KR20210000802A - Artificial intelligence voice recognition processing method and system - Google Patents

Abstract

The present invention relates to an artificial intelligence voice recognition method and a system thereof, capable of processing automatic response with a voice of an opposite sex different from a voice of a user based on gender recognition of a voice of a user. More specifically, the present invention relates to the artificial intelligence recognition method comprising the steps of: detecting a voice of a user; analyzing the meaning of the voice and the gender and age of the user through the detected voice of the user; automatically selecting a preset response mode according to the analyzed gender and age of the user; and outputting a response corresponding to the meaning of the voice in the selected response mode. When selecting the response mode, a voice of the opposite sex different from the gender of the user is selected.

Description

Artificial intelligence speech recognition processing method and system {ARTIFICIAL INTELLIGENCE VOICE RECOGNITION PROCESSING METHOD AND SYSTEM}

The present invention relates to an artificial intelligence speech recognition processing method and system, and in particular, to an artificial intelligence speech recognition processing method and system configured to enable automatic response processing with a voice of a reason different from the user's voice based on the gender recognition of the user's voice. About.

In general, the response (secretary) voice through artificial intelligence (AI) voice recognition is mostly output as female voices.

Artificial intelligence basically has no gender, but gender is classified by name or voice. Most voice recognition assistants have female names such as "Siri", "Cortana", and Alexa. On the other hand, AI for lawyer work and high-level work response has male names such as "Watson" and "Ross."

AI starts with no gender, but depending on what data is accumulated and how the algorithm is structured, gender can be built.

The fact that such AI assistants have more female voices than males cited some research findings that female voices are more comfortable than male voices, but some argued that it was the result of reflecting stereotypes about gender roles, and the issue of social gender discrimination. And it emerged as a problem of strengthening stereotypes.

In other words, the use of female names and voices for AI voice assistants is an opportunity to reinforce these stereotypes, even if there is no bad meaning, and in medical, legal, and quiz fields, sexual differentiation is further enhanced by using smart, superior male voices. It is encouraging.

Currently, the manufacturer is releasing the product by determining the name (calling language) and gender of the AI voice assistant, and the voice is also heard in the specified gender voice. Therefore, it is not possible to reflect the voices of favorable AI voice assistants according to gender and age of various users.

1 shows a configuration diagram of a conventional general artificial intelligence speech recognition system.

Most of the speech recognition systems have the configuration as shown in FIG. 1, and among the input signals 11 of the speaker 10, only the speech signals uttered by a real person are detected, features are extracted (30), and the reference acoustic model and It is a principle of classifying the pattern (50) by measuring the similarity of (40), and processing it as a language (70) based on the language model (60) to recognize it as a final sentence.

This technology focuses only on the part of improving speech recognition errors, and consists of a system that generates and synthesizes a language stored in a database of conversation processing through recognition sentences, and outputs it as a voice through a speaker. That is, the feature extraction 30 simply recognizes only the gender of the speaker, and focuses only on the technology itself that increases the speech recognition rate with an algorithm that classifies patterns and processes language.

(0001) Korean Patent Publication No. 10-2019-0026518 (0002) Domestic registration patent No. 10-0806025

The technical problem to be solved by the present invention is to provide an artificial intelligence speech recognition processing method and system configured to respond with a voice of a different gender (the opposite sex) from the user based on the gender identified through the user's voice.

Another technical problem to be solved by the present invention is to provide an artificial intelligence speech recognition processing method and system configured to respond with a voice of an age group having favorable sensitivity depending on the age group as well as the gender of the user as described above.

An artificial intelligence speech recognition processing method according to the present invention for achieving the above technical problem comprises: detecting a user's speech; Analyzing the meaning of the voice and the user's gender and age group through the detected user's voice; Automatically selecting a preset response mode according to the analyzed user's gender and age; And outputting a response corresponding to the meaning of the voice in the selected response mode, wherein when the response mode is selected, a voice of a reason different from the user's gender is selected.

In addition, the artificial intelligence speech recognition processing system of the present invention, a voice detection unit for detecting a user's voice; A voice analysis unit that analyzes the meaning of the voice and the user's gender and age group from the user's voice detected through the voice detection unit; A voice processing unit that automatically selects a preset response mode according to the gender and age group of the user analyzed through the voice analysis unit; And a voice output unit that outputs a response corresponding to the meaning of the voice in the selected response mode, wherein when the response mode is selected, a voice of a reason different from the user's gender is selected.

