Speech Coding Research Articles

Effect of Number of Electrodes Used to Elicit Electrical Stapedius Reflex Thresholds in Cochlear Implants

Introduction: When mapping cochlear implant (CI) patients with limited reporting abilities, the lowest electrical stimulus level that produces a stapedial reflex (i.e., the electrical stapedius reflex threshold [eSRT]) can be measured to estimate the upper bound of stimulation on individual or a subset of CI electrodes. However, eSRTs measured for individual electrodes or a subset of electrodes cannot be used to predict the global adjustment of electrical stimulation levels needed to achieve comfortable loudness sensations that can be readily used in a speech coding strategy. In the present study, eSRTs were measured for 1-, 4-, and 15-electrode stimulation to (1) determine changes in eSRT levels as a function of the electrode stimulation mode and (2) determine which stimulation mode eSRT levels best approximate comfortable loudness levels from patients’ clinical maps. Methods: eSRTs were measured with the 3 different electrical stimulation configurations in 9 CI patients and compared with behaviorally measured, comfortable loudness levels or M-levels from patients’ clinical maps. Results: A linear, mixed-effects, repeated-measures analysis revealed significant differences (p < 0.01) between eSRTs measured as a function of the stimulation mode. No significant differences (p = 0.059) were measured between 15-electrode eSRTs and M-levels from patients’ clinical maps. The eSRTs measured for 1- and 4-electrode stimulation differed significantly (p < 0.05) from the M-levels on the corresponding electrodes from the patients’ clinical map. Conclusion: eSRT profiles based on 1- or 4-electrode stimulation can be used to determine comfortable loudness level on either individual or a subset of electrodes, and 15-electrode eSRT profiles can be used to determine the upper bound of electrical stimulation that can be used in a speech coding strategy.

Enhancement by postfiltering for speech and audio coding in ad hoc sensor networks

Enhancement algorithms for wireless acoustic sensor networks (WASNs) are indispensable with the increasing availability and usage of connected devices with microphones. Conventional spatial filtering approaches for enhancement in WASNs approximate quantization noise with an additive Gaussian distribution, which limits performance due to the non-linear nature of quantization noise at lower bitrates. This work proposes a postfilter for enhancement based on Bayesian statistics to obtain a multidevice signal estimate, which explicitly models the quantization noise. The experiments using perceptual signal-to-noise ratio, perceptual evaluation of speech quality, and MUSHRA (multistimulus with hidden reference and anchors) scores demonstrate that the proposed postfilter can be used to enhance signal quality in ad hoc sensor networks.

Communicative and Emotive Characteristics of the Diplomatic Idiolect

This article is devoted to the study of communicatively conditioned professional idiolect and the emotional speech of a diplomat by means of quantitative computer content analysis and semantic analysis. The characteristics of a professional diplomat’s idiolect, changes in speech that have occurred over time are determined. Particular attention is paid to lexical units, a change in connotations, the degree of overall emotional performance, as well as clichés and aphorisms. The concept of “implicitly emotive diplomatic clichés” is introduced, which is characteristic of the type of discourse under study. According to the results of the study, it is concluded that a professional idiolect is expressed in the skillful switching of speech codes, the rejection of pragmatic appraisal, the ability to explicitly express an appraisal using facts and the exact use of changing connotations, as well as implicitly emotive diplomatic clichés, which allows a diplomat through a “delegated subject” express the country’s attitude to current events, but remain within the framework of the diplomatic speech protocol. The study allows us to come closer to a clearer and deeper understanding of the phenomenon of a professional diplomat’s idiolect as a complex set of speech and textual practices aimed at solving complex international problems in a peaceful way, i.e. verbally. The research material was S.V. Lavrov as the Permanent Representative to the UN Security Council speeches on the theme of ‘peacekeeping’ stored in the Security Council digital archive for 1994-2004, as well as the material of the press conference on 01/17/2020 as the Minister of Foreign Affairs of the Russian Federation.

