Vector Quantization of LPC Parameters in the Presence of Channel Errors

The authors explore the benefits of time-varying bit allocation to excitation and LPC (linear predictive coding) parameters for the case of codebook-excited LPC. The overall bit rate in the experiment was 4.8, 6.4, or 8.0 kb/s. In each case, permissible bit rates for the LPC component were 0, 24, 36, or 48 bits per frame, one of which was selected for each speech frame using a brute-force search maximum performance. Average SNR gains over conventional time-invariant methods were modest, on the order of 1 to 2 dB, but gains for certain speech segments were as high as 3 to 5 dB. Perceptually, gains due to variable bit allocation were most noticeable in the 6.4 kb/s system, especially with female speakers. However, even in this case, the benefits of flexible bit allocation were somewhat offset by distortions due to other inadequacies in the coding algorithm. >

This paper presents a class of quadratically weighted distortion measures which provide optimal performance for the high rate vector quantization (VQ) of linear predictive coding (LPC) parameters. It is shown that the quantization distortion of a high rate VQ converges to a quadratically weighted measure, where the quadratic weighting matrix is a matrix, which is a generalization of the scalar sensitivity concept to the vector case. The sensitivity matrix is the second order term of the Taylor series expansion of the original distortion measure. Closed form expressions and computationally efficient methods for computing the sensitivity matrices of the different LPC parameterizations are given, which involve no numerical integration and can be implemented in real-time on modern DSP chips. In the general case, the sum of sensitivity weighted scalar errors is not equivalent to the original distortion measure. However the sensitivity matrix of the line spectral pair (LSP) frequencies is exactly diagonal, demonstrating that for LSPs only a sum of sensitivity weighted scalar errors will result in optimal performance.

음성코딩 시 성도는 Linear Predictive Coding (LPC) 계수를 이용해서 모델링 한다. 일반적으로 LPC 계수는 양자화와 선형보간 관점에서 유리한 Line Spectral Frequency (LSF) 파라미터로 변경하여 사용한다. 10차 이상의 다차원 LSF 데이터를 벡터 양자화를 이용하여 직접 코딩하게 되면 벡터 내 상관관계 (intra-frame correlation)를 모두 이용할 수 있으므로 rate-distortion 관점에서는 높은 효율을 기대할 수 있다. 하지만, 계산량과 메모리 요구량이 높아져서 실제 코딩 시스템에서는 사용할 수 없게 되므로, 차원을 나누어 압축하는 Split Vector Quantization (SVQ)이 이용된다. 또한, LSF 데이터는 과거 벡터와의 벡터 간 상관관계 (inter-frame correlation)가 높으므로, 이를 이용한 Predictive Split Vector Quantization (PSVQ)이 사용되고 있다. PSVQ는 SVQ 보다 높은 rate-distortion 성능을 보인다. 본 논문에서는 음성 저장 장치를 위한 최적의 PSVQ를 구현하기 위해서 다수의 과거 프레임 정보와의 벡터 간상관관계 (inter-frame correlation)를 고려한 Multi-Frame AR-model 기반 SVQ (MF-AR-SVQ)를 제안하였다. 기존 PSVQ와 비교해 보았을 때, MF-AR-SVQ는 계산량과 메모리 요구량의 큰 증가 없이, 평균 spectral distortion 관점에서 약 1비트의 성능 향상을 보였다. For speech coding, a vocal tract is modeled using Linear Predictive Coding (LPC) coefficients. The LPC coefficients are typically transformed to Line Spectral Frequency (LSF) parameters which are advantageous for linear interpolation and quantization. If multidimensional LSF data are quantized directly using Vector-Quantization (VQ), high rate-distortion performance can be obtained by fully utilizing intra-frame correlation. In practice, since this direct VQ system cannot be used due to high computational complexity and memory requirement, Split VQ (SVQ) is used where a multidimensional vector is split into multilple sub-vectors for quantization. The LSF parameters also have high inter-frame correlation, and thus Predictive SVQ (PSVQ) is utilized. PSVQ provides better rate-distortion performance than SVQ. In this paper, to implement the optimal predictors in PSVQ for voice storage devices, we propose Multi-Frame AR-model based SVQ (MF-AR-SVQ) that considers the inter-frame correlations with multiple previous frames. Compared with conventional PSVQ, the proposed MF-AR-SVQ provides 1 bit gain in terms of spectral distortion without significant increase in complexity and memory requirement.

