TW201008222A - A mobile phone for emotion recognition of incoming-phones and a method thereof - Google Patents

Abstract

A mobile phone for emotion recognition of incoming-phones is provided. The mobile phone may include a speech recording unit, an A/D converter, a character extracting unit, an emotion classifier, and an emotion report outputting unit. The mobile phone may further include a memory that stores a plurality of emotion data corresponding to characters of speeches. The mobile phone can recognize emotions of speaks, including anger, happiness, sadness, boredom and neutral, during the communication of the talk. A related method for emotion recognition of incoming-phones is also provided.

Description

201008222 IX. Description of the Invention: [Technical Field of the Invention] The present invention relates to voice-based technology, and more particularly to the method of emotion recognition. m [Prior Art] According to research, humans have a total of five basic erroneous libraries, including Anger, Bored, Happy, and Sadness. At present, busy modern people 彝Μ6·1) and Dan taxpayers, friends, colleagues, often use the phone as a medium for communication and contact feelings, such as the long-term U-management is not face-to-face, so often do not know the other party at the current = on the phone The state of affairs, sometimes because of the failure to correctly understand the emotions of the other party's speech, so that the wrong words cause the σ angle of both sides, 1 misunderstanding the other party will. Today's mobile phones, if they can provide Μ information in this respect, thus recognizing the ambiguity of the other party's speech _ step emotions between the emotions of the mosquitoes to enhance the lack of results (four) for people and people [invention content] in view of the above, it is necessary梃 也 ^ _ ^ ^ To & for a realization of the caller's illegitimate, eve state. _^ Cheng _ 崎 之 情绪 - 种 种 种 种 种 种 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _: 语二3 ^ material and silent voice: (four), and the root of the voice voice signal frequency 201008222 - rate size from the voice signal to extract different feature parameters; emotional classifier, used to read according to different characteristic parameters Taking the emotional feature data corresponding to the voiced voice signal, and classifying and counting the read emotional feature data to generate classified statistical data of the emotional feature; and an emotion output unit for generating the incoming caller according to the classified statistical data generated by the emotional classifier Emotional analysis report. A method for recognizing a phone call emotion, the method comprising the steps of: recording a call voice of the other party as an analog voice signal; converting the analog voice signal into a digital voice signal; and sounding the digital voice signal by using the endpoint detection principle The voice data and the silent voice data are cut out; the different feature parameters are extracted from the voiced voice signal according to the frequency of the voiced voice signal; the emotional feature data corresponding to the voiced voice signal is read according to different feature parameters; The feature data is classified and statistically generated to generate classified statistical data of the emotional features; and the sentiment analysis report of the incoming caller is generated according to the classified statistical data. • Compared with the prior art, the mobile phone and the method for realizing the emotional recognition of the incoming call can recognize the emotional state of the other party during the mobile phone call, thereby improving the call quality between the two parties. [Embodiment] Referring to FIG. 1, a structural diagram of a preferred embodiment of a mobile phone 8 for implementing call emotion recognition according to the present invention is shown. In this embodiment, the mobile phone 8 includes a voice recording unit 1, a digital-to-analog (A/D) converter 2, a feature capturing unit 3, a memory 4, an emotion classifier 5, an emotion output unit 6, and a display fluorescent device. Curtain 7. The voice recording unit 1 is used to record the incoming call voice of the other party as an analogy 201008222 audio signal ' and transmit the analog voice signal to her converter 2. The converter 2 is configured to convert the analog voice signal into a digital signature feature capture unit 3 for detecting the original voice through the endpoint: the number: the voiced voice data and the voice data in the language are cut, the digital voice signal The cap takes (four) voice 峨, and according to the frequency of the audio signal ^ frequency, different characteristic parameters are extracted from the voiced voice signal. How to use the principle of endpoints to compare the voiced data in digital voice signals with and without

