Speech Coding Research Articles

Log Likelihood Spectral Distance, Entropy Rate Power, and Mutual Information with Applications to Speech Coding

We provide a new derivation of the log likelihood spectral distance measure for signal processing using the logarithm of the ratio of entropy rate powers. Using this interpretation, we show that the log likelihood ratio is equivalent to the difference of two differential entropies, and further that it can be written as the difference of two mutual informations. These latter two expressions allow the analysis of signals via the log likelihood ratio to be extended beyond spectral matching to the study of their statistical quantities of differential entropy and mutual information. Examples from speech coding are presented to illustrate the utility of these new results. These new expressions allow the log likelihood ratio to be of interest in applications beyond those of just spectral matching for speech.

A subspace based progressive coding method for speech compression

In this study, two novel methods, which are based on Karhunen Loeve Transform (KLT) and Independent Component Analysis (ICA), are proposed for coding of speech signals. Instead of immediately dealing with eigenvalue magnitudes, the KLT- and ICA-based methods use eigenvectors of covariance matrices (or independent components for ICA) by geometrically grouping these vectors into fewer numbers of vectors. In this way, a data representation compaction is achieved. Further compression is achieved through discarding autocovariance eigenvectors corresponding to the small eigenvalues and applying vector quantization on the remaining eigenvectors. Additionally, this study proposes an iterative error refinement process, which uses the rest of the available bandwidth in order to transmit an efficient representation of the description error for better SNR. The overall process constitutes a new approach to efficient speech coding, with ICA being used in subspace speech coding for the first time. Constant bit rate (CBR) and variable bit rate (VBR) coding algorithms are employed with the proposed methods. TIMIT speech database is used in the experimental studies. Speech signals are synthesized at 2.4 kbps, 8 kbps, 12.2 kbps, 16 kbps, 16.4kbps and 19.85 kbps rates by using various frame lengths. The qualities of synthesized speech signals are compared to those of available speech codecs, i.e., LPC (2.4 kbps), G.728 (LD-CELP, 16 kbps), G.729A (CS-CELP, 8 kbps), EVS (16.4 kbps), AMR-NB (12.2 kbps) and AMR-WB (19.85 kbps).

Proposed modifications in ITU-T G.729 8 kbps CS-ACELP speech codec and its overall performance analysis

This paper proposes modification in the transmission of excitation codevector and its non-zero pulse sign magnitude using "codebook partition and label assignment" approach, which in turn reduces the number of bits required to transmit it through the communication channel in legacy CS-ACELP 8 kbps speech codec. The proposed approach uses the excitation codebook structure of forward mode standard G.729E 11.8 kbps with two non-zero pulses per track which avoids the use of two algebraic codebook structure for forward mode as well as for backward mode of G.729E with least significant pulse replacement approach for finding optimized excitation codevector. Proposed modification in legacy 8 kbps CS-ACELP (80 bits/10 ms) speech codec actuates the bit rate of 10.6 kbps (106 bits/10 ms) with a better objective and subjective analysis in stark contrast with legacy 8 kbps CS-ACELP speech coder and also avoids the switching of codebook modes of standard 11.8 kbps (G.729E) CS-ACELP speech coder. This paper also aims to propose the reduction in the number of searches in the final codevector of excitation structure by considering initial codevector as a final codevector which improves the quality of the speech compared to the output speech quality of legacy G.729 CS-ACELP working at 8 kbps. Both legacy CS-ACELP 8 kbps speech codec and proposed CS-ACELP 10.6 kbps are implemented in MATLAB. Subjective and objective analysis are carried out on a proposed CS-ACELP 10.6 kbps speech codec in order to evaluate its performance and the results obtained are then cross- compared with the results of legacy CS-ACELP (8 kbps) using set of tables and graphs. It is evident from obtained results that both PESQ and MOS scores are quite comparable for each set of wave files even though bitrates are reduced. Consistency and efficiency of proposed algorithm is assured by calculating the population mean of 95% confidence interval based on obtained objective and subjective parameter results.

