Design and implementation of a web-based software framework for real time intelligent audio coding based on speech/music discrimination

摘要

In this work a software framework based on client-server architecture is implemented for real time intelligent audio coding. A speech/music discrimination scheme analyzes the input audio signal and takes a decision about the nature of the audio signal (speech or music) on a frame by frame basis. According to the decision of the speech/music discriminator, a suitable coder is selected at each frame. The designed software framework makes use of the speech and audio coders incorporated into the MPEG4 audio standard (HVXC or CELP for speech frames and TwinVQ or AAC for music frames) to evaluate the performance of an intelligent multi-mode audio coder. The framework supports several types of audio features (timbral texture features and rhythmic content features) and classifiers (classical Statistical Pattern Recognition (SPR) classifiers, Multilayer Perception Neural Networks (MLPNN), Support Vector Machines (SVM), Fuzzy Expert Systems (FES), Hidden Markov Models (HMM)) Comparison between a speech/music discrimination based-intelligent audio coder and MPEG4-AAC has been performed using audio signals representative of the two corresponding classes (speech and music). Subjective and objective tests have been accomplished aiming at assessing the behaviour of the intelligent audio coding scheme.