Analysis-by-synthesis Research Articles

Analysis-by-synthesis based training target extraction of the DNN for noise masking

In conventional speech enhancement methods, the target of noise mask in the time-frequency domain is based on deep neural networks (DNN), such as ideal ratio mask and phase-sensitive mask, in which they do not consider the dependency of spectrum. In this paper, an ideal real-valued ratio mask (IRVRM) extraction method is proposed based on the analysis-by-synthesis (ABS) for utilizing the dependency of spectrum. In the synthesis process, the enhanced speech is obtained by inverse short-time Fourier transform (ISTFT) of the masked spectrum, whereas in the analysis process, the IRVRM is determined by maximizing speech quality of the reconstructed speech from mask space. The ABS loop algorithm is proposed to reduce computational complexity, namely, the best mask in the specifically generated subspace is conducted in each loop. After the ABS loop, the approximated IRVRM is conducted. This IRVRM is further utilized as the training target of the DNN. The experimental results show that when the extracted IRVRM with the ABS loop is employed as the training target of the DNN, the speech quality is effectively improved in the DNN-based noise masking.

Improving the Visual Quality of Size-Invariant Visual Cryptography for Grayscale Images: An Analysis-by-Synthesis (AbS) Approach.

In visual cryptography (VC) for grayscale image, size reduction leads to bad perceptual quality to the reconstructed secret image. To improve the quality, the current efforts are limited to the design of VC algorithm for binary image, and measuring the quality with metrics that are not directly related to how the human visual system (HVS) perceives halftone images. We propose an analysis-by-synthesis (AbS) framework to integrate the halftoning process and the VC encoding: the secret pixel/block is reconstructed from the shares in the encoder and the error between the reconstructed secret and the original secret images is fed back and compensated concurrently by the error diffusion process. In doing so, the error between the reconstructed secret and original secret is pushed to high frequency band, thus producing visually pleasing reconstructed secret image. This framework is simple and flexible in that it can be combined with many existing size-invariant VC algorithms, including probabilistic VC, random grid VC and vector/block VC. More importantly, it is proved that this AbS framework is as secure as the traditional VC algorithms. Experimental results demonstrate the effectiveness of the proposed AbS framework.

Cepstrum-Based Bandwidth Extension for Super-Wideband Coders

This letter proposes a bandwidth extension (BWE) method using the cepstral envelope coding and duplication of quantized wideband (WB) signals by means of analysis -by-synthesis (AbS) for super-wideband (SWB) coders. In the proposed method, a high frequency band is generated by utilizing the quantized cepstral coefficients extracted from the envelope and the quantized modified discrete cosine transform (MDCT) shape of the wideband signal. The proposed method is compared with the latest G.718 SWB codec and the experimental results show that the proposed method outperforms the baseline codec both in subjective listening tests and objective performance measures.

A physiologically and perceptually motivated voice source model

Many glottal source models have been proposed, but none has been systematically validated perceptually. Our previous work showed that model fitting of the negative peak of the flow derivative is the most important predictor of perceptual similarity to the target voice. In this study, a new voice source model motivated by high-speed laryngeal videoendoscopy is proposed to capture perceptually-important source shape aspects. Six voice source models (the proposed model, two previous models developed at UCLA, as well as the Fujisaki-Ljungkvist, Liljencrants-Fant, and Rosenberg models) were fitted to 40 natural voices obtained by inverse filtering and analysis-by-synthesis (AbS). We generated synthetic copies of the voices using each modeled source pulse, with all other parameters held constant, and then conducted a visual sort-and-rate task in which listeners assessed the extent of perceived match between the original natural voice samples and each copy. Model fitting results showed that the proposed model provides a more accurate fitting to the AbS-derived source than the other models. Perceptual experiments showed that the proposed model provides a close match to the original natural voice. Perceptual studies examining the extent to which each model matches the target tokens will also be reported. [Work supported by NSF grant IIS-1018863 and NIH/NIDCD grant DC01797.]

분석/합성 구조의 저 전송률 변환여기 광대역 음성/오디오 부호화기 설계

본 논문은 음성과 오디오의 혼합 콘텐츠를 대상으로 하는 9.2 kbps 저 전송률 변환여기 부호화기 설계를 목표로 한다. 저 전송률 환경을 위하여 주파수 영역 신호의 중요도에 따른 Band-Selection 방법과 Gain-Shape 양자화를 이용한 Analysis by Synthesis(AbS) 구조를 이용한다. AbS 구조를 이용하는 과정에서 발생하게 되는 연산량을 감소시키기 위하여 밴드별 IDFT를 취하고 합성하는 방법을 이용한다. 전송 되지 않는 구간에 대하여 Comfort Noise를 삽입함으로써 음질 저하 없이 설계한다. 기존의 TCX모드와 비교하여 저 전송률에서 성능의 저하 없는 부호화기를 제안한다. This paper is aimed to design 9.2 kbps low bits late transform excitation coder that target to voice and audio signal. To set up low bit rate, we used Band-selection in frequency domain and gain-shape quantization and AbS structure. To decrease lots of calculation from ABS structure, we used each band IDFT and synthesis. And we designed non-transfer band for performance by inserting comfort noise. We propose coder that has low bit rate and similar performance comparing with original 10.4 kbps AMR-WB+ TCX mode.

