A scalable MELP coder based on embedded quantization of line spectral frequencies

摘要

We modify the federal standard mixed-excitation linear prediction (MELP) speech coder by replacing its multistage vector quantizer (MSVQ) for line spectral frequencies (LSF) with a multistage tree-structured vector quantizer (MTVQ), which is embedded in nature. We show that the modified coder becomes bit-rate scalable with gradual change in quality when the bit-rate varies from 1288.9 bps to 2400 bps at step of 44.4 bps. The same approach can be applied to other linear prediction (LP) based speech coder to achieve fine-grain scalability.