Effects of age on psychophysical measures of auditory temporal processing and speech reception at low and high levels

Age-related cochlear synaptopathy (CS) has been shown to occur in rodents with minimal noise exposure, and has been hypothesized to play a crucial role in age-related hearing declines in humans. It is not known to what extent age-related CS occurs in humans, and how it affects the coding of supra-threshold sounds and speech in noise. Because in rodents CS affects mainly low- and medium-spontaneous rate (L/M-SR) auditory-nerve fibers with rate-level functions covering medium-high levels, it should lead to greater deficits in the processing of sounds at high than at low stimulus levels. In this cross-sectional study the performance of 102 listeners across the age range (34 young, 34 middle-aged, 34 older) was assessed in a set of psychophysical temporal processing and speech reception in noise tests at both low, and high stimulus levels. Mixed-effect multiple regression models were used to estimate the effects of age while partialing out effects of audiometric thresholds, lifetime noise exposure, cognitive abilities (assessed with additional tests), and musical experience. Age was independently associated with performance deficits on several tests. However, only for one out of 13 tests were age effects credibly larger at the high compared to the low stimulus level. Overall these results do not provide much evidence that age-related CS, to the extent to which it may occur in humans according to the rodent model of greater L/M-SR synaptic loss, has substantial effects on psychophysical measures of auditory temporal processing or on speech reception in noise.

A novel voice activity detection algorithm using modified global thresholding

Voice activity detection is currently a challenging task that is applicable in real time applications such as speech coding and recognition. It is due to the low signal-to-noise ratio that affected the structural properties. Voice activity detection helps in detecting the speech region that is present in various nonstationary noises. The literature associated with Voice activity detection suggests that numerous works use unbalanced classification approach with higher and poor, speech and non-speech detection rates, respectively. This leads to the condition that majority of the noisy segments are categorized as speech. Hence, to overcome this issue, we propose a novel modified global thresholding scheme that has a fuzzy entropy tool. Our proposal can effectively identify both regions by locating the transition from non-speech to speech areas and vice versa. This will improve the detection rates as misclassification error of noisy segments as speech segments are minimized. The performance of the proposed algorithm is tested on various additive non-stationary noises at different SNR levels. In most of the existing research, it is often assumed that the noise is stationary for a particular instant in order to estimate the noise information. But in real scenario this is impossible. Our significant contribution is in developing an algorithm that handles the signals which possess nonstationary noises and various complex events which can be a mixture of different noises. As the characteristics of speech vary over time (nonstationary), when additively mixed with nonstationary noises becomes more challenging especially at low SNR levels (− 5 dB, − 10 dB). Therefore, the problem becomes more complicated like that in the real-time scenario. Our proposed method produces 91.98% and 87.38% of speech and non-speech detection rates in low SNR levels, respectively. It also obtains an accuracy of 93.39% for speech babble noises against the state-of-art algorithms which varied between 50 and 80% only. Similarly, NDS rates of the proposed algorithm is very minimal, i.e. less than 10% compared to the bench mark algorithms which had at least 50% or more of the noise detected as speech segments. The significance of our invention is in precisely locating where a speech begins and ends in a given noisy speech. We believe that we have produced a path breaking approach that can be helpful in real time applications in speech processing.

Noninvasive Measures of Distorted Tonotopic Speech Coding Following Noise-Induced Hearing Loss.

