Linear Prediction Speech Coder Research Articles

Single and multiple frame coding of LSF parameters using deep neural network and pyramid vector quantizer

In linear predictive speech coders, the linear predictive coding (LPC) parameters are usually transformed into the line spectral frequency (LSF) representation for quantization. In this paper, the single and multiple frame coding of LSF parameters using deep neural network (DNN) and pyramid vector quanrizer (PVQ) are proposed. In the single-frame scheme, a non-linear DNN predictor which demonstrates much better prediction performance than the autoregressive (AR) model is applied to exploit the inter-frame dependency of LSF parameters. The prediction residual signal has Laplacian distribution and can be efficiently quantized by the PVQ. The performance evaluation using spectral distortion shows that the proposed DNN predictive PVQ outperforms the AR predictive split vector quantization (SVQ). In the multiple-frame scheme, a deep autoencoder possessing linear coder-layer units with Gaussian noise is used to compress and de-correlate multiple LSF frames. The deep autoencoder shows a high degree of modelling flexibility for multiple LSF frames. To quantize the coder-layer vector effectively, a PVQ is considered. The experimental results show that the proposed multi-frame scheme with determined optimal coder-layer dimension outperforms the discrete cosine model (DCM)-based approach in terms of spectral distortion performance and robustness across different speech segments.

Pitch and Fourier magnitude based steganography for hiding 2.4 kbps MELP bitstream

In this study, the authors present a new steganographic technique called random least significant bits of pitch and Fourier magnitude steganography (RLPFS). It is based on hiding a secret speech coded by mixed excitation linear prediction (MELP) speech coder in speech bitstream (cover signal), which is also encoded by MELP coder. First, the RLPFS leaks the hidden speech in the following modes: pitch-based steganography, Fourier magnitude-based steganography or both. These modes are selected randomly. Second, during transmission, the stego speech, the mode number, and the number of embedded bits would be transmitted either through a covert channel created in the transmission protocol or through the cover speech. In this work, the authors have dealt with the challenge of embedding a secret speech into a cover speech coded by a very low bit rate speech coder while maintaining a reasonable level of speech quality. They have shown that RLPFS was able to create hidden channels with maximum steganographic bandwidths up to 266.64 bit/s at the cost of a steganographic noise between 0.031 and 0.62 mean opinion score. Also, this study takes into account the security of the parameters, the synchronisation of the receiver to deal with a packet loss during transmission and the resistance of the proposed method against steganalysis.

Switched orthogonalization of fixed-codebook search in code-excited linear-predictive speech coder: Derivation of conditions for switching

Switched orthogonalization of fixed-codebook search in code-excited linear-predictive speech coder: Derivation of conditions for switching

Algebraic codebook search strategy for an algebraic code‐excited linear‐prediction speech coder by means of reduced candidate mechanism and iteration‐free pulse replacement

This work aims to present a combined version of reduced candidate mechanism (RCM) and iteration-free pulse replacement (IFPR) as a novel and efficient way to enhance the performance of algebraic codebook search in an algebraic code-excited linear-prediction speech coder. As the first step, individual pulse contribution in each track is given by RCM, and the value of N is then specified. Subsequently, the replacement of a pulse is performed through the search over the sorted top N pulses by IFPR, and those of 2–4 pulses are carried out by a standard IFPR. Implemented on a G.729A speech codec, this proposal requires as few as 20 searches, a search load tantamount to 6.25% of G.729A, 31.25% of the global pulse replacement method (iteration = 2), 41.67% of IFPR, but still provides a comparable speech quality in any case. The aim of significant search performance improvement is hence achieved in this work.