高能率符号励振線形予測符号化と帯域選択型変形離散余弦変換符号化を用いた32kbit/sスケーラブル広帯域音声・オーディオ符号化方式の開発

高能率符号励振線形予測符号化と帯域選択型変形離散余弦変換符号化を用いた32kbit/sスケーラブル広帯域音声・オーディオ符号化方式の開発

スピーチ・コードを解読する : 乳児はどのように言語を学習するか

スピーチ・コードを解読する : 乳児はどのように言語を学習するか

Single-Mode-Based Unified Speech and Audio Coding by Extending the Linear Prediction Domain Coding Mode

Unified speech and audio coding (USAC) is one of the latest coding technologies. It is based on a switchable coding structure, and has demonstrated the highest levels of performance for both speech and music contents. In this paper, we propose an extended version of USAC with a single-mode of operation—which does not require a switching system—by extending the linear prediction-coding mode. The main concept of this extension is the adoption of the advantages of frequency-domain coding schemes, such as windowing and transition control. Subjective test results indicate that the proposed scheme covers speech, music, and mixed streams with adequate levels of performance. The obtained quality levels are comparable with those of USAC.

Postprocessing Synthetic Speech With a Complex Cepstrum Vocoder for Spoofing Phase-Based Synthetic Speech Detectors

State-of-the-art speaker verification systems are vulnerable to spoofing attacks. To address the issue, high-performance synthetic speech detectors (SSDs) for existing spoofing methods have been proposed. Phase-based SSDs that exploit the fact that most of the parametric speech coders use minimum-phase filters are particularly successful when synthetic speech is generated with a parametric vocoder. Here, we propose a new attack strategy to spoof phase-based SSDs with the objective of increasing the security of voice verification systems by enabling the development of more generalized SSDs. As opposed to other parametric vocoders, the complex cepstrum approach uses mixed-phase filters, which makes it an ideal candidate for spoofing the phase-based SSDs. We propose using a complex cepstrum vocoder as a postprocessor to existing techniques to spoof the speaker verification system as well as the phase-based SSDs. Once synthetic speech is generated with a speech synthesis or a voice conversion technique, for each synthetic speech frame, a natural frame is selected from a training database using a spectral distance measure. Then, complex cepstrum parameters of the natural frame are used for resynthesizing the synthetic frame. In the proposed method, complex cepstrum-based resynthesis is used as a postprocessor. Hence, it can be used in tandem with any synthetic speech generator. Experimental results showed that the approach is successful at spoofing four phase-based SSDs across nine parametric attack algorithms. Moreover, performance at spoofing the speaker verification system did not substantially degrade compared to the case when no postprocessor is employed.

Cochlear tuning and phase-locking assessed with complementary techniques within the same species, and extrapolation to humans

The cochlea decomposes sound into bands of its constituent frequencies and encodes the temporal waveform of these bands. These two properties generate frequency-tuning and phase-locking in individual neurons of the auditory nerve. The relative roles of frequency tuning and phase-locking toward important aspects of perception, e.g., the coding of speech and pitch, are heavily debated. Characterization of the limits of these processes in humans may help to clarify these relative roles. We studied threshold and suprathreshold frequency tuning, as well as phase-locking to the fine-structure of sound, by recording from single-fibers in the auditory nerve of different species, including chinchilla, cat, and macaque monkey. We also developed stimulus and analysis paradigms to study frequency tuning and phase-locking via mass potentials recorded near the cochlear round window, and applied these techniques to these same species as well as to normal hearing human subjects. The comparison of bandwidths of single fibers with mass potentials within the same species allows us to infer bandwidths of single fibers from human mass potential data. Combined with behavioral and otoacoustic emission data, the evidence suggests that the human auditory nerve is unusual in its sharpness of frequency tuning but not in its range of phase-locking.

‘Transborder conversation’ and oppositional codes: Mediatised diffusion of social movement discourse between Hong Kong and China

Since 2003, Hong Kong has gradually transformed into a ‘social movement society’. With the help of new technologies, mediatised ‘transborder conversation’ about social movement across the border with mainland China, as a verbal practice, has become routine. Sojourn students from the mainland, who usually stay in Hong Kong for several years, actively participate in this verbal practice. Taking a ‘discourse-centred online ethnography’ approach, this study aims to unpack the verbal practice of transborder conversations driven by sojourn students during the Umbrella Movement from the theoretical perspective of speech codes. By conducting participant observation and in-depth interviews with 30 participants, two oppositional codes, cynicism and idealism, have been identified. Each code assumes different relations of the self and the state and different political efficacy, which manifests as different interactional practices. The presence of these oppositional codes has endangered the counter-hegemonic nature of transborder conversations.