Animal models of noise-induced hearing loss (NIHL) show a dramatic mismatch between cochlear characteristic frequency (CF, based on place of innervation) and the dominant response frequency in single auditory-nerve-fiber responses to broadband sounds (i.e., distorted tonotopy, DT). This noise trauma effect is associated with decreased frequency-tuning-curve (FTC) tip-to-tail ratio, which results from decreased tip sensitivity and enhanced tail sensitivity. Notably, DT is more severe for noise trauma than for metabolic (e.g., age-related) losses of comparable degree, suggesting that individual differences in DT may contribute to speech intelligibility differences in patients with similar audiograms. Although DT has implications for many neural-coding theories for real-world sounds, it has primarily been explored in single-neuron studies that are not viable with humans. Thus, there are no noninvasive measures to detect DT. Here, frequency following responses (FFRs) to a conversational speech sentence were recorded in anesthetized male chinchillas with either normal hearing or NIHL. Tonotopic sources of FFR envelope and temporal fine structure (TFS) were evaluated in normal-hearing chinchillas. Results suggest that FFR envelope primarily reflects activity from high-frequency neurons, whereas FFR-TFS receives broad tonotopic contributions. Representation of low- and high-frequency speech power in FFRs was also assessed. FFRs in hearing-impaired animals were dominated by low-frequency stimulus power, consistent with oversensitivity of high-frequency neurons to low-frequency power. These results suggest that DT can be diagnosed noninvasively. A normalized DT metric computed from speech FFRs provides a potential diagnostic tool to test for DT in humans. A sensitive noninvasive DT metric could be used to evaluate perceptual consequences of DT and to optimize hearing-aid amplification strategies to improve tonotopic coding for hearing-impaired listeners.

CZĘSTOTLIWOŚCIOWE ODPOWIEDZI WYWOŁANE U DOROSŁYCH, KTÓRZY MOGĄ LUB NIE MOGĄ ŚPIEWAĆ CZYSTO

BackgroundIt is believed that auditory processing occurs normally in people who can sing in tune and improperly in people who cannot. Auditory feedback seems to be a crucial factor in the way the voice is produced and monitored. Evaluation of auditory processing using the Frequency Following Response (FFR) allows fine-grained neural processing to be objectively identified and might be a way of differentiating between those who sing in-tune and those who sing off-tune.The aim of this study was to analyse the FFR responses of people who can sing in-tune and compare them to those who sing off-tune.Material and methodsFFR responses were recorded in 37 adults who were assigned to one of two groups: (i) a control group (CG) consisting of 17 adults who could sing in-tune (ii) an experimental group (EG) consisting of 20 adults who sang off-tune.ResultsThere were statistically significant differences in the electro-physiological responses of the EG compared to the CG for the latencies of waves A, C, D, and F in the right ear. In contrast, FFR amplitude measurements did not seem to be a suitable parameter for identifying changes in the coding of speech sounds.ConclusionsFFR responses in the EG showed a different pattern from the CG group, with a number of longer latencies in the EG. However, FFR amplitude did not differ significantly between the groups.

Evaluation of Potential Efficiency of Speech Coding Using Different Parameters of Linear Prediction

The paper presents the results for estimating the potential coding efficiency of spectrum envelope waveform (SEW) of speech signals (SS) by using the linear prediction (LP) method and using different sets of alternative equivalent parameters (AEP) that include line spectral pairs/projections (LSP), line spectral frequencies (LSF), and alternative line spectral parameters of highest splitting (LSP-HS and LSF-HS). These results are derived by the proposed approach based on using the LP method for SS SEW coding with maximum frame overlap during its analysis. It includes the examination of such coding scheme as approaching an appropriate analog vector source in each of AEP spaces, the stepwise designing of an appropriate vector codebook in each of spaces with gradual increase of its size and employing the ideal scheme of vector quantization with exhaustive search at each stage. The distortions versus rate relationships have been calculated based on the results of analysis in each of AEP spaces. In addition, a generalized function is proposed for approximation of the specified relationships. The technique is presented that makes it possible in each AEP space to estimate Shannon’s lower bound, the dispersion of the Gaussian equivalent source, differential entropy, redundancy, values of the weighting constant in the generalized formula of entropy, and other entropy characteristics of coding the equivalent sources (parameters of SS SEW) in these spaces. The efficiency indicators of real and potential coding of corresponding AEP are proposed and calculated. It has been shown that in conjunction with the combination of proposed efficiency indicators, the best results demonstrate spaces of line spectral parameters of highest splitting (LSP-HS and LSF-HS).