The voice data into (4) #m is described in detail in the lower ® 2 towel. The: sign parameter is used to describe the acoustic parameters of the speech feature, such as the cepstral coefficient CMel-FrequencyCepstmm C〇efficients, MFCC). The memory 4 is used to store emotional feature data corresponding to different feature parameters. For example, a feature parameter A corresponds to an emotional profile (eg, angry "angry"). The emotional feature data is predefined by the mobile phone manufacturer. In the embodiment, the emotional feature data is directly stored in the memory 4 of the mobile phone 8. In other embodiments, the emotional profile data can be stored in a network of mobile phone operators. The emotion classifier 5 is configured to read the emotional feature data corresponding to the voiced voice signal from the memory 4 according to different feature parameters, and classify and calculate the read emotional feature data to generate classified statistical data of the emotional feature. The emotion classifier 5 classifies the read emotional feature data by using the principle that the similar data has the same characteristics. For example, if the MFCC values of the two voice signals are not more than a preset value, the two voice signals are A similar voiced voice signal corresponds to the same emotional feature (eg, angry "angry"). In the present embodiment, the emotion classifier 5 judges the current emotion of the opponent according to the emotional feature with the highest statistical value in the classified statistical data of the 201008222 emotion, for example, if the classification statistics of the emotional feature are: sadness = 4 , angryness (angry) = 2, happiness (happy) =1 'neutrality (n_al) =1 and tiredness (bored) = 〇, the emotion classifier 5 determines that the emotion category is "sadness" ,,.

The emotion output unit 6 is configured to generate an emotion analysis report of the calling party based on the classification statistics of the emotion characteristics, and output the emotion analysis report and display it on the display screen 7 of the mobile phone 8. The sentiment analysis report includes anger, tiredness, happiness, normality, and sadness, so that the user can understand the emotional state of the other party during the call. Referring to Fig. 2, the feature capturing unit 3 in Fig. 1 cuts the schematic diagram of the voiced voice and the voiceless voice by using the endpoint detection principle. In this embodiment, the main purpose of the endpoint detection is to cut the voiced data and the unvoiced data in the voice signal by the energy or the zero rate in the voice signal at a certain time. As shown in FIG. 2, "Enl" represents a conservative value of energy. If the energy of the voice signal is less than or equal to the conservative value "Enl" of the energy, the feature extraction unit 3 determines that the voice signal is silent voice; if the energy of the voice signal is greater than The energy conservation value "Enl", the feature extraction unit 3 determines that the voice signal is voiced speech. "En2" indicates a starting energy value larger than "Enl". If the voice signal energy of "U" at a certain time is greater than the energy value "En2", then the time "tl" is the beginning of the voiced voice signal. "EnEnd" indicates an end energy value that is smaller than "Enl,". If the voice signal energy of a certain time "t2" is less than the energy value EnEnd", the time "t2" is the end of the voiced signal. The feature extracting unit 3 cuts out the voice data and the silent voice data from the time value "tl" to the time "t2" according to the energy value of the 201008222 slang. In the figure, the principle that the zero-rate "ZCR" is used to cut out the voice signal is the same. Therefore, the method of the present invention is not described in detail. Analog map. The five-tone recording unit 1 records the incoming call voice of the other party as an analog voice signal, and 曰% the parameter 2 (step S31). A/n charm ratio than "communication" is transmitted to the A/D conversion bit voice signal (step negotiation) change theft 2 to convert the analog voice signal into the number two = detection principle will digital voice signal midrange signal t get 4; _cutting, so as to extract the characteristic parameters of the unit 3 according to the frequency size characteristics of the voiced voice signal from the digital language (step S34), and picking up the voice in the voice signal from the voice signal, using the terminal preparation principle Cutting digital emotion classification is not = acoustic speech data as depicted in Figure 2. Hundreds of hearts. The diminishing corresponding to the emotional singer" Jia Jia 5 pairs of reading emotional characteristics of the data ^). The classification of statistical classification of emotional classification (steps of the emotional characteristics of the same characteristics of the principle of reading the same use of the data. For example, Feature selection list ^ 3撷月声特^Bei material for classification

