Abstract
A variable-bit-rate characteristic waveform interpolation (VBR-CWI) speech coder with about 1.89 kbit/s average bit rate which integrates speech classification based on GMM(Gaussian mixture model) is proposed. Each input frame is classified into one of 4 classes. Different CW representation modes are adopted for different classes. Experimental results showed that the proposed code can eliminate most buzzy and noisy artifacts existing in the fixed-bit-rate 3.8 kbit/s characteristic waveform interpolation (FBR-CWI) speech codec, thus improving the reconstructed speech quality while lowering the average bit rate for about 15%.
| Original language | English |
|---|---|
| Pages (from-to) | 23-27 |
| Number of pages | 5 |
| Journal | Beijing Ligong Daxue Xuebao/Transaction of Beijing Institute of Technology |
| Volume | 27 |
| Issue number | SUPPL. 1 |
| Publication status | Published - May 2007 |
Keywords
- Characteristic waveform interpolation
- Gaussian mixture model
- Source controlled variable bit rate