An efficient algebraic codebook search for ACELP speech coder

Abstract In a bid to enhance the search performance, this paper presents an improved version of reduced candidate mechanism (RCM), an algebraic codebook search conducted on an algebraic code-excited linear prediction (ACELP) speech coder. This improvement is made based on two findings in a piece of our prior work. The first finding is that a pulse with a high contribution in the associated track is more likely to serve as an optimal pulse in the optimal codevector and the second is that the speech quality can be well maintained at a search accuracy above 50% approximately. Subsequently, a new finding in this study concerning a structured algebraic codebook in G.729 indicates that there is a 0.8321 probability that the number 1 ranked pulse in a global sorting by pulse contribution is indeed one of the optimal pulses. Hence, the number 1 pulse in the global sorting is labeled as one of the optimal pulses, following which a sequence of search tasks are fulfilled through RCM. This proposed complexity reduction algorithm, implemented on a G.729A speech codec, takes as few as eight searches, a search load tantamount to 2.5% of G.729A, 12.5% of global pulse replacement method (iteration = 2), 16.7% of iteration-free pulse replacement method, and 50% of RCM (N = 2). This proposal is thus found to successfully reduce the required computational complexity to a great extent as intended.

An Efficient Search Strategy for ACELP Algebraic Codebook by Means of Reduced Candidate Mechanism and Iteration-Free Pulse Replacement

This work aims to present a combined version of reduced candidate mechanism (RCM) and iteration-free pulse replacement (IFPR) as a novel and efficient way to enhance the performance of algebraic codebook search in an algebraic code-excited linear-prediction (ACELP) speech coder. As the first step, individual pulse contribution in each track is given by RCM, and the number N of candidate pulses is then specified. Subsequently, the replacement of a pulse is performed through the search over the sorted top N pulses by IFPR, and those of 2 to 4 pulses are carried out by a standard IFPR. Implemented on a G.729A speech codec, this proposal requires as few as 24 searches, a search load tantamount to 7.5% of G.729A, 37.5% of the global pulse replacement method (iteration=2), 50% of IFPR, but still provides a comparable speech quality in any case. The aim of significant search performance improvement is hence achieved in this work. DOI: http://dx.doi.org/10.5755/j01.itc.43.2.4946

Bitrate Scalability for Multi-Pulse Based Code Excited Linear Prediction Speech Coder

Problem statement: The bitrate scalability is an important functionality that is defined in the Moving Picture Expert Group (MPEG-4) requirements. Since the traffic in a communication network varies with time, the capability of varying the bitrate of a speech coder is needed. Approach: This study proposes an approach to vary the bitrate by using a bitrate scalable tool which is attached with the core section of Multi-Pulse based Code Excited Linear Predictive (MP-CELP) coder. The bitrates scalable tool employs multi-stage excitation coding based on an embedded-coding approach. The multi-pulse excitation codebook at each stage is adaptively produced depending on the selected excitation signal at the previous stage. Results: The experimental results show that the speech quality of the proposed coder with the bitrate scalable tools is improved above the speech quality of the conventional coder. Conclusion: From the study, the proposed approach gives a tool to improve the speech compression and can also support a variety of coder bitrates.

Analysis of Fundamental Frequency Contour of Coded Speech Based on Multi-Pulse Based Code Excited Linear Prediction Algorithm

Problem statement: In low-bit-rate speech communication, speech coding deteriorates the characteristics of the coded speech significantly. An important feature of the speech is the fundamental frequency contour which determines the pitch information of the speech. It has been known that pitch information is one of the core parameter of the multi-pulse based code excited linear prediction (MPCELP) speech coder. Therefore the study of the deteriorated fundamental frequency contour should be conducted properly. Approach: This study proposes an analysis of the fundamental frequency contour of the coded speech based on MP-CELP speech coder. The comparison of the fundamental frequency contour of the natural speech and that of the coded speech has been performed. The MP-CELP with three levels of bitrate scalability is selected as the core speech coder. The speech material includes a hundred of male speech utterances and a hundred of female speech utterances. Results: The experimental results show that the speech coder causes the deterioration of the fundamental frequency contour empirically. The Root Mean Square Error (RMSE) between the fundamental frequency contour of the natural speech and that of the coded speech for three different bitrates has been conducted. The lower bitrate causes the higher value of RMSE. Conclusion: From the study, it is a proved that the MP-CELP speech coder deteriorates the fundamental frequency contour of the transmitted speech.

