A dual-band excitation LSP codec for very low bit rate transmission

Abstract

In multiband excitation (MBE) vocoders, the excitation spectrum is a series of voicedunvoiced (viuv) bands. This allows each speech segment to be partially voiced and partially unvoiced. It has been found that excitation in low frequency region is mostly voiced whereas in high frequency area, it is usually unvoiced. Although dividing the excitation spectrum into several viuv bands (12 bands are commonly used) provides better output speech quality, its implementation for very low bit rate transmission is very difficult, if not impossible. In this paper, a modified version called dualband excitation LSP vocoder is proposed. The excitation spectrum, in this case, contains either one or two viuv bands. Line spectral pairs (LSP) are used to represent the spectral envelope because of its effectiveness. Speech quality at a transmission rate of 2 kbps has been found to be clear and intelligible.

High Rate Data Hiding in Speech Signal

“…Some speech coding algorithms based on MBE, such as INMARSAT-M (Kondoz, 1994) and MELP (McCree et al, 1997), substantially improve quality of the synthetic speech, as compared to non-MBE vocoders in low bit rates. In an MBE coder, the excitation spectrum is taken as a series of voiced/unvoiced (v/uv) bands that are computed and arranged based on the original signal spectrum for each frame of the signal (Chiu and Ching, 1994). This allows each speech segment to be partially voiced and partially unvoiced in the frequency domain.…”

“…This allows each speech segment to be partially voiced and partially unvoiced in the frequency domain. Although there is basically no limits to the number and patterns of v/uv bands, it has been shown in (Chiu and Ching, 1994) that a small number of v/uv bands can adequately reconstruct a near natural and intelligible speech signal. Many other findings in low-rate speech coding confirmed this assertion (see e.g.…”

Very Low Rate Scalable Speech Coding through Classified Embedded Matrix Quantization

“…The original MBE model, however, is inapplicable to speech coding at very low rates, that is, below 4 kbps, due to the large number of frequency bands it employs. On the other hand, dual-band excitation, as the simplest possible MBE model, has attracted lots of attention by the research community [26]. It has been shown that most (more than 70%) of the speech frames can be represented by only two bands [26].…”

“…On the other hand, dual-band excitation, as the simplest possible MBE model, has attracted lots of attention by the research community [26]. It has been shown that most (more than 70%) of the speech frames can be represented by only two bands [26]. Further analysis of the speech spectra revealed that the low frequency band is usually voiced, where the high-frequency band usually contains a noise-like signal (i.e., unvoiced) [26].…”

“…It has been shown that most (more than 70%) of the speech frames can be represented by only two bands [26]. Further analysis of the speech spectra revealed that the low frequency band is usually voiced, where the high-frequency band usually contains a noise-like signal (i.e., unvoiced) [26]. In our coding system, we use the dual-band MBE model proposed in [27], in which the two bands join at a variable frequency determined based on the voicing characteristics of speech signals on a frame-by-frame basis in the LPC model.…”