Mapping the Speech Code: Cortical Responses Linking the Perception and Production of Vowels.

The acoustic realization of speech is constrained by the physical mechanisms by which it is produced. Yet for speech perception, the degree to which listeners utilize experience derived from speech production has long been debated. In the present study, we examined how sensorimotor adaptation during production may affect perception, and how this relationship may be reflected in early vs. late electrophysiological responses. Participants first performed a baseline speech production task, followed by a vowel categorization task during which EEG responses were recorded. In a subsequent speech production task, half the participants received shifted auditory feedback, leading most to alter their articulations. This was followed by a second, post-training vowel categorization task. We compared changes in vowel production to both behavioral and electrophysiological changes in vowel perception. No differences in phonetic categorization were observed between groups receiving altered or unaltered feedback. However, exploratory analyses revealed correlations between vocal motor behavior and phonetic categorization. EEG analyses revealed correlations between vocal motor behavior and cortical responses in both early and late time windows. These results suggest that participants' recent production behavior influenced subsequent vowel perception. We suggest that the change in perception can be best characterized as a mapping of acoustics onto articulation.

Speech Codes Theory Applied to Problematic Situations in Intercultural Communication A Six-Step Iterative Training Cycle

In this article, we propose a six-step iterative training cycle for intercultural trainers and trainees to understand the cultural codes underlying conflict in intercultural communication. We offer this training cycle, grounded in speech codes theory and ethnographic research methods, as a complementary alternative to traditional prescriptive cultural values approaches. Speech codes theory is a descriptive theory that conceptualizes culture as a system of norms, premises, and symbols about communication that is shared by members of a given speech community. Further, we argue that a corollary of understanding speech codes through our six-step iterative training cycle is the development of emic intercultural communication competence and sensitivity.

Language change and network games

AbstractStudies of language change and variation in sociolinguistics investigate the correlations between social variables and phenomena like vernacular speech norms, code switching, and dialect continua. In multiple studies, researchers claim one variable as (a) particularly decisive and correlative for a number of phenomena and (b) almost universally applicable: the social network structure (Milroy, ). This article summarizes previous work incorporating network theory in questions of language change and discusses a practice noticeably absent from classical sociolinguistics: the simulation of language change. In simulation experiments, sociolinguistic theories of language change – especially those employing social network structure – can be tested in a virtual society free from the hindrance of data sparseness. In this context, the model of game theory can be utilized to construct individuals' interaction for more robust and feasible results.

Echo State Network による音声符号化を用いた音響信号生成

Echo State Network による音声符号化を用いた音響信号生成

音声音響符号化技術の進展

多様な信号処理の全般で重要な役割を果たしている線形予測分析技術の利用の観点で，音声音響符号化技術の進展を紹介する．開発当初から線形予測分析による合成フィルタは音声生成の声道モデルと親和性が高く，音声合成や電話音声に特化した音声符号化で広く使われてきた．一方，現在普及している典型的な音響符号化には線形予測符号化技術が使われていないが，低ビットレート化，音声音響統合符号化の要請に合わせて，スペクトル包絡を効率的に表現する手法として広く使われるようになった．これらの経緯を説明し，あわせて最近実用段階にあるロスレス音響符号化MPEG-4 ALS，携帯電話用の符号化3GPP EVS について紹介する．

An Improved Version of Algebraic Codebook Search Algorithm for an AMR-WB Speech Coder

An Improved Version of Algebraic Codebook Search Algorithm for an AMR-WB Speech Coder

Take-Home Trial Comparing Fast Fourier Transformation-Based and Filter Bank-Based Cochlear Implant Speech Coding Strategies.

Previous studies have demonstrated no improved or deteriorated speech intelligibility with the HiResolution Fidelity 120™ speech coding strategy (HiResF120) over the original HiRes strategy. Improved spectral and deteriorated temporal sensitivities have been shown, making it plausible that the beneficial effect in the spectral domain was offset by the worsened temporal sensitivity. We hypothesize that the implementation of fast Fourier transform (FFT) processing, instead of the traditionally used bandpass filters, explains the reduction of temporal sensitivity. In this study, spectral ripple discrimination, temporal modulation detection, and speech intelligibility in noise were assessed in a two-week take-home trial with 3 speech coding strategies: one with conventional bandpass filters (HiRes), one with FFT-based filters (HiRes FFT), and one with FFT-based filters and current steering (HiRes Optima). One participant dropped out due to discomfort with both research programs. The 10 remaining participants performed equally well on all tasks with all three speech coding strategies, implying that FFT processing does not change the ability of CI recipients to discriminate spectral or temporal information or speech understanding.