Performance Evaluation of Multi-Pulse Based Code Excited Linear Predictive Speech Coder with Bitrate Scalable Tool over Additive White Gaussian Noise and Rayleigh Fading Channels

Performance Evaluation of Multi-Pulse Based Code Excited Linear Predictive Speech Coder with Bitrate Scalable Tool over Additive White Gaussian Noise and Rayleigh Fading Channels

Performance Evaluation of Multi-Pulse Based Code Excited Linear Predictive Speech Coder with Bitrate Scalable Tool over Additive White Gaussian Noise and Rayleigh Fading Channels

Problem statement: In the modern speech communication technology, the speech coding with bitrate scalability was needed. However, various types of noises in the communication channels cause damages in the transmitted information especially speech data. Tonal-language speech was also affected by this situation. Approach: Based on the high pitch delay resolution technique, the MPCELP speech coding was proposed over the environment with CDMA AWGN and Rayleigh fading channels for tonal language. The proposed coder supports multiple bitrates and also has the functionality of bitrate scalability. Results: Through performance analysis and computer simulation, the quality of the proposed coding was presented with an improvement from conventional scalable MP-CELP in the specific-noise environments. The HPDR technique was applied to the MP-CELP to use for tonal language, meanwhile it can support the core coding rate of 5.6, 8.2, 12.2 kbps and additional scaled bitrates. Conclusion: By applying the high pitch delay resolution technique with the MP-CELP speech coding, we can improve the quality of tonal encoded speech. Moreover, the coding quality of the proposed coder was better than the conventional coder for Thai language over both AWGN channel and Rayleigh fading channel.

Multi-Pulse Based Code Excited Linear Predictive Speech Coder with Fine Granularity Scalability for Tonal Language

Problem statement: The flexible bit-rate speech coder plays an important role in the modern speech communication. The MP-CELP speech coder which is a candidate of the MPEG4 natural speech coder supports a flexible and wide bit-rate range. However, a fine scalability had not been included. To support finer scalability of the coding rate, it had been studied in this study. Approach: In this study, based on the MP-CELP speech coding with HPDR technique, Fine Granularity Scalability was introduced by adjusting the amount of transmitted fixed excitation information. The FGS feature aim at changing the bit rate of the conventional coding more finely and more smoothly. Results: Through performance analysis and computer simulation, the quality of scalability of the MP-CELP coding was presented with an improvement from conventional scalable MP-CELP. The HPDR technique is also applied to the MP-CELP to use for tonal language, meanwhile it can support the core coding rate of 4.2, 5.5, 7.5 kbps and additional scaled bit rates. Conclusion: The core coder with high pitch delay resolution technique and adaptive codebook for tonal speech quality improvement has been conducted and the FGS brings about further efficient scalability.

Early linear predictive speech coding at Bell Telephone Laboratories and Electrical Communication Laboratory

At the 6th ICA in Tokyo (1968), two papers on the linear predictive coding of speech were presented by Manfred Schroeder and by me in the same session. The first paper was on the adaptive linear predictive waveform coding for medium band coding (time domain modeling) and the latter was on the LPC vocoder based on the autoregressive spectral estimation for narrow‐band coding (frequency domain modeling). Although their approaches were very different in both concept and scope, there were dramatic common points: the covariance and autocorrelation methods were used to solve normal equation to determine the LPC parameters and also the pitch period was extracted to reproduce speech periodicity in both systems. Subsequently the multipulse (1982), stochastic (1984), and code (1985) excitation LP were proposed at AT&T BTL, whereas the transformation of LPC parameters to PARCOR (1969) and LSP (1975) and their efficient quantization were intensively studied at ECL of NTT. These efforts were eventually combined for practical implementations in the secure, cellular, and IP telephone in the 1980s. The brilliant ideas by Manfred Schroeder and his associates at BTL, in particular Bishnu Atal, have always been a helpful lighthouse to navigate the world‐wide speech coding research in the right direction since 1980s.