At this time, the gender of the user is characterized by analyzing the vocal cord vibration frequency and jitter.

In addition, the age of the user is characterized by analyzing through a shimmer and a noise-to-harmonics ratio (NHR).

In addition, the analysis of the user's gender and age group includes at least one of a linear prediction coefficient method, a capstrum method, a Melfrequin's capstrum method, an energy spectrum method for each frequency band, a Gaussian mixed model, a neural network model, a support vector machine, and a hidden Markov model. It is characterized by using any one to analyze.

In addition, in the response mode, when the user's voice is neutral, a neutral response is provided.

In addition, it is characterized in that it responds with a voice of an age group lower than that of the user.

The present invention described above has the following effects.

First, by configuring to respond with a voice of the opposite sex different from the user's gender, it is possible to increase the user's affinity and intimacy.

Additionally, by providing a voice of an age lower than that of the user, the likelihood and intimacy can be further increased.

In addition, it is possible to solve the problem of social gender discrimination and gender stereotypes compared to the prior art by the above configuration.

1 is a configuration diagram of a conventional general artificial intelligence speech recognition system,
Figure 2 is a configuration diagram according to an embodiment of the present inventors artificial intelligence speech recognition processing system,
3 is a view showing the vibration frequency of the vocal cords according to sex and age according to the present invention;
4 is a diagram showing jitter by gender and age according to the present invention;
5 is a view showing a seamer by gender and age according to the present invention,
6 is a view showing the NHR by gender and age according to the present invention.
7 is a configuration diagram according to an embodiment of the present inventors artificial intelligence speech recognition processing method,
8 is a diagram showing a method for manually selecting a voice assistant according to an artificial intelligence voice recognition processing method according to the present invention;
9 is a diagram showing a method for automatically selecting a voice assistant according to an artificial intelligence voice recognition processing method according to the present invention.

Hereinafter, some embodiments of the present invention will be described in detail through exemplary drawings. In adding reference numerals to elements of each drawing, it should be noted that the same elements are assigned the same numerals as possible even if they are indicated on different drawings. In addition, in describing the embodiments of the present invention, if it is determined that a detailed description of a related known configuration or function interferes with the understanding of the embodiments of the present invention, the detailed description thereof will be omitted.

In addition, in describing the constituent elements of the embodiment of the present invention, terms such as first, second, A, B, (a), and (b) may be used. These terms are only used to distinguish the component from other components, and the nature, order, or order of the component is not limited by the term. When a component is described as being "connected", "coupled" or "connected" to another component, that component may be directly connected or connected to that other component, but another component between each component It should be understood that may be “connected”, “coupled” or “connected”.

First, the present invention is applicable to all publicly disclosed artificial intelligence speech recognition processing systems, and since artificial intelligence itself is a known technology, detailed descriptions will be omitted.

2 is a configuration diagram according to an embodiment of an artificial intelligence speech recognition processing system according to the present invention, FIG. 3 is a diagram showing vocal cord vibration frequencies by gender and age according to the present invention, and FIG. 4 is a jitter by gender and age according to the present invention. 5 is a view showing a seamer by gender and age according to the present invention, and FIG. 6 is a view showing NHR by gender and age according to the present invention. An artificial intelligence speech recognition processing system according to the present invention will be described with reference to FIGS. 2 to 6 as follows.

Referring to FIG. 1, the artificial intelligence (AI) speech recognition processing system according to the present invention includes a speech detection unit 100, a speech analysis unit 200, a speech processing unit 300, and a speech output unit 400. It is composed by

The voice detection unit 100 detects only the voice of a user (speaker) input through a microphone or the like. The voice detection unit 100 is a technology widely used for voice recognition and plays an important role as a means of operating machines and devices using human voice.

Speech signal recognition technologies are largely classified into speech recognition and speaker recognition. Speech recognition is again divided into a “speaker dependent system” that recognizes only a specific speaker, and a “speaker independent system” that recognizes regardless of the speaker.

