Decomposition of a speech signal wavelet packet using an entropy criterion
Sonia Moussa, Zied Hajaiej, Ali Garsallah · 2017
The sound signals caterers have long do an analysis and time-frequency representation of sounds, but they still faced the problem of compromise between adaptation to different signal components. Indeed, the audio signals may include components with very different time-frequency characteristics and which each require a different window size. As a result, all these physical properties justify the need for time-frequency representation with adaptive resolution. In this paper, we propose a new time-frequency plane representations for finite energy signals. Firstly, we describe the notion of entropy as a time-frequency representation criterion and the Heisenberg principle. Then we present an extension of the wavelet-packets, and the corresponding time-frequency representations. We called them Σ-dyadic discreet wavelet bases. Connection between Finite Impulse Response filters, Prefect Reconstruction Filter Banks and discreet wavelet bases have been established by many authors. Consequently, Σ-dyadic discreet wavelet bases allow conceiving a new family of non uniform filter banks.