Is Free Speech in Danger on University Campus? Some Preliminary Evidence from a\xa0Most Likely Case

Although universities play a key role in questions of free speech and political viewpoint diversity, they are often associated with the opposite of a free exchange of ideas: a proliferation of restrictive campus speech codes, violent protests against controversial speakers and even the firing of inconvenient professors. For some observers these trends on university campuses are a clear indicator of the dire future for freedom of speech. Others view these incidents as scandalized singular events and regard campus intolerance as a mere myth. We take an empirical look at some of the claims in the debate and present original survey evidence from a most likely case: the leftist social science studentship at Goethe University Frankfurt. Our results show that taking offense is a common experience and that a sizable number of students are in favor of restricting speech on campus. We also find evidence for conformity pressures on campus and that both the desire to restrict speech and the reluctance to speak openly differ significantly across political ideology. Left-leaning students are less likely to tolerate controversial viewpoints and right-leaning students are more likely to self-censor on politically sensitive issues such as gender, immigration, or sexual and ethnic minorities. Although preliminary, these findings may have implications for the social sciences and academia more broadly.

Cross-cultural communication of the Indonesian students in China

This paper aims to find out how the process and the steps taken by Indonesian students studying in China to overcome culture shock during their study in China. This research seeks to provide answers to cultural shock events that arise from a process of cross-cultural communication. The data were obtained by conducting an in-depth interview process and then processed optimally using the Speech Codes Theory (SCT) initiated by Gerry Philipsen. This study shows that Indonesian students tend to cope with cultural shock events from cross-cultural communication processes when mingling with other community members.

Inside Alcohol Behavioral Couple Therapy (ABCT): In-session speech trajectories and drinking outcomes

The alcohol treatment literature has established in-session client speech as a mechanism of change that therapist behavior can influence and that can predict drinking outcomes. This study aimed to explore temporal patterns of in-session speech in Alcohol Behavioral Couple Therapy (ABCT), including the unique interplay between client and partner speech and the role of speech trajectories in predicting client drinking outcomes. Participants were 165 heterosexual couples receiving ABCT in one of four clinical trials. We coded client speech on an utterance-by-utterance basis using the System for Coding Couples' Interactions in Therapy–Alcohol. We focused on individual-level speech codes of change talk and sustain talk and couple-level variables of positive and negative interactions. We segmented the initial and midtreatment sessions into quartiles to conduct path analyses and latent growth curve models. Path analyses suggested that clients and partners may not have been aligned in terms of treatment goals at the start of the therapy. This misalignment within couples was pronounced during the initial session and decreased by the midtreatment session, reflecting progression toward treatment goals. Of the latent growth curve models, only client sustain talk during the midtreatment session predicted greater client drinking at the end of treatment. Results provide insight into the inner workings of ABCT and suggest recommendations for ABCT therapists. This study also supports a growing consensus that sustain talk may be a stronger mechanism of change than change talk in various alcohol treatment interventions.

Comprehensive Survey of FIR-Based Sample Rate Conversion

Sample rate conversion (SRC) is ubiquitous and critical function of software defined radio and other signal processing systems (speech coding and synthesis, computer simulation of continuous-time systems, etc..). In this paper, we present a survey on linear phase finite impulse response (FIR) based sampling rate conversion. Many different FIR-based SRC solutions exist, such as classical FIR, polyphase, Farrow, cascaded-integrator-comb, and Newton structures. Each one of these solutions is presented differently in the literature, and SRC reference books introducing the subject are often missing hardware implementation aspects. The main objective of this paper is to provide a simple and comprehensive overview of main FIR-based SRC techniques from theoretical to hardware implementation aspects. The state of the art of FIR-based SRC filters is summed-up through a concise derivation of the different solutions from a common root: linear phase FIR filters. Each SRC solution is presented from both theoretical and practical implementation points of view. The paper provides a succinct tutorial that introduces SRC, and helps identifying and implementing the appropriate FIR-based SRC architecture for any given applications.

