Method and device for recognizing human voices in audio

Abstract

The invention discloses a method and device for recognizing human voices in audio. The method comprises the steps that framing processing is carried out on audio data; each frame of the audio data after the framing processing is analyzed by using LPC with an order of P, and audio characteristics are extracted, and comprise a short-time zero-crossing rate, a P-order LPC predication coefficient and the skewness and the kurtosis of an LPC predicted residual magnitude spectrum; a P+3-order characteristic vector is formed according to the audio characteristics; an SVM algorithm is used for training the characteristic vector to obtain a corresponding support vector machine; whether the human voices are included in each frame of the audio data or not is recognized according to the support vector machine. According to the method and device for recognizing the human voices in the audio, the high-precision high-confidence reorganization on the human voices in the audio can be achieved, fundamental services are provided for song content analysis, and therefore the functions of lyric synchronization, song classification, song recommendation and the like are further achieved.

Description

technical field [0001] The invention relates to the field of multimedia information, in particular to the field of audio signal analysis, in particular to a method and device for recognizing human voice in audio. Background technique [0002] With the continuous development of multimedia technology, audio and video information plays an increasingly important role in people's work, style and entertainment. For example, major music websites on the Internet classify songs or recommend songs, so that each user can search for songs as soon as possible or recommend good songs to users. [0003] At present, the song classification and recommendation of major music websites are mostly based on text analysis and user behavior collaborative filtering, and there is no application of in-depth audio content analysis technology. The audio content analysis technology classifies the audio according to the extracted audio features, so that the user can retrieve the desired audio more accura...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to provide a method and device for identifying human voice in audio, which is formed by extracting effective audio features, namely short-term zero-crossing rate, P-order LPC prediction coefficient, and skewness and kurtosis of LPC prediction residual amplitude spectrum. Feature vectors, and use machine learning to identify human voices in audio, solve the problem of low precision and low accuracy in human voice recognition research, and realize high-precision and high-confidence recognition of human voices in audio

Images