Description: Implementation of a speech codec based on coding of speech at 8 kbit/s
using conjugate-structure algebraic-code-excited linear-prediction (CS-ACELP)
- We took .wav files that is sampled at 8000 Hz using 16-bit linear PCM. The encoding process is done every 10ms frame or 80 samples. For the preprocessing stage, the samples are high passed with cut-off frequency of 140 Hz and scaled down by 2. A total of 240 samples are buffer for windowing and autocorrelation computation. The autocorrelation coefficients are used to calculate the LP filter coefficients using the Levinson-Durbin algorithm. The LP filter coefficients are converted to Line Spectral Pair (LSP) coefficients. LSP coefficients are converted back to the LP filter coefficients, which is just the reverse process of the conversion from LP to LSP. This module is exactly what the decoder will need in order to convert the LSP coefficients to LP coefficients. We decided not to implement the LSF quantization module because we did not have the codebook information when we designed our system.
The open-loop pitch delay is calculated first for each frame. Then the closed-loop pitch Platform: |
Size: 40960 |
Author:coco |
Hits:
Search - line spectral pair