Speaker-dependent voice recognition is a technology that stores and registers the user's voice before use, and compares the pattern of the input voice with the pattern of the stored voice when real recognition is performed.

As for the voice input, when a voice signal of a talker (user) is inputted from a microphone, etc., only the voice section should be detected, but the voice detection section has a great influence on recognition performance. In a noisy environment, since noise is often included in the detection section of the speaker's voice signal, it is preferable to configure the voice detection unit 100 to increase the voice recognition rate.

The voice analysis unit 200 grasps the meaning of the voice based on the user's voice information detected through the voice detection unit 100 and analyzes the user's gender and age range. The meaning of voice is to understand what the user intends and can be analyzed through natural language processing. Natural Language Processing (NLP) refers to a technology that transforms human language into a mechanical form through various analysis methods so that a computer can understand it. In addition, it includes the technology to make morphemes that can be interpreted by humans again. Natural language is a sub-field of artificial intelligence. It is a field of study in language engineering, artificial intelligence, and computational linguistics. Natural language refers to a language such as Korean or English that has been used for communication that has occurred naturally over a long time in the past, not a language created artificially by humans like programming languages. Generally speaking, in engineering, programming languages such as C and JAVA come to mind, so the natural language used by humans is called separately. Since the technology itself for grasping the meaning of speech is also a known technology, a description of a specific methodology for grasping the meaning of speech will be omitted.

On the other hand, an example of an artificial intelligence device capable of natural language processing is a Chat Bot. Chat bot is literally a compound word of'chatting' and'robot' and can mean a robot that communicates (chat) like a human. The chatbot may be a home device such as Google Home, Alexa, and Siri manufactured and sold by Google, Amazon, and Apple, or it may be an enterprise device that responds to customers.

The voice analysis unit 200 may include a feature vector calculation unit, a gender detection module unit, and an age detection module unit to analyze a user's gender and age group.

For feature vector calculation, gender detection and age detection, linear prediction coefficient method, capstrum method, Melfrequin's capstrum (MFCC) method, energy spectrum method for each frequency band, Gaussian mixed model (GMM), neural network model (NNM), Analysis can be performed by using at least one of a support vector machine (SVM) and a hidden marker model (HMM), and since the methods and models listed above are also known techniques, detailed descriptions will be omitted.

Referring to FIG. 3,'Fo' denotes a vocal cord vibration frequency and perceptually corresponds to a pitch. Jitter shown in FIG. 4 represents a number showing how constant the period of vibration is.

The shimmer of FIG. 5 is a value showing how constant the amplitude of the vibration is. The more the period or amplitude is irregular, the greater the value of the seamer, and the noise-to-harmonics ratio (NHR) of FIG. 6 exists between 70 and 4,500 Hz. This is the average value of the ratio between the overtones and the abnormal overtones that exist between 1,500 and 4,500Hz. The larger the value, the higher the ratio of noise.

3 and 4, it is the vocal cord vibration frequencies'Fo' and'jitter' that show a significant difference in the sex of the talker.

-Vibration frequency Fo: male 119.02 ± 22.71 Hz, female 199.60 ± 26.93 Hz

-Jitter: Male 0.24 ± 0.15%, female 0.14 ± 0.11%

In addition, referring to FIGS. 5 and 6, it is'Simmer' and'NHR' as follows that show a significant difference according to the age of the talker.

-Seamer: 6.05 ± 5.16% for men, 5.90 ± 4.69% for women

-NHR: Male 0.0192 ± 0.02, Female 0.013 ± 0.01

The feature of the talker may be calculated through a threshold value capable of distinguishing gender and age based on the difference in features as described above.

The voice processing unit 300 automatically selects a preset response mode according to the gender and age group of the user analyzed by the voice analysis unit 200. The response mode is configured to select a voice of another gender opposite to the user's gender, select a neutral voice when the user's voice is neutral, and select a voice of an age lower than the user's age.

In other words, if the user is male, a female voice, if the user is female, a male voice, and if the user is neutral, a neutral voice. In addition, if the user's age range is in their 20s to 30s, they respond with a voice in their 20s. If you are in your 40s to 50s, you can respond with a voice in your 30s, and if your age is in your 60s, you can configure it to respond with a voice in your teens. In this case, when the user's voice is neutral, it may be processed to select a neutral voice in his twenties by default.