A modified regular-pulse excitation (RPE) linear predictive coding (LPC) codec is used to encode audio signals of 15 kHz bandwidth. With appropriate perceptual masking and carefully selected excitation structure, a fixed-pulse excitation (FPE) LPC codec is able to provide high-fidelity audio output. The quantization of both the LPC parameters and the excitation pulse sequence is examined. It is found that the LPC parameters can be quantized efficiently by adaptive differential PCM, while the excitation pulse sequences need more bits to quantize the amplitudes compared to narrowband speech encoding. Informal listening tests show that high-fidelity audio can be achieved at bit rates less than 100 kb/s.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

A class of reverberant speech enhancement techniques involve processing of the linear prediction residual signal following Linear Predictive Coding (LPC). These approaches are based on the assumption that reverberation is mainly confined to the prediction residual and affects the LPC coefficients to a lesser extent. This chapter begins with a study on the effects of reverberation on the LPC parameters where mathematical tools from statistical room acoustics are used in the analysis. Consequently, a general framework for dereverberation using LPC is formulated and several existing methods utilizing this approach are reviewed. Finally, a specific method for processing a reverberant prediction residual is presented in detail. This method uses a combination of spatial averaging and larynx cycle-based temporal averaging. Experiments with a microphone array in a small office demonstrate the dereverberation and noise suppression of the spatiotemporal averaging method, showing up to a 5 dB improvement in segmental SRR and 0.33 in the normalized Bark spectral distortion score.

Speech coders operating at low bit rates necessitate efficient encoding of the linear predictive coding (LPC) coefficients. In this work, we developed an optimized switched split vector quantization (SSVQ) scheme applied for encoding the LPC coefficients represented by the line spectral frequencies (LSF) parameters. It will be shown that the SSVQ provides better performance in terms of bit-rate, spectral distortion and computational complexity than the traditional split vector quantizer. We further applied our SSVQ encoding system, called LSF-SSVQ encoder, to quantize the LSF parameters of the 2.4 Kbits/s normalised speech coder MELP operating over an ideal noiseless channel.

An algorithm for LPC (linear predictive coding) parameter optimization in multipulse (MP)-LPC based speech coders is presented. It is shown that, by taking into account the nature of the MP-excitation signal into LPC parameter computation, it is possible to improve the effectiveness of the LPC model. This results in a better quality of the reconstructed signal in terms both of objective and subjective criteria. The implementation details of the algorithm are discussed and experimental results are presented. In particular a comparison with standard MP-LPC techniques is given. >

A cascade formant model is well suited to describe certain speech segments, such as vowels and vowel-like sounds. The formant model is also useful because the relationship between formants and the vocal tract configurations are well understood; however, this model is not adequate for other speech sounds, such as stops, fricatives, nasals, etc. On the other hand, LPC (linear predictive coding) analysis, of sufficiently high order, can adequately describe the spectrum of any speech sound, but the relationship between the LPC parameters and the spectrum or vocal tract configuration is not obvious. This paper describes a speech analysis/synthesis scheme that uses both formants and LPC parameters for different sections of a speech signal. Thus, in some regions, the benefit of the formant model can be utilized, while in other regions, the LPC representation can be used to obtain a good description of the speech spectrum. The analysis algorithm resolves the problem of discontinuities that arise from using the two different spectral representations. When provided with the same multipulse signal, the method of speech analysis described in this paper produces resynthesized speech of the quality of multipulse LPC.

An efficient stochastically excited linear predictive coding algorithm for high quality low bit rate transmission of speech

A new form of line spectral frequency (LSF), bounded line spectral frequency, is presented. It is shown that the new representation is more efficient than the direct line spectral frequency and the differential line spectral frequency (DLSF). By using a vector measure, the scalar quantisation of tenth-order linear predictive coding (LPC) parameters can be coded at 28 bit/frame with a transparent quantisation quality.

