scispace - formally typeset
Search or ask a question
Topic

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.


Papers
More filters
Proceedings ArticleDOI
02 Dec 1990
TL;DR: In this paper, error protection strategies for a high-quality code-excited linear predictive (CELP) vocoder to be used for mobile satellite communications were examined. But, the proposed coding technique can provide good performance under random error conditions as high as 10/sup -2/ BER.
Abstract: The authors examine the error protection strategies for a high-quality code-excited linear predictive (CELP) vocoder to be used for mobile satellite communications. The speech coder has a data rate of 4.8 kb/s and an overall rate of 6.4 kb/s, including error protection coding. The error protection strategies involve selective error correction and error detection along with speech parameter smoothing. Candidate coding strategies are evaluated by examining their performance against burst errors common to a mobile channel and against random errors with fairly high bit error rates (BERs). It is found that a hybrid parity check coding scheme performs best under these channel conditions. The error protection strategies have been found to work well under burst error conditions designed to simulate a land mobile environment. The proposed coding technique can provide good performance under random error conditions as high as 10/sup -2/ BER. >

2 citations

01 Jan 2007
TL;DR: A variety of structured quantizers which strike different balances between complexity and performance are presented and a study of the performance benefits of speaker-dependent coding in the CELP framework is undertaken.
Abstract: Modern coding applications, such as wideband speech, are characterized by sources with large dimensions and unknown statistics, complicated distortion measures, and the need for high-quality quantization. However, the complexity of quantization systems must be kept in check as the dimension grows, requiring flexible quantization structures. These structures, in turn, require an automatic training method that can infer statistics from example data and balance the various factors to optimize performance. The development of efficient, flexible quantization structures also opens up new coding applications, such as speaker-dependent coding. This approach promises improved performance but presents a variety of implementational challenges. The first part of this dissertation presents a variety of structured quantizers which strike different balances between complexity and performance. This includes the scalar transform coder, which is augmented with a flexible companding scalar quantizer based on Gaussian Mixtures. Next, a variety of extensions to the Gaussian Mixture Vector Quantizer (GMVQ) system for recursive coding are examined. Training techniques for these systems are developed based on High-Rate quantization theory, which provides a tractable objective function for use in automatic design. This replaces ad-hoc methods used for design of structured quantizers with a data-driven approach which is able to incorporate various distortion measures and structures. The performance of the systems is demonstrated on the problem of wideband speech spectrum coding. The second part of this dissertation considers speaker-dependent wideband speech coding. Using the GMVQ system and training approach developed in the first portion, a study of the performance benefits of speaker-dependent coding in the CELP framework is undertaken. The three main types of CELP parameters (spectrum, adaptive codebook and fixed codebook) are all investigated, and the gains quantified. Next, a number of implementational issues related to speaker-dependent coding are addressed. A safety-net approach is utilized to provide robustness, and its implementation in the context of GMVQ is explored. A variety of online training architectures are presented which strike different balances between training complexity, communications overhead and performance. As components of these architectures, techniques for training on quantized data and recursive learning are examined.

2 citations

Proceedings ArticleDOI
26 Nov 1996
TL;DR: This study proposes a novel approach based on a Bayesian network and the LSP frequencies to generate syllable prosody and the coarticulation between two concatenated syllables respectively.
Abstract: This study proposes a novel approach based on a Bayesian network and the LSP frequencies to generate syllable prosody and the coarticulation between two concatenated syllables respectively. The Bayesian network is employed to model the relation between the prosodic information and the linguistic features. Given a Chinese character sequence, the Bayesian network can provide appropriate prosodic information, including pitch contour, syllable intensity, syllable duration and pause duration. Furthermore, the coarticulation is generated by adjusting the LSP frequencies in a CELP-based synthesizer. The synthesized speech is tested on twenty subjects. The test results indicate that the average correct rate is 95.8% for intelligibility, and the mean opinion score (MOS) is 3.2 for naturalness.

2 citations

Proceedings ArticleDOI
26 Sep 2013
TL;DR: Experimental results show that performance of the SE-GC method is better compared to COMB-ESM method under speech coding, which is the one of the major degradation in mobile environment.
Abstract: Vowel onset and vowel offset points are the instants at which the onset and offset of vowel take place in the speech signal. Vowel regions start with the vowel onset point and end with the vowel offset point. This paper discusses the effect of speech coding on detection of vowel offset point. Speech coding is the one of the major degradation in mobile environment. In this work, effect of speech coding is studied by using two recently developed methods, COMB-ESM and SE-GC for vowel offset point detection. COMB-ESM method uses the combination of evidences from excitation source, spectral peaks energy and modulation spectrum. SE-GC method uses spectral energy within glottal closure region for detecting the vowel offset points. Speech coders used in this study are GSM full rate (ETSI 06.10), CELP (FS-1016), and MELP (TI 2.4 kbps). Performance of vowel offset point detection methods is evaluated using TIMIT database and consonant-vowel (CV) units collected from the broadcast news corpus. Experimental results show that performance of the SE-GC method is better compared to COMB-ESM method under speech coding.

2 citations

Journal ArticleDOI
TL;DR: This paper discusses two speech-coding algorithms at around 8 kbit/s that both use pre-selection in the codebook search and training for the conjugate structured random codebook and meets all the requirements of the ITU-T (formerly CCITT) 8 k bit/s speech coding standard.
Abstract: This paper discusses two speech-coding algorithms at around 8 kbit/s. These algorithms both use pre-selection in the codebook search and training for the conjugate structured random codebook. One has short-delay time that can dispense with echo-control equipment. This algorithm is robust against channel errors by using LPC analysis separated from pitch excitation, although LPC parameters are derived from the locally decoded signal. The other algorithm has medium-delay time and meets all the requirements of the ITU-T (formerly CCITT) 8 kbit/s speech coding standard. The performance of these codecs is evaluated using signal to noise ratio (SNR), pair comparison tests with a modulated noise reference unit (MNRU), and Mean Opinion Score (MOS). The major applications of these codecs are personal handy phone systems and FPLMTS.

2 citations


Network Information
Related Topics (5)
Decoding methods
65.7K papers, 900K citations
83% related
Data compression
43.6K papers, 756.5K citations
83% related
Signal processing
73.4K papers, 983.5K citations
83% related
Feature vector
48.8K papers, 954.4K citations
80% related
Feature extraction
111.8K papers, 2.1M citations
79% related
Performance
Metrics
No. of papers in the topic in previous years
YearPapers
20226
20213
20207
201915
201810
201713