The voice output unit 400 outputs a response corresponding to the meaning of the above-described voice in a response mode selected by the voice processing unit 300 through a speaker or the like.

That is, the artificial intelligence speech recognition processing system according to the present invention recognizes the user's gender and age group through the user's voice, and automatically outputs the voice of a gender different from the user's, as well as voices of an age lower than the user's age group. The key technology is to automatically print out. This is based on the fact that men are attracted to the clear voices in the high-pitched range, and women feel attractive and intimate with the male voices in the mid-to-low range.

7 is a configuration diagram according to an embodiment of an artificial intelligence speech recognition processing method of the present inventor, FIG. 8 is a diagram showing a manual selection method of a voice assistant according to the artificial intelligence speech recognition processing method of the present inventor, and FIG. A diagram showing a method of automatically selecting a voice assistant according to an intelligent voice recognition processing method.

First, an artificial intelligence speech recognition processing method according to the present invention will be described with reference to FIG. 7 as follows.

The present inventors' artificial intelligence speech recognition processing method includes the steps of detecting voice (S100), analyzing the user's gender and age (S200), automatically selecting a response mode (S300), and outputting a response ( S400).

The step of detecting the voice (S100) may refer to the description of the above-described voice detection unit 100, and the step of analyzing the user's gender and age range (S200) may refer to the description of the aforementioned voice analysis unit 200. I can. In addition, the step of automatically selecting the response mode (S300) may refer to the description of the above-described voice processing unit 300, and the step of outputting the response (S400) refer to the description of the above-described voice output unit 400. This can be done, so a detailed description will be omitted.

The artificial intelligence speech recognition processing method according to the present invention may be configured in a manual mode and an automatic mode.

An algorithm implementation state of the artificial intelligence speech recognition processing method according to the passive mode of the present invention will be described with reference to FIG. 8 as follows.

First, when the physical button of the product's hardware (the secretary gender selection button) is pressed, the user decides whether or not to select'manual secretary'. At this time, it may be configured to select the'secretary gender selection 3 guidance voice' by listening to the user.

For example, among the voices of artificial intelligence voice assistants registered in the database (DB) in advance, voices in their 20s are output through a speaker to be selected by the user. When the selection of the manual mode is completed, the voice of the artificial intelligence voice assistant selected by the user in advance is output through the speaker, regardless of the gender of the user's voice (calling language).

Next, an algorithm implementation state of the artificial intelligence speech recognition processing method according to the automatic mode of the present invention will be described with reference to FIG. 9.

When the user utters the artificial intelligence voice assistant call word in the'gender detection module unit', it detects the gender with the user's voice command voice regardless of the gender of the caller.

First, the gender is classified by extracting the features of the speaker's voice by referring to the gender reference value of the vocal cord vibration frequency (Fo) and the jitter value, and the classified gender is stored in a'gender recognition register'.

Secondly, in order to classify the age group, the age detection module unit extracts the speaker's features by referring to the age group threshold value of the caller's caller voice and the NHR value, and then classifies the age group and stores it in the'age recognition register'. .

When the classification of the sex and age group of the talker is completed, the voice processing unit 300 outputs the selected voice using the seven voice mode databases (DB) previously stored.

In other words, if the speaker's voice is male, the voice of a female artificial intelligence voice assistant of Lee Sung-hyung responds. Therefore, if the talker is in their 30s, it is treated as a female 20s AI voice assistant, if the talker is in their 40s to 50s, it is treated as a female 30s AI voice assistant, and if the talker is over 60, it is treated as a female teenage AI voice assistant. .

In addition, if the speaker's voice is female, the voice of Lee Sung-hyung's male artificial intelligence voice assistant responds. Therefore, if the talker is in his 30s, it is treated as an artificial intelligence voice assistant for men in his 20s, if the talker is in his 40s to 50s, it is treated as an AI voice assistant in his thirties, and if the talker is in his 60s or older, it is treated as a male teenage AI voice assistant .

In addition, when the speaker's voice is neutral, it is treated as a neutral voice in his 20s.

