LPC VOCODER Using Instants of Significant Excitation and Pole Focusing

Download Book (19,150 KB) As a courtesy to our readers the eBook is provided DRM-free. However, please note that Springer uses effective methods and state-of-the art technology to detect, stop, and prosecute illegal sharing to safeguard our authors’ interests.
Download Chapter (470 KB)


Vocoders are designed/used to reduce the bit rate requirement for speech signal transmission without significant degradation in the quality of the resultant speech. In most of the speech coding techniques, the system and the source parameters are coded separately. The system is generally coded using a codebook which demands only less number of bits compared to source coding. Coding of the source parameters requires significant computational complexity and memory to preserve the identity and naturalness of the speaker. In this work, a LPC vocoder is designed which uses the instants of significant excitation estimated from the speech signal to code the source information. Traditional pitch excited LPC vocoders produce intelligible speech at a bit rate of 2400 bps, but they are often synthetic. Hence source is coded by identifying the locations of instants of significant excitation and the corresponding strength at those instants. Thus the bit rate requirement is reduced significantly in the order of 1.6Kbps when a code book of size 1024 is used. The number of bits used for coding the system is further reduced by using the codebook of reduced size, namely 64. But the quality of resynthesized signal decreases as the poles representing the system are defocused due to high averaging. Hence, pole focusing, increasing the radius of the poles is done. This causes the increment in the gain/magnitude of the corresponding frequency component in the spectrum thereby improving the signal quality. Finally, the bit rate requirement is further reduced in the order of 1.3kbps. The performance analysis shows that the resultant synthesized speech is intelligible and quite natural preserving the identity of the speaker.