Adaptive two-direction pitch tracking algorithm for MBE vocoder

In this paper, an adaptive two-direction pitch tracking algorithm for multiband excitation (MBE) vocoder is proposed. Instead of the multiple constraint equations, a difference inequation is constructed. The adaptive adjusting model of the difference threshold is established by utilizing the information including voicing decisions, pitch estimate, and multiple pitch error identification of the previous frame. To identify the multiple pitch error, an approximate corrective matching error measure from the spectral analysis-bysynthesis (AbS) method is utilized. Because the difference threshold can be adjusted adaptively, the proposed algorithm can efficiently avoid continuous multiple and continuous submultiple pitch errors. Moreover, the proposed algorithm almost needs no additional time complexity and computational complexity, so it can easily be implemented in actual systems. From the experimental results, it is justified that, the proposed algorithm outperforms the conventional algorithm, improving the accuracy of the pitch estimation and displaying excellent robustness to different speakers and white Gaussian noise conditions <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup> .

Speech authentication by semi-fragile speech watermarking utilizing analysis by synthesis and spectral distortion optimization

This paper proposes an improved semi-fragile speech watermarking scheme by quantization of linear prediction (LP) parameters, i.e., the inverse sine (IS) parameters. The spectral distortion due to watermark embedding is controlled to meet the `transparency' criterion in speech coding. A modified bit allocation algorithm combined with watermarking is developed to determine the quantization step so that the `transparency' requirement is satisfied. Due to the statistical nature, the LP coefficients estimated from the watermarked speech signal are different from the watermarked LP coefficients even in the absence of attacks. This effect is the cause of increase in decoding error and minimum authentication length. To tackle this problem, an Analysis by Synthesis (AbS) scheme is developed to reduce the difference between the estimated LP coefficients and the watermarked ones. The watermark detection threshold and minimum authentication length are then derived according to the probability of error and the signal to noise ratio (SNR) requirements. Experimental results show that the proposed AbS based method can effectively reduce the difference between the watermarked IS parameter and the extracted IS parameter when there is no attacks. In addition, the modified bit allocation algorithm can automatically find the appropriate quantization step used in the odd-even modulation so that the transparency requirement is satisfied.

Multichannel Audio Coding Based on Analysis by Synthesis

Spatial hearing enables translation of an auditory scene into a perceived 3-D image by interpreting the acoustic cues related to the sounding objects, their locations, and the physical characteristics of the space. Spatial audio production requires multichannel audio signals in order to convey this information and increase the realism of a real or virtual environment for applications such as the home entertainment, virtual reality, and remote collaboration. As demand to spatial audio continues to expand, efficient coding of multichannel audio content becomes more and more important. This paper provides an overview of some well-known multichannel audio coding techniques and presents a new coding framework for improving the objective fidelity of the decoded signals. A closed-loop encoding system based on analysis-by-synthesis (AbS) principle applied on the MPEG surround (MPS) architecture is described. Comparison results are presented, which show that significant improvements can be achieved with a closed-loop system instead of the conventional open-loop system.

LPC parameters substitution for speech information hiding

Information hiding techniques adopted in secret communication should meet the requirements of high hiding capacity, real-time and high robustness. For the purpose of real-time speech secure communication, an information hiding algorithm based on the analysis-by-synthesis (ABS) speech coding scheme is presented in this article. This algorithm substitutes secret speech data bits for linear predictive coefficients (LPCs) in linear predictive coding (LPC). Statistics show that the change of voiced signal track is slow and the codebook for LPCs vector is large in ABS coding scheme. Therefore, a new concept, filter similarity, is proposed to determine the LPC parameters for substituting with secret speech information, and to generate multi-codebook for saving storage space to store secret speech information. To achieve the best effects of information hiding, a dynamic threshold is built to make an optimal trade-off among hiding-capacity, security, robustness, transparence and real time. Experiments show that the proposed approach meets the requirements of information hiding and secret communication speech quality. The test results indicate that this approach reaches high hiding capacity with an excellent speech quality and complicating speakers' recognition.

Analysis-by-Synthesis Sinusoidal Model without an Overlapping Scheme

A new sinusoidal modeling approach for the analysis-by-synthesis (AbS) of parameters that characterize a linear combination of damped sinusoids is proposed. In addition to the typical sinusoidal parameters, two different damping factors, which represent the time-varying nature of speech, were used to efficiently reduce the modeling error. Even though the proposed model does not employ the overlap-adding synthesis or smoothly interpolative synthesis scheme, it shows substantially better modeling performance in the synthesis of voiced and transient segments.

A novel approach of secure communication based on the technique of speech information hiding

The proposed secure communication approach adopts the proposed algorithm of Analysis-By-Synthesis (ABS) speech information hiding to establish a Secret Speech Subliminal Channel (SSSC) for speech secure communication over PSTN (Public Switched Telephone Network), and employs the algorithm of ABS speech information extracting to recovery the secret information. This approach is more reliable, covert and securable than traditional and chaotic secure communication.