The emotion classifier 5 calculates the MFCC value for the adjacent distance, and takes an L from the shortest emotional data to define the voice. The value of the sentiment is from the injury degree (1) = 4, angry households (when the characteristics, if the repair 5' grief Ungry)=2, happiness (201008222 • =1, neutrality =1 and bored =0, the emotion classifier 5 determines that the emotion category is “sadness”. The output unit 6 generates an emotional analysis report of the incoming caller according to the classified statistical data generated by the emotion classifier 5. The sentiment analysis report describes the emotional state of the other party's call, including anger, tiredness, happiness, and normality. And the degree of sadness (step S37). Finally, the emotion output unit 6 outputs and displays the sentiment analysis report on the display screen 7 of the mobile phone 8 for the user to know the emotional state of the other party's call (step S38). The present invention is not limited to the scope of the present invention, and may be modified and retouched, and thus the protection of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a structural diagram of a preferred embodiment of a mobile phone for implementing call emotion recognition according to the present invention. FIG. 2 is a feature capture of FIG. The unit uses the endpoint detection principle to cut the schematic diagram of the voiced voice and the voiceless voice. Fig. 3 is a flow chart of a preferred embodiment of the method for realizing the emotional recognition of the incoming call of the mobile phone according to the present invention. [Description of main component symbols] Voice recording unit 丄 A/D conversion Device 2 feature extraction unit 3 memory, 201008222 emotion classifier emotion output unit display screen mobile phone

Claims (1)

201008222 X. Applying for Patent Fan Park 1 · A mobile phone that realizes the emotional recognition of incoming calls. The mobile phone includes: a voice recording unit for recording the incoming call voice as an analog voice A/D conversion n, and changing the analog voice signal to digital Zhayuan, used to cut the digital voice by the end of the grammar: the number of the 艏Γ rape (4) and the silent voice (four), the silk according to the voice signal, 'small from the sound s voice signal to draw different characteristic parameters; = Classification n 'Poetry (4) Different characteristic parameters read the emotional characteristics of the voiced speech signal, and classify the read emotional feature data to generate the classified statistical data of the emotional characteristics; and the unit 'used according to the emotion The classified statistics generated by the classifier generate an sentiment analysis report of the incoming caller. ^ As claimed, the hand of the silk electric emotion identification mentioned in Item 1 of the patent (4): the data in the memory is stored in the memory of the mobile phone, or stored in the network database of the operator. Machine, special: the hand of the range of the first to achieve the emotional recognition of the caller's emotional classifier according to the similar data with similar characteristics of the principle, s emotional characteristics of the data for classification and statistics. = = as claimed in the patent _ the first item of the real county electric sentiment identification hand out and display unit is also used to the emotional analysis report, the shell is not on the display screen of the mobile phone. [·If the application (4) _ 4 items are 敎 丝 , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , 12 201008222 6 · A method for emotional recognition of a mobile phone call, the method comprising the steps of: recording an incoming call voice as an analog voice signal; converting the analog voice signal into a digital voice signal; and using the endpoint detection principle to digitize the voice signal The voiced voice data and the voiced voice data are cut out; the different feature parameters are extracted from the voiced voice signal according to the frequency of the voiced voice signal; and the emotional features corresponding to the voiced voice signal are read according to different feature parameters. The classified emotional feature data is classified and statisticed to generate classified statistical data of the emotional characteristics; and the sentiment analysis report of the incoming caller is generated according to the classified statistical data. 7. The method for identifying the emotional characteristics of a mobile phone as described in claim 6 of the patent application, wherein the emotional feature data is stored in a memory of the mobile phone or stored in a network database of the mobile phone operator. 8 · As claimed in claim 6 of the patent application scope, the endpoint detection principle cuts out the voice data and the silent voice in the voice signal according to the energy or the zero rate in the voice signal. data. 9. The method as claimed in claim 6, wherein the method further comprises the step of: outputting and displaying the sentiment analysis report on a display screen of the mobile phone. 10. The method of emotional recognition of a mobile phone call as described in claim 9 of the patent scope, wherein the sentiment analysis report describes the emotional state of the other party's call, including anger, tiredness, happiness, normality and sadness. 13