|
Common description
Low bit rate MELP (Mixed Excitation Linear Prediction)
speech coders with rate 2200/2000/1800 bps use classical parametric model
of description of speech signals. This model is implemented on the base
of linear prediction filter with mixed excitation in several frequency
bands.
Independent estimation of voiced/unvoiced decision in
different frequency bands is providing more natural synthetic speech sounding
then quality of classical LPC vocoders with binary voiced/unvoiced decision
for single band.
All vocoders have fixed-point ANSI-C models for PC. Vocoder
MELP 2200 bps has real-time implementation on the base of DSP TMS320C54x/C55X.
Real-time implementations for other rates will be done on demand. Please
ask for their parameters.
DSP implementations have following features:
- C call interface for top level procedures
- real time work
- multichannel support - independent processing of several duplex channels
on single DSP
- code compatibility within DSP family
- high optimization of program code - all internal signal processing
procedures are implemented on assembler language
Parameters of MELP 2200 bps TMS320C54x implementation:
| |
Coder |
Decoder |
DSP peak load per channel, MIPS |
21.1 |
5.8 |
Program memory, K words |
7.5 |
Data memory, K words |
11.6+2.6*N |
N – number of channels
Parameters of MELP 2200 bps TMS320C55x implementation:
| |
Coder |
Decoder |
DSP peak load per channel, MIPS |
30 |
8 |
Program memory, K bytes |
17.2 |
Data memory, K words |
11.6+2.6*N |
N – number of channels
Here you can test speech quality of MELP vocoders with rate 2200/2000/1800
bps by listening following speech samples:
Original english speech |
|
MELP vocoder 2200 bps |
|
MELP vocoder 2000 bps |
|
MELP vocoder 1800 bps |
|
|
 Common description
Implementation
Speech samples
All technical and commercial questions about products, prices,
terms of delivery, software libraries, source code etc. please send us by e-mail:
|
