Efficient coding of LSP parameters using Compressed Sensing on Approximate KLT Domain

An efficient LSP parameters quantization scheme is proposed using the compressed sensing (CS). The LSP parameters extracted from consecutive speech frames are compressed by CS on the approximate KLT domain to produce a measurement vector, which is quantized using the split vector quantizer. Then, from the quantized measurements, the original LSP parameters are reconstructed by the orthogonal matching pursuit method. Experiments show that the scheme can obtain "transparent quality" at 5 bits/frame with drastic bits reduction compared to other methods.

---

Zaproponowano kwantyzację parametru LSP (Linear prediction coefficient) przy użyciu metody compressed sensing CS. Oryginalna wartośc LSP może być zrekonstruowana przy zastosowaniu metody ortogonalnego dopasowania. Uzyskano dobrą jakość ramki 5 bitów/ramka ze znacząca redukcją bitów w porównaniu z innymi metodami.