Optimized trellis coded vector quantization of LSF parameters, application to the 4.8 kbps FS1016 speech coder

Speech coders operating at low bit rates necessitate efficient encoding of the linear predictive coding (LPC) coefficients. Line spectral frequencies (LSF) parameters are currently one of the most efficient choices of transmission parameters for the LPC coefficients. In this paper, an optimized trellis coded vector quantization (TCVQ) scheme for encoding the LSF parameters is developed. When the selection of a proper distortion measure is the most important issue in the design and operation of the encoder, an appropriate weighted distance measure has been used during the TCVQ construction process. Using this distance, we will show that the LSF TCVQ encoder performs better than the encoder conceived with the unweighted distance and a reduction of about 1–2 bits/frame is obtained while maintaining the same performance. We further applied the TCVQ encoder system for encoding the LSF parameters of the US federal standard (FS1016) 4.8 kbps code excited linear prediction (CELP) speech coder. At lower bit rates, our objective and subjective evaluation results show that the incorporated LSF TCVQ encoder performs better than the 34 bits/frame LSF scalar quantizer used originally in the FS1016 coder. The subjective tests reveal also that the 27 bits/frame scheme produces equivalent perceptual quality to that when the LSF parameters are unquantized.

An intelligibility enhancement for the mixed excitation linear prediction speech coder

We recently proposed a technique to increase the robustness of perceptually important acoustic cues in speech, based on modification of the phoneme timing structure. We now apply the technique as a preprocessor to improve the intelligibility of the mixed excitation linear prediction (MELP) coder. The enhancement strategy, however, requires adaptations to minimize the introduction of unnatural speech characteristics that are introduced and intensified by parametric coding schemes. The enhanced 2.4 kb/s fixed-point MELP coder achieves a level of speech intelligibility comparable to the 8 kb/s G.729 Annex A coder.

Robust transmission of melp-compressed speech: an illustrative example of joint source-channel decoding

A mixed-excitation linear predictive (MELP) speech coder was selected as the US federal standard for 2400 b/s speech compression. This paper examines the quality of MELP-compressed speech when transmitted over noisy communication channels in conjunction with a variety of error-control schemes. The focus is on channel decoders that exploit the inherent in the MELP bitstream. This residual redundancy, which is manifested by the correlation in time and the nonuniform distribution of various MELP parameters, can be quantified by modeling the parameters as one-step Markov chains and computing the entropy rate of the Markov chains based on the relative frequencies of transitions. Moreover, this residual redundancy can be exploited by an appropriately tuned channel decoder to provide substantial coding gain when compared with decoders that do not exploit it. Channel coding schemes include conventional binary convolutional codes and iteratively-decoded parallel concatenated convolutional (turbo) codes.

A joint source-channel speech coder using hybrid digital-analog (HDA) modulation

A joint source-channel coding system for transmitting speech on a bandlimited additive white Gaussian noise (AWGN) channel is presented. The proposed method uses a hybrid of digital and analog modulation techniques. The digital part of the system consists of a Federal Standard 1016 code-excited linear predictive (FS 1016 CELP) speech coder followed by a rate-3/5 parallel concatenated (turbo) error correcting code. The analog part, which transmits the quantization error due to the FS 1016 CELP coder, consists of a linear encoder and decoder. The advantage of the proposed system is that it achieves excellent rate-distortion/capacity performance that is common in digital systems while maintaining a graceful degradation characteristic that is common in analog systems. Comparisons are made with three purely digital systems and an analog system-with all systems operating at the same overall rate. A formal listening test shows that, at high channel SNR, the improved performance of the proposed hybrid system (versus the purely digital systems) is noticeable to the average listener. Finally, an informal listening test indicates that at low channel SNR (where the error correcting code breaks down) the decoded speech of the hybrid system is still intelligible.

