Skip to Main Content
This paper presents a 3.3kbps Characteristic Waveform Interpolation (CWI) speech coding algorithm based on biorthogonal Wavelet Transform (WT). The amplitude spectral surface of characteristic waveforms is decomposed into a series of reduced time resolution surfaces with B-spline biorthogonal wavelet filter banks. This multi-scale characteristic leads to more efficient quantization by exploiting each surface's perceptual importance and inherent transmission rate requirements. At the CWI encoder, the complex alignment operation is removed and the amplitude spectrum is decomposed without phase spectrum. At the decoder, different scale space is reconstructed separately and random or fixed phase is combined to the spectrum based on voiced degree. Compared with traditional CWI algorithm using lowpass filtering CW decomposition, the proposed WT-CWI algorithm can get lower coding rate and improve coding result. Subjective A-B listening tests indicate that the synthesis speech quality of the proposed 3.3kbps WT-CWI coding algorithm is far better than that of traditional 4.25kbit/s CWI speech coder.