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.

A 4 kb/s toll quality harmonic excitation linear predictive speech coder

Abstract: The harmonic excitation linear predictive speech coder (HE-LPC) is a technique derived from MBE and MB-LPC type of speech coding algorithms. The HE-LPC coder has the potential of producing high quality speech at 4.8 kb/s and below. This coder employs a new pitch estimation and voicing technique. In addition, new DCT based LPC and residual amplitude quantization techniques have been developed. The 4 kb/s HE-LPC coder with a 14th order LPC filter was found to produce much better speech quality than the various low rate speech coding standards, including 3.6 kb/s INMARSAT Mini-M AMBE vocoder. During formal ITU ACR test, the 4 kb/s HE-LPC vocoder was found to produced equivalent performance to 32 kb/s ADPCM and G.729 for both flat and modified IRS filtered clean input speech conditions. The HE-LPC algorithm can also be extended to cover bit rates between 1.2 and 8 kb/s range depending on the application.

Speech coding apparatus and method

Abstract: Provided is a speech coding apparatus and method. A band divider divides an input signal into a high-band signal and a low-band signal, a narrowband encoder encodes the low-band signal using a Code Excited Linear Prediction (CELP)-based narrowband speech codec, a frequency characteristic collector converts the high-band signal to a signal in a frequency domain and obtains Modified Discrete Cosine Transform (MDCT) coefficients, a subband determiner determines subbands in a final stage based on the MDCT coefficients and determines subbands for quantization based on the subbands in a final stage, a gain quantizer performs gain quantization of the subbands, a bit assignment unit assigns bits to the subbands according to the magnitude of the gain quantization, and a shape quantizer performs shape quantization of the subbands in an algebraic method. Accordingly, algorithm consistency can be maintained and a complexity can be reduced by extending a bandwidth with a small number of bits in a speech codec.

A robust speech/music discriminator for switched audio coding

Abstract: Switching between speech coding and generic audio coding schemes was recently proven to be very efficient for coding a large range of audio materials at low bit-rates. However, it strongly relies on a robust classification of the input signal. The aim of the paper is to design a reliable speech and music discriminator (SMD) for such an application. Main attention was laid on getting a good tradeoff between accuracy, reactivity and stability of the decision while keeping the delay and complexity reasonably low. To this end, short-term and long-term features are dissociated before being conveyed to two different classifiers. The two classifier outputs are combined in a final decision using a hysteresis. Objective measures show that a more reliable switching decision is achievable. The SMD was successfully implemented in MPEG Unified Speech and Audio Coding (USAC). It allows the codec to show unprecedented audio quality.

Celp voice encoder

Abstract: A CELP type speech coder performs quantization of pitch differential value on pitch information between subframes. The coder limits the number of preliminary selected candidates using threshold processing. The coder includes specialized pitches for a subframe on which quantization of pitch differential value is not applied. When pitch preliminary selection is performed on such a subframe, the coder limits the number of preliminarily selected candidates using threshold processing to avoid outputting, as a preliminarily selected candidate, the above-mentioned specialized pitches. The coder improves the accuracy of the pitch search (adaptive codebook search) while avoiding adverse effects on the quantization of pitch differential value.

High-quality harmonic coding at very low bit rates

Abstract: The paper presents a harmonic vocoder to produce high-quality speech at very low bit rates (below 2 kb/s). Voiced speech is decomposed into forward and backward signals which consist of interpolated harmonics. Unvoiced speech is reconstructed in the time domain with an approach similar to CELP. To remove the "buzzy" quality and avoid the "hoarse" quality, three methods are presented: the randomness of the harmonic phases is controlled according to pitch value and the continuity of synthetic speech is maintained; the spectral envelope determined by the LP model is modified; some noise components can be introduced for voiced synthetic speech. The harmonic vocoder produces quite natural, clear speech. Its perceptual quality is much better than that of the LPC-10 vocoder. >