Recent advances in speech coding algorithms and techniques based on the use of linear prediction now permit high quality voice reproduction at remarkably low bit rates. This paper reviews some of the main ideas underlying the algorithms of major interest today. The concept of removing redundancy by linear prediction is reviewed, first in the context of predictive quantization or DPCM. Then linear predictive coding, adaptive predictive coding, and vector quantization are discussed. The concepts of excitation coding via analysis-by-synthesis linear predictive coding is explained and some important enhancements such as vector sum excitations, and adaptive postfiltering are described. Low-delay coding by backward computation of LPC parameters is explained. The concept of phonetic segmentation of speech for closed-loop coding systems is also presented. Linear prediction is the key technique that underlies almost all of the important algorithms for speech coding of interest today. Finally, we discuss some recent work on nonlinear prediction of speech and its potential for the future of speech coding.

A method for generating an artificial conversational speech signal as the input test signal in measuring the characteristics of devices operated by speech signals, such as speech detectors, voice switches, and echo controllers is proposed. A state transition model among talkspurt, pause, double‐talk, and mutual silence in conversational speech signals was introduced to simulate the statistical characteristics of these states in real conversational speech signals. This model assumed that the cumulative distribution of the duration time in each state is exponential. During talkspurt intervals, an artificial voice was generated using a speech synthesizer that controlled the spectrum and source characteristics independently. This artificial signal generation method can generate a signal with given characteristics by changing the values of system parameters, such as the average duration time of each state, transition probability from one state to another, and LPC (linear predictive coding) parameters that expr...

Quantization of data is performed to reduce the bit rate required either for storage of data, or transmission between two or more communicators. Linear prediction (LP) is an efficient way to represent the short term spectrum of the speech signal. The line spectral frequency (LSF) transformation provides benefits over than linear prediction representations such as reflection coefficients, arc sine reflection coefficients, or log area ratios. Vector quantization of the filter parameters allows for a larger reduction in the bit rate to represent a set of parameters over scalar quantization. This reduction comes at the expense of using more computational complexity and greater amounts of storage. To reduce the burden of both storage and computation, various techniques and procedures have been developed. These techniques include full, multistage and split vector quantization, as well as the adaptive, variable dimension and finite state vector quantizers [1-8].

Compared to scalar quantization (SQ), vector quantization (VQ) has memory, space-filling, and shape advantages. If the signal statistics are known, direct vector quantization (DVQ) according to these statistics provides the highest coding efficiency, but requires unmanageable storage requirements if the statistics are time varying. In code-excited linear predictive (CELP) coding, a single compromise codebook is trained in the excitation-domain and the space-filling and shape advantages of VQ are utilized in a nonoptimal, average sense. In this paper, we propose Karhunen-Loe/spl grave/ve transform (KLT)-based adaptive classified VQ (CVQ), where the space-filling advantage can be utilized since the Voronoi-region shape is not affected by the KLT. The memory and shape advantages can be also used, since each codebook is designed based on a narrow class of KLT-domain statistics. We further improve basic KLT-CVQ with companding. The companding utilizes the shape advantage of VQ more efficiently. Our experiments show that KLT-CVQ provides a higher SNR than basic CELP coding, and has a computational complexity similar to DVQ and much lower than CELP. With companding, even single-class KLT-CVQ outperforms CELP, both in terms of SNR and codebook search complexity.

The aim of the study was to inspect the acoustic properties and sleep characteristics of a preapneic snoring sound. The feasibility of forecasting upcoming respiratory events by snoring sound was also investigated. Participants with habitual snoring or a heavy breathing sound during sleep were recruited consecutively. Polysomnography was conducted, and snoring-related breathing sound was recorded simultaneously. Acoustic features and sleep features were extracted from 30-second samples, and a machine learning algorithm was used to establish 2 prediction models. A total of 74 eligible participants were included. Model 1, tested by 5-fold cross-validation, achieved an accuracy of 0.92 and an area under the curve of 0.94 for respiratory event prediction. Model 2, with acoustic features and sleep information tested by Leave-One-Out cross-validation, had an accuracy of 0.78 and an area under the curve of 0.80. Sleep position was found to be the most important among all sleep features contributing to the performance of the 2 models. Preapneic sound presented unique acoustic characteristics, and snoring-related breathing sound could be deployed as a real-time apneic event predictor. The models, combined with sleep information, serve as a promising tool for an early warning system to forecast apneic events. Wang B, Yi X, Gao J, etal. Real-time prediction of upcoming respiratory events via machine learning using snoring sound signal. J Clin Sleep Med. 2021;17(9):1777-1784.