SPR Classic TM 600 bps vocoder is based on Sinusoidal Pulsed Representation (SPR) model, where excitation signal includes mixture of pitch-harmonics, noisy and aperiodic impulses.
It allows to synthesize correctly any speech sounds, including mixed voiced+unvoiced and complex transients.
Additionally, choice of optimal quantization schemes and high algorithmic optimization provided superior characteristics of this vocoder in comparison with others for the same bit rate
Features:
We used ITU-T P.50 multilanguage speech base and ITU-T P.862 utility to estimate speech quality.
SPR 600 bps vocoder doesn't concede and even exceeds MELPe 600 bps standard vocoder for majority from twenty languages. Average PESQ is: 2.6 vs 2.55.
At the same time, all other characteristics of the SPR 600 vocoder excel considerably appropriate characteristics of the standard vocoder.
SPEECH QUALITY
We passed encoded bit streams through AWGN channel simulator with various SNR and estimated quality of decoded speech.
SPR 600 shows strong superiority over MELPe 600.
ROBUSTNESS
Time delay is very important characteristic of communication system. Algorithmic delay of vocoder does play determinative role in total delay. Well known, the delay more than 100 ms is not acceptable for normal conversation.
SPR 600 provides 100 ms delay, MELPe - 110 ms.
Frame size is 80 ms (640 speech samples).
ALGORITHMIC DELAY
SPR 600 requires three times lower computing resources and shows strong superiority over MELPe 600.
Only 14 MIPS versus 44 MIPS for TI's C55xx.
COMPUTING COMPLEXITY
We have estimated and compared resources of SPR 600 bps vocoder for TI's C55xx DSP.
SPR 600 shows strong superiority over MELPe 600.
MEMORY USAGE
SPR 600 bps is available right now for TI's C55xx DSP and for PC (Win32) platforms and also is available for any other DSP, RISC or general-purpose platform during two-three months.
AVAILABILITY
REQUEST
REQUEST
