Application of single frequency filtering for speech and speaker specific tasks

Abstract

Speech produced in practical environments is affected by several sources of degradations, which include noises, reflections, reverberation and sounds from other sources such as background speech, vehicles, etc. These degradations reduce the performance of several speech-based applications, such as speech and speaker recognition systems. It is difficult to characterize these degradations, as they are not known in advance, and also they are time varying. Hence, it is necessary to explore and extract speech-specific and speaker-specific information in the collected speech signal to improve the performance of speech-based systems. The objective of the studies proposed in this thesis is to explore methods based on single frequency filtering (SFF) analysis of speech signals and artificial neural network (ANN) models for some speech-specific and speaker-specific tasks. The SFF analysis gives flexibility in the representation of speech information at different spectral and temporal resolutions. The ANN models are used for capturing the implicit features, specific to a given application. The SFF analysis can be used to provide a representation of the excitation source and the vocal tract system characteristics of the dynamic speech production mechanism. The ANN models help in extracting the discriminative and descriptive features as needed for a given task. The speech-specific tasks explored in this thesis include speech/nonspeech detection from degraded speech, discrimination of natural and synthetic speech. The speaker-specific tasks examined in this thesis include speaker separation in two speaker conversation data, mimicked voice detection and speaker verification. The speech/nonspeech detection task is addressed using approaches based on signal processing and on ANN models. The signal processing approach exploits speech-specific characteristics in the degraded signal. The results of speech/nonspeech detection are given for the speech data from TIMIT corpus, corrupted by synthetically adding noises of different types