Word length effects on pictorial memory encoding in 5- to 7-year-old children: An eye-tracking study

How speech coding is used for visual material is critical to understanding the link between language and memory development. Using eye-tracking, the present study examined whether the number of syllables of pictured objects’ names, monosyllabic versus multisyllabic, predicts looking time at the stimuli-objects in the context of a memory task, thereby indicating verbal recoding. The children’s (N = 39, ages 5–7 years) language ability was also considered. Younger children (5;1–6;3 years) did not appear to sub-vocally recode the verbal labels of visual stimuli during encoding, whereas older children (6;4–7;3 years) who were already attending elementary school looked longer at objects with multisyllabic labels. Notably, it was primarily the less verbally competent among those children who contributed to the effect. Thus, our results demonstrate, for the first time, that young children’s sub-vocal verbal processing when trying to memorize visual stimuli may be contingent upon verbal competence.

Nicotinic Receptor Subunit Distribution in Auditory Cortex: Impact of Aging on Receptor Number and Function.

The presence of novel or degraded communication sounds likely results in activation of basal forebrain cholinergic neurons increasing release of ACh onto presynaptic and postsynaptic nAChRs in primary auditory cortex (A1). nAChR subtypes include high-affinity heteromeric nAChRs commonly composed of α4 and β2 subunits and low-affinity homomeric nAChRs composed of α7 subunits. In young male FBN rats, we detail the following: (1) the distribution/expression of nAChR subunit transcripts in excitatory (VGluT1) and inhibitory (VGAT) neurons across A1 layers; (2) heteromeric nAChR binding across A1 layers; and (3) nAChR excitability in A1 layer (L) 5 cells. In aged rats, we detailed the impact of aging on A1 nAChR subunit expression across layers, heteromeric nAChR receptor binding, and nAChR excitability of A1 L5 cells. A majority of A1 cells coexpressed transcripts for β2 and α4 with or without α7, while dispersed subpopulations expressed β2 and α7 or α7 alone. nAChR subunit transcripts were expressed in young excitatory and inhibitory neurons across L2-L6. Transcript abundance varied across layers, and was highest for β2 and α4. Significant age-related decreases in nAChR subunit transcript expression (message) and receptor binding (protein) were observed in L2-6, most pronounced in infragranular layers. In vitro patch-clamp recordings from L5B pyramidal output neurons showed age-related nAChR subunit-selective reductions in postsynaptic responses to ACh. Age-related losses of nAChR subunits likely impact ways in which A1 neurons respond to ACh release. While the elderly require additional resources to disambiguate degraded speech codes, resources mediated by nAChRs may be compromised with aging.SIGNIFICANCE STATEMENT When attention is required, cholinergic basal forebrain neurons may trigger increased release of ACh onto auditory neurons in primary auditory cortex (A1). Laminar and phenotypic differences in neuronal nAChR expression determine ways in which A1 neurons respond to release of ACh in challenging acoustic environments. This study detailed the distribution and expression of nAChR subunit transcript and protein across A1 layers in young and aged rats. Results showed a differential distribution of nAChR subunits across A1 layers. Age-related decreases in transcript/protein expression were reflected in age-related subunit specific functional loss of nAChR signaling to ACh application in A1 layer 5. Together, these findings could reflect the age-related decline in selective attention observed in the elderly.