According to the present invention described above, the voice of the user (speaker) is automatically analyzed, the user's gender and age group are identified through the analyzed voice information, and then a voice of a gender different from the user's gender or a neutral voice is output. And, by outputting a voice of an age lower than that of the user, it is possible to increase the user's affinity and intimacy with the voice of an artificial intelligence voice assistant, and solve the social gender discrimination issue and gender stereotype as in the prior art.

In the above, even if all the constituent elements constituting the embodiments of the present invention are described as being combined into one or operating in combination, the present invention is not necessarily limited to these embodiments. That is, within the scope of the object of the present invention, all of the constituent elements may be selectively combined and operated in one or more. In addition, terms such as "include", "consist of" or "have" described above mean that the corresponding component may be present unless otherwise stated, excluding other components Rather, it should be interpreted as being able to further include other components. All terms, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the art, unless otherwise defined. Terms generally used, such as terms defined in the dictionary, should be interpreted as being consistent with the meaning of the context of the related technology, and are not interpreted as ideal or excessively formal meanings unless explicitly defined in the present invention.

The above description is merely illustrative of the technical idea of the present invention, and those of ordinary skill in the art to which the present invention pertains will be able to make various modifications and variations without departing from the essential characteristics of the present invention. Accordingly, the embodiments posted in the present invention are not intended to limit the technical idea of the present invention, but to explain the technical idea, and the scope of the technical idea of the present invention is not limited by these embodiments. The scope of protection of the present invention should be interpreted by the following claims, and all technical ideas within the scope equivalent thereto should be interpreted as being included in the scope of the present invention.

100: voice detection unit 200: voice analysis unit
300: audio processing unit 400: audio output unit
S100: detecting voice
S200: Analyzing the user's gender and age group
S300: Step of automatically selecting a response mode
S400: Step of outputting a response

Claims (12)

Detecting a user's voice;
Analyzing the meaning of the voice and the user's gender and age group through the detected user's voice;
Automatically selecting a preset response mode according to the analyzed user's gender and age; And
And outputting a response corresponding to the meaning of the voice in the selected response mode; and
Artificial intelligence speech recognition processing method for selecting a voice of the opposite sex different from the user's gender when the response mode is selected. The method of claim 1,
The user's gender is,
Artificial intelligence speech recognition processing method that analyzes the vocal cord vibration frequency and jitter. The method of claim 1,
The user's age is,
Artificial intelligence speech recognition processing method that analyzes through shimmer and noise-to-harmonics ratio (NHR). The method of claim 1,
The user's gender and age group analysis is at least one of a linear prediction coefficient method, a capstrum method, a Melfrequin's capstrum method, an energy spectrum method for each frequency band, a Gaussian mixed model, a neural network model, a support vector machine, and a hidden Markov model. Artificial intelligence speech recognition processing method that is analyzed by using. The method of claim 1,
The response mode,
Artificial intelligence speech recognition processing method that responds with a neutral voice when the user's voice is neutral. The method of claim 5,
An artificial intelligence speech recognition processing method that responds with a voice of an age lower than the user's age. A voice detection unit for detecting a user's voice;
A voice analysis unit that analyzes the meaning of the voice and the user's gender and age group from the user's voice detected through the voice detection unit;
A voice processing unit that automatically selects a preset response mode according to the gender and age group of the user analyzed through the voice analysis unit; And
Includes; a voice output unit for outputting a response corresponding to the meaning of the voice in the selected response mode, and
An artificial intelligence speech recognition processing system that selects a voice of the opposite sex different from the user's gender when the response mode is selected. The method of claim 7,
The user's gender is,
An artificial intelligence speech recognition processing system that analyzes the vocal cord vibration frequency and jitter. The method of claim 7,
The user's age is,
An artificial intelligence speech recognition processing system that analyzes through a shimmer and noise-to-harmonics ratio (NHR). The method of claim 7,
The user's gender and age group analysis is at least one of a linear prediction coefficient method, a capstrum method, a Melfrequin's capstrum method, an energy spectrum method for each frequency band, a Gaussian mixed model, a neural network model, a support vector machine, and a hidden Markov model. Artificial intelligence speech recognition processing system that analyzes by utilizing. The method of claim 7,
The response mode,
Artificial intelligence speech recognition processing system that responds with neutral when the user's voice is neutral. The method of claim 11,
Artificial intelligence speech recognition processing system that responds with voices of an age lower than the user's age.

Images