesizing the voice of the voice text to be synthesized with the timbre characteristics of the customer service staff according to the voice characteristic parameters.

8. The service device according to claim 7, wherein the second synthesizing subunit is specifically configured to search a speech parameter model corresponding to each term in the term annotation file in a speech parameter model library previously established according to a tone of a currently called customer service person; and according to the speech parameter model corresponding to each word, determining LF0 obtained by converting the fundamental frequency information corresponding to the speech text to be synthesized into a log domain, an average value BAP of the non-periodic component spectrum information on different frequency bands and an 18-dimensional line spectrum pair parameter LSP extracted from the sound channel spectrum information in a frame through a parameter generation algorithm.

9. The service device according to claim 8, wherein the third synthesis subunit is specifically configured to form a hybrid excitation source corresponding to the speech text to be synthesized using the determined LF0 and the BAP; inputting the determined mixed excitation source into a filter, controlling the filter through the determined LSP, and synthesizing the voice of the voice text to be synthesized with the timbre characteristics of the customer service staff.

10. The service apparatus according to any one of claims 6 to 9, further comprising: the modeling unit is used for decomposing an original voice waveform file contained in a voice database of customer service personnel to obtain fundamental frequency information, aperiodic component spectrum information and vocal tract spectrum information of each syllable in the original voice waveform file; converting the fundamental frequency information of each syllable into a log domain to obtain LF 0; respectively averaging the aperiodic component spectrum information of each syllable in each preset frequency band to obtain BAP; extracting 18-dimensional line spectrum pair parameters LSP from the sound channel spectrum information of each syllable in a frame; according to the word label file corresponding to the original voice waveform file, establishing a voice parameter model for LF0, BAP and LSP determined for each syllable according to a hidden Markov model; and carrying out model clustering and model training on each established voice parameter model to obtain a voice parameter model library with the tone of the customer service staff.

Images