FOR THE QUESTION OF LEARNING LOCALIZATION AND COMPOSITION OF TOPONYM OF THE PRONE VALLEY

Ущелье реки Проне расположено в среднем Картли, на правом берегу Куры. Включает в себя Дванис-цкали, Алис-цкали, Пцис-цкали и Сурамис-цкали. В соответствии с историческим термином и сегодня эти реки население называет Проне. Для диахронного изучения существующих в этом ущелье топонимов, с точки зрения лингвистики существенными справками располагают приходская книга Руиси или летопись Руиси, составленная Николозом Мровели Орбелишвили или Капланишвили в 1715 году; «Описание царства грузинского (География Грузии)» Вахушти Багратиони (середина XVIII века); «Описание Картли и Кахетии» Иоанна Багратиони (1794-99 годы) и материалы, в разное время записанные и проанализированные грузинскими исследователями (С. Макалатия, Ал. Глонти, М. Коберидзе, Н. Отинашвили) с точки зрения диалектологии. Речь деревень Дванис-цкали и Пцис-цкали относится к шида картлийскому говору, а речь жителей сел Алис-цкали и Сурамис-цкали относится к западно-картлийскому. Топонимы ущелья реки Проне по своему содержанию и образованию разнообразны, существующие средства образования слов подобны грузинскому литературному языку и другим диалектам грузинского языка. Встречаются как субстантивированные, так и атрибутивные субстантивные микро (Накедлареби, Сакидула ...), макро (Дванисцкали, Проне…) и собственные топонимы (Двани, Дирби, Тахтидзири ...), где составляющей частью композита встречаются «цкали»(вода), «тба»(озеро), «геле»(речка), «чала» (роща), «хеви» (ущелье), «чаоби» (болото), «мореви»(водоворот). Детерминантой к топонимам такого типа подтверждаются как общественные, так и собственные имена. Фонетические процессы, характерные для грузинской языковой системы, также встречаются в топонимах ущелья реки Проне. Например: потеря звука, замещение звуков, включение звука… Активно используется суффиксальное (- ет, - ит , - ат, ис-и, ат (а), - aт (a), - та, - еул, - ул , - ур, - ован, -осан, -ин-а, -еб, -мар (нар) и преффиксально-суффиксальное образование (на-ев, са-е, са-ур, са-иа. Различие создает ареал распространения топонимов и частота производного средства. Различные случаи указывают на тенденцию унификации и случаи переключения речевого кода. Встречается территороиальная локализация использования диалектной лексики в топонимии. Проанализированный материал по содержанию и образованию разнообразен. Образование грузинское, не встречаются заимствованные термины. По частоте использования выделяются топонимы с основой на «цкал» и «цкар». Большинство топонимов сложные слова. К топонимам такого типа определителем выступают как нарицательные, так и собственные имена.

New perspectives on speech motor planning and programming in the context of the four- level model and its implications for understanding the pathophysiology underlying apraxia of speech and other motor speech disorders

ABSTRACT Background: The complexity of speech motor control, and the incomplete conceptualisation of phases in the transformation of the speech code from linguistic symbols to a code amenable to a motor system, tend to obscure the understanding of acquired apraxia of speech (AOS). The four-level framework (FLF) of speech sensorimotor control suggests the differentiation between speech motor planning, programming and execution and locate the locus of disruption in AOS in the motor planning phase. Currently, terminological confusion and uncertainty regarding phases in speech motor control still complicate the characterisation of AOS. This neuromotor disorder is inconsistently described in the literature as a “planning or programming”, “planning and programming”, or as a “planning and/or programming” disorder. Purpose: To describe a new version of the FLF, the FL (four-level) model, which further explicates and differentiates between speech motor planning, programming, and execution levels or phases of processing; to integrate concepts from computational modelling into the FL model and propose distinct control architectures for both the planning and programming levels; and to identify the loci and nature of disruption in the motor planning phase which could explain the pathophysiology and core features of AOS. Discussion and Conclusions: A four-level model is presented that differentiates two pre-execution phases and an execution phase. The first pre-execution phase is controlled by a motor planner and involves an inverse model, an efference copy, and a forward model for each sound or over-learnt utterance. This phase also involves a forward predictive planner which enables the system to handle the planning of several sounds and to plan coarticulation of sounds. The motor planner is operated according to an auxiliary forward model architecture. AOS is depicted as a breakdown at several possible points in the motor planning phase. The second pre-execution phase is driven by a motor program generator and predictive controller that is governed by an integral forward model architecture. The final execution phase is portrayed as being driven by closed loop control. The conceptualization of the programmer challenges the traditional view of execution and not only that of planning as is generally accepted. The implications for the classification of motor speech disorders are discussed. Future research should address the exact nature of articulatory movements and other features of speech across the range of planning, pure programming, programming-execution and pure execution disorders.