Vector quantization of excitation gains in speech coding

In this paper, we consider vector quantization of excitation gains in code-excited linear predictive (CELP) speech coder using the average error in reconstruction of the excitation signal as the distortion measure and use the same measure to design the codebooks. We have derived a generalized Lloyd's algorithm (GLA) to design a codebook for quantization so that the average of the above criterion over the training vectors is minimized. We have also derived an algorithm, referred to as the Genetic GLA (GGLA), that can be shown to converge to the global optimum of the associated functional with probability one. The performance of ACELP using the codebooks obtained by the proposed algorithms is compared with that of the conjugate-structured ACELP-based ITU-T G.729 coder. Qualitative and quantitative comparisons show that their qualities are comparable.

Mobile audio-visual terminal: system design and subjective testing in DECT and UMTS networks

It is anticipated that there will shortly be a requirement for multimedia terminals that operate via mobile communications systems. This paper presents a functional specification for such a terminal operating at 32 kb/s in a Digital European Cordless Telecommunications (DECT) and Universal Mobile Telecommunications System (UMTS) radio network. A terminal has been built, based on a PC with digital signal processor (DSP) boards for audio and video coding and decoding. Speech coding is by a phonetically driven code-excited linear prediction (CELP) speech coder and video coding by a block-oriented hybrid discrete cosine transform (DCT) coder. Separate channel coding is provided for the audio and video data. The paper describes the techniques used for audio and video coding, channel coding, and synchronization. Methods of subjective testing in a DECT network and in a UMTS network are also described. These consisted of subjective tests of first impressions of the mobile audio-visual terminal (MAVT) quality, interactive tests, and the completion of an exit questionnaire. The test results showed that the quality of the audio was sufficiently good for comprehension and the video was sufficiently good for following and repeating simple mechanical tasks. However, the quality of the MAVT was not good enough for general use where high-quality audio and video was needed, especially when transmission was in a noisy radio environment.

Analysis-by-synthesis linear predictive speech coder with restricted-position multipulse and transformed binary pulse excitation

An analysis-by-synthesis linear predictive speech coder is described. This speech coder has a synthesis part including an adaptive codebook for generating an adaptive excitation, means for generating a multi-pulse excitation and means for generating a transformed binary pulse excitation. The multi-pulse excitation generating means comprises means for generating pulses in restricted pulse positions. The multi-pulse excitation and the transformed binary pulse excitation are combined.

Sequence MAP decoding of trellis codes for Gaussian and Rayleigh channels

This paper considers the use of sequence maximum a posteriori (MAP) decoding of trellis codes. A MAP receiver can exploit any "residual redundancy" that may exist in the channel encoded signal in the form of memory and/or a nonuniform distribution, thereby providing enhanced performance over very noisy channels, relative to maximum likelihood (ML) decoding. The paper begins with a first-order two-state Markov model for the channel encoder input. A variety of different systems with different source parameters, different modulation schemes, and different encoder complexities are simulated. Sequence MAP decoding is shown to substantially improve performance under very noisy channel conditions for systems with low-to-moderate redundancy, with relative gain increasing as the rate increases. As a result, coding schemes with multidimensional constellations are shown to have higher MAP gains than comparable schemes with two-dimensional (2-D) constellations. The second part of the paper considers trellis encoding of the code-excited linear predictive (CELP) speech coder's line spectral parameters (LSPs) with four-dimensional (4-D) QPSK modulation. Two source LSP models are used. One assumes only intraframe correlation of LSPs while the second one models both intraframe and interframe correlation. MAP decoding gains (over ML decoding) as much as 4 dB are achieved. Also, a comparison between the conventionally designed codes and an I-Q QPSK scheme shows that the I-Q scheme achieves better performance even though the first (sampler) LSP model is used.