Code-excited linear prediction

About: Code-excited linear prediction is a research topic. Over the lifetime, 2025 publications have been published within this topic receiving 28633 citations. The topic is also known as: CELP.

PACELP: a high quality and low complexity scheme for narrow band coding of speech

Abstract: The paper proposes a new Pulse Adaptive Code-Excited Linear Predictive (PACELP) speech coding algorithm. The optimal excitation search procedures become simple when the pulse adaptive method is used in matching the codebook. The codebook of PACELP is obtained from Regular-Pulse Excitation sequences using a varying threshold clustering algorithm. The codebook efficiency is 4 times as much as the basic CELP codebook, and its dimensions are reduced from 40 to 10. Finally computer simulation results show that at 4.8 kb/s, PACELP can produce a clear and natural-sounding synthetic speech. >

A differential linear predictive voice coder for 1200 bps

Abstract: The concept of differential linear prediction is presented It is shown that for voiced speech, differential linear prediction results in a theoretical coding reduction of over twelve bits per frame compared to standard linear prediction Intelligibility test scores for a 1200 bit per second differential linear predictive coder demonstrate the potential for this technique even in channel error rates up to five percent

Very low bit rate MELP speech coding algorithms

Abstract: In speech coding algorithms,the Mixed Excitation Linear Prediction (MELP) algorithm can do better to imitate the natural speech characteristic and can synthesize a high quality speech in the low bit rate,so it becomes one of the most potential algorithms in the modern low bit rate speech coding. In the wireless communication,satellite communication,the military and secure communication,the channel bandwidth is a serious problem,so it is necessary to study the compressed (even very low bit rate) speech coding technique. This paper analyses several low bit rate speech coding algorithms based on MELP,summarizes the theories and the key techniques,and finally makes a comparison of the speech qualities.

Hierarchical multi-channel audio coding based on time-domain linear prediction

Abstract: A novel hierarchical multi-channel coding scheme is proposed which exhibits a significant decrease in decoding complexity compared to earlier proposals. The new coding scheme is based on a single channel downmixing process followed by predictions of the multichannel input signals. Symmetries in the prediction filter coefficients and the prediction errors allow for a reduced number of channels which need to be transmitted. A detailed evaluation of the achievable prediction gain and the impact of quantization on the perceived quality leads to insights into the appropriate choice of system parameters. Besides the attractive feature of being usable as a hierarchic extension to existing single channel communication systems and its very low additional algorithmic delay, the transmission quality of the proposed design also scales very well with the available data rate.

Performance evaluation of analysis-by-synthesis homomorphic vocoders

Abstract: Previous research has shown that the homomorphic filtering procedure combined with analysis-by-synthesis excitation coding is a promising alternative to linear predictive coding (LPC) for low bit rate vocoding. In particular, static vector excited and dynamic vector excited homomorphic vocoders have been designed. In this work, the performances of recently developed homomorphic vocoders are evaluated through formal subjective listening tests, using a variation of the paired acceptability rating method. The subjective test results show that the vocoder framework using analysis-by-synthesis excitation analysis is capable of producing good speech quality at 4.8 kb/s or lower. >