Complete optimization of excitation and model parameters in parametric speech coders

In a speech coding system, synthesis error (the difference between the original speech at the encoder input and reproduced speech at the decoder output) is a more relevant measure of signal distortion than linear prediction (LP) error. By minimizing the synthesis error instead of the linear prediction error, the analysis and synthesis stages become more compatible. In a previous paper (J. Acoust. Soc. Am. 109, 2491), an Analysis-by-Synthesis (AbS) technique for joint optimization of the excitation and filter parameters in parametric speech coders were presented. Using a gradient search in the root domain, synthesis error is minimized by reoptimizing the filter parameters for a given excitation. This paper provides a major improvement over the previous work, resulting in more than 3 dB of gain in the segmental signal-to-noise ratio. The improvement is due to a new algorithm for computing the gradient vector in the gradient search procedure. The paper describes the new algorithm and report on the improved results. By adding an extra minimization step, this technique can be incorporated into the existing LPC, multipulse LPC, and CELP-type speech coding standards.

Formant estimation method using inverse-filter control

This paper proposes a new method for estimating formant frequencies of speech signals, based on inverse-filter control and zero-crossing frequency distributions. In this method, which is called the inverse-filter control (IFC) method, we use 32 basic inverse filters that are mutually controlled by weighted means of zero-crossing frequency distributions. After quick convergence of the inverse filters, we can gain four to six formant frequencies as final mean-values of the zero-crossing frequencies. The proposed method (IFC) has a specific feature that it directly estimates resonant frequencies of a vocal tract, unlike analysis-by-synthesis (A-b-S) or linear predictive coding (LPC) as a spectral matching method. Therefore, spectral shapes influence indirectly alone the formant estimation in the IFC. Although the superiority of IFC to LPC was not necessarily prominent in the systematic evaluation using synthetic speech, the estimates showed satisfactorily small errors for the practical analysis. On the other hand, when observing some analysis examples of real speech, we found many fewer gross errors in IFC than in LPC. Last, we describe in brief a method for estimating a spectral envelope (or formant bandwidths) based on the obtained formant frequencies and the spectrum to be analyzed. According to the results, it is understandable that the existence of the wide-band formants also contributes to stable formant trajectories.

Enhanced waveform interpolative coding at low bit-rate

This paper presents a high quality enhanced waveform interpolative (EWI) speech coder at low bit-rate. The system incorporates novel features such as optimization of the slowly evolving waveform (SEW) for interpolation, analysis-by-synthesis (AbS) vector quantization (VQ) of the SEW dispersion phase, dual-predictive AbS quantization of the SEW, efficient parameterization of the rapidly-evolving waveform (REW) magnitude, and VQ of the REW parameter, a special pitch search for transitions, and switched-predictive analysis-by-synthesis gain VQ. Subjective tests indicate that the 2.8 kb/s EWI coder's quality exceeds that of G.723.1 at 5.3 kb/s, and it is slightly better than that of G.723.1 at 6.3 kb/s.

Revisiting analysis by synthesis

Analysis by synthesis (AbS)—analyzing a speech signal by determining parameters for its synthesis—has been visited by investigators for several decades, but this approach has not resulted in commonly used speech acoustic analysis tools. The most prominent reasons for this are the substantial computational demands of AbS and the availability of fast analysis methods like LPC that are ‘‘good enough’’ in most cases. However, the availability of substantially more powerful, yet inexpensive, computers and the need for analysis methods that work well in all cases prompted this revisit to the use of AbS. To that end, an AbS program was developed that performs a pitch synchronous fit of formant synthesis parameters to acoustic speech signals. This program, called GASP, uses the Levenberg–Marquardt method to adjust synthesis parameters to achieve an accurate approximation to the input speech spectrum for each analysis epoch. At present, GASP fits five formants and one antiresonance along with corresponding bandwidths and two source parameters (amplitude and spectral slope). The talk will illustrate advantages and some disadvantages of the GASP approach, and describe an extension of GASP that incorporates temporal as well as spectral constraints on parameter fitting. [Work supported by Nemours Research Programs and NIH.]

Spectral linear prediction: Properties and applications

Linear prediction (LP) is presented as a spectral modeling technique in which the signal spectrum is modeled by an all-pole spectrum. The method allows for arbitrary spectral shaping in the frequency domain, and for modeling of continuous as well as discrete spectra (such as filter bank spectra). In addition, using the method of selective linear, prediction, all-pole modeling is applied to selected portions of the spectrum, with applications to speech recognition and speech compression. LP is compared with traditional analysis-by-synthesis (AbS) techniques for spectral modeling. It is found that linear prediction offers computational advantages over AbS, as well as better modeling properties if the variations of the signal spectrum from the desired spectral model are large. For relatively smooth spectra and for filter bank spectra, AbS is judged to give better results. Finally, a sub-optimal solution to the problem of all-zero modeling using LP is given.