Voice emotion recognition method based on glottal wave signal feature extraction

Abstract

The invention discloses a voice emotion recognition method based on glottal wave signal feature extraction. According to the method, a spectrogram and TEO are mainly adopted as input of CRNN, a low-level descriptor and a high-level statistical function are combined, emotional speech features are extracted, subjected to dimensionality reduction and subjected to an identification algorithm, and finally output is conducted through an HSF channel. The glottal wave signal feature extraction mainly realizes signal acquisition in a complex cepstrum phase decomposition form, a PCA method is adopted toreduce the dimension of a feature vector, and a BP neural network algorithm is adopted to identify and output the feature vector. According to the method, vocal cord vibration characteristics can bebetter reflected, glottis open-phase and closed-phase information is clear, sound source harmonic components and sound channel interference are effectively reduced, and the recognition accuracy is high.

Description

technical field [0001] The invention belongs to the fields of speech recognition, intelligent processing and human-computer interaction, and in particular relates to a speech emotion recognition method based on glottal wave signal feature extraction. Background technique [0002] Speech is widely used as an important information resource transmission and communication medium. The acoustic signal of speech contains a large amount of user information, semantic information and rich emotional information. The development direction of phonetic tasks mainly includes voiceprint recognition and speech recognition. And emotion recognition, speech emotion recognition aims to identify the correct emotional state of the speaker through speech signals. Since speech is not a complete expression of emotional physiological signals, how to efficiently and accurately recognize the emotions expressed by users under the premise of ignoring other sensory results , is a hot topic in phonetic rese...

AI Technical Summary

Problems solved by technology

The glottal wave signal feature extraction is the key to optimizing speech emotion recognition. The traditional method of extracting glottal wave is based on inverse filtering IF (Inverse Filtering). For example, according to the input signal, the free field outside the lips passes through a specially designed breathing air The flow meter records the gas volume velocity in the oral cavity, and obtains the glottal wave through IF. Although it is not easily affected by low-frequency noise, the design of the flow meter has high requirements for the accuracy of the sensor, which is difficult to meet the requirements of ordinary factory production. The extraction method is the residual signal harmonic sum (SRH algorithm), and its accuracy needs to be further broken through. In addition, the traditional speech signal processing technology uses a wide range of emotion-related feature parameters, resulting in too many dimensions and high redundancy. And it is easy to cause the disaster of dimensionality to affect the recognition effect of speech emotion

Images