AN EFFICIENT SPEECH GENERATIVE MODEL BASED ON DETERMINISTIC/STOCHASTIC SEPARATION OF SPECTRAL ENVELOPES

The paper presents a speech generative model that provides an efficient way of generating speech waveform from its amplitude spectral envelopes. The model is based on hybrid speech representation that includes deterministic (harmonic) and stochastic (noise) components. The main idea behind the approach originates from the fact that speech signal has a determined spectral structure that is statistically bound with deterministic/stochastic energy distribution in the spectrum. The performance of the model is evaluated using an experimental low-bitrate wide-band speech coder. The quality of reconstructed speech is evaluated using objective and subjective methods. Two objective quality characteristics were calculated: Modified Bark Spectral Distortion (MBSD) and Perceptual Evaluation of Speech Quality (PESQ). Narrow-band and wide-band versions of the proposed solution were compared with MELP (Mixed Excitation Linear Prediction) speech coder and AMR (Adaptive Multi-Rate) speech coder, respectively. The speech base of two female and two male speakers were used for testing. The performed tests show that overall performance of the proposed approach is speaker-dependent and it is better for male voices. Supposedly, this difference indicates the influence of pitch highness on separation accuracy. In that way, using the proposed approach in experimental speech compression system provides decent MBSD values and comparable PESQ values with AMR speech coder at 6,6 kbit/s. Additional subjective listening testsdemonstrate that the implemented coding system retains phonetic content and speaker’s identity. It proves consistency of the proposed approach.

Comparative analysis of speech coders

In this paper a comparative analysis of some of the most popular speech coders is presented. Qualitatively and quantitatively are tested Linear Prediction Coding in its implementation LPC-10e and with the use of auto-correlation and covariance,

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.

Trademarks, Hate Speech, and Solving a Puzzle of Viewpoint Bias

In this article, I argue that in the seemingly straightforward ruling in Iancu v Brunetti, striking down a provision of the law governing trademarks, the Court revealed a significant clarification of the limits of the doctrine of viewpoint discrimination. In free speech doctrine, the Court is unanimous in condemning viewpoint discrimination, but its contours remain “slippery” because viewpoint bias is rarely a game changer in a given case. One enduring puzzle is whether a limit on the mode or manner of communication – a ban on racial epithets, for example – embodies viewpoint discrimination. This question has been unresolved for almost thirty years, ever since the Court’s murky opinion in R.A.V. v St. Paul struck down, as viewpoint based, an ordinance aimed at fighting words that “arouse[d] anger, alarm, or resentment in others on the basis of race, color, creed, religion, or gender.” The Court was not clear whether the ordinance was viewpoint biased because it regulated one side of a public debate or limited a mode or manner of debate for both sides. The difference between these two possible readings matters: if limits on modes or manners of speech are deemed to be viewpoint discrimination, then it may be virtually impossible to enact, for example, bans on racial epithets at a public university. But Brunetti clarified matters considerably. The Court struck down the provisions of the Lanham Act prohibiting the registration of “immoral” and “scandalous” marks as viewpoint biased. The conflict between Justice Elena Kagan’s opinion for the Court and the lead opposition opinion written by Justice Sonia Sotomayor illuminated an important area of agreement that appears to control a majority of the Court. That agreement is this: that worries about viewpoint bias do not ordinary come into play when the government regulates the mode and manner of communication as opposed to the ideas conveyed. Such a principle has a number of implications. Perhaps the most important is that R.A.V. is less significant in First Amendment doctrine than it has seemed for thirty years. Also, some kinds of speech codes could survive First Amendment challenge, as long as they apply in certain fora and are aimed at the mode and manner of communication rather than the ideas expressed. Another implication would be that it would be possible for Congress to rewrite the now-defunct provisions of the Lanham Act to survive First Amendment challenge and also satisfy much of Congress’s original goals.