A Fixed Rate Speech Coder Based on the Filter Bank Method and the Inflection Point Detection

A fixed rate speech coder based on the filter bank and the non-uniform sampling technique is proposed. The non-uniform sampling is achieved by the detection of inflection points (IPs). A speech block is band passed by the filter bank, and the subband signals are processed by the IP detector, and the detected IP patterns are compared with entries of the IP database. For each subband signal, the address of the closest member of the database and the energy of the IP pattern are transmitted through channel. In the receiver, the decoder recovers the subband signals using the received addresses and the energy information, and reconstructs the speech via the filter bank summation. As results, the coder shows fixed data rate contrary to the existing speech coders based on the non-uniform sampling. Through computer simulation, the usefulness of the proposed technique is confirmed. The signal-to-noise ratio (SNR) performance of the proposed method is comparable to that of the uniform sampled pulse code modulation (PCM) below 20 kbps data rate.

Auditory brainstem responses to stop consonants predict literacy.

Precise temporal coding of speech plays a pivotal role in sound processing throughout the central auditory system, which, in turn, influences literacy acquisition. The current study tests whether an electrophysiological measure of this precision predicts literacy skills. Complex auditory brainstem responses were analysed from 62 native German-speaking children aged 11-13years. We employed the cross-phaseogram approach to compute the quality of the electrophysiological stimulus contrast [da] and [ba]. Phase shifts were expected to vary with literacy. Receiver operating curves demonstrated a feasible sensitivity and specificity of the electrophysiological measure. A multiple regression analysis resulted in a significant prediction of literacy by delta cross-phase as well as phonological awareness. A further commonality analysis separated a unique variance that was explained by the physiological measure, from a unique variance that was explained by the behavioral measure, and common effects of both. Despite multicollinearities between literacy, phonological awareness, and subcortical differentiation of stop consonants, a combined assessment of behavior and physiology strongly increases the ability to predict literacy skills. The strong link between the neurophysiological signature of sound encoding and literacy outcome suggests that the delta cross-phase could indicate the risk of dyslexia and thereby complement subjective psychometric measures for early diagnoses.

Speech evoked auditory brainstem response and cognitive disorders

Speech evoked auditory brainstem response(s-ABR)is evoked by compound syllable, and those stimulus are similar to the daily language which convey both semantic information and non-semantic information. Speech coding program can take place at brainstem. As a new method, s-ABR may reveal the mystery of speech coding program. Many tests have proved that s-ABR is somehow related to cognitive ability. We mainly illustrated the possibility of grading the cognitive ability using s-ABR, the abnormal test result from those cognitive disorders, and the family factors that contribute to cognitive disorder.

Composition of Deep and Spiking Neural Networks for Very Low Bit Rate Speech Coding

Most current very low bit rate (VLBR) speech coding systems use hidden Markov model (HMM) based speech recognition/synthesis techniques. This allows transmission of information (such as phonemes) segment by segment that decreases the bit rate. However, the encoder based on a phoneme speech recognition may create bursts of segmental errors. Segmental errors are further propagated to optional suprasegmental (such as syllable) information coding. Together with the errors of voicing detection in pitch parametrization, HMM-based speech coding creates speech discontinuities and unnatural speech sound artefacts. In this paper, we propose a novel VLBR speech coding framework based on neural networks (NNs) for end-to-end speech analysis and synthesis without HMMs. The speech coding framework relies on phonological (sub-phonetic) representation of speech, and it is designed as a composition of deep and spiking NNs: a bank of phonological analysers at the transmitter, and a phonological synthesizer at the receiver, both realised as deep NNs, and a spiking NN as an incremental and robust encoder of syllable boundaries for coding of continuous fundamental frequency (F0). A combination of phonological features defines much more sound patterns than phonetic features defined by HMM-based speech coders, and the finer analysis/synthesis code contributes into smoother encoded speech. Listeners significantly prefer the NN-based approach due to fewer discontinuities and speech artefacts of the encoded speech. A single forward pass is required during the speech encoding and decoding. The proposed VLBR speech coding operates at a bit rate of approximately 360 bits/s.