Artificial Speech Research Articles

Genos Dikanikon: Amateur and Professional Speech in the Courtrooms of Classical Athens (review)

Reviewed by: Genos Dikanikon: Amateur and Professional Speech in the Courtrooms of Classical Athens Edmund M. Burke Victor Bers . Genos Dikanikon: Amateur and Professional Speech in the Courtrooms of Classical Athens. Hellenic Studies 33. Washington, D.C.: Center for Hellenic Studies, 2009. Dist. by Harvard University Press. ix + 159 pp. Paper, $15.95. That the Athenians were a litigious people is a commonplace. Yet the extant dicastic speeches from the fifth and fourth centuries are the product exclusively of elite hands and overwhelmingly deal with litigation involving elites. Indeed, a significant percentage of those speeches were paid for, as individuals wealthy enough retained the services of professional speechwriters, logographoi, to help negotiate the hazards of the courts. With a few noteworthy exceptions, then, the poor are underrepresented in the corpus of the Attic orators, although invariably they constituted a plurality of the city's citizenry. In this slim volume, Victor Bers, the author a quarter century ago of an important article on the impact of juror outbursts in the Athenian courts ("Dikastic Thorubos," in P. A. Cartledge and F. D. Harvey, eds. Crux: Essays Presented to G. E. M. de Ste. Croix on his 75th Birthday [Exeter 1985] 1-15), returns to the dicastic arena, this time to examine the practice of self-presentation by amateur and professional speakers. Explicitly Bers argues "that the professional component of the genos dikanikon represents only a portion of the speechmaking that went on in the Athenian courts; (and) that many men constrained to rely entirely or mainly on their own resources also spoke in court." Beyond the demonstration of a sociological phenomenon, Bers argues further "that their (viz., amateurs') speech in court resembled routine speech in a number of ways, and that in these they differ from professional speech; [and] that professional speech was crafted to avoid certain features of amateur speech that seemed to cause a speaker's failure . . . in particular those that manifested excessive emotion when it was in his interest to appear unafraid and unperturbed." These avoidance strategies Bers labels evitanda (5). The issue of what Bers characterizes as "the concrete reality and drama of the Attic courts" (3), including the use of powerful affective speech and gesture, has received much scholarly attention since the mid-1980s. A feature of Bers' work, then, is polemical, as he stakes out his positions in several scholarly debates. His appeal to ancient evidence is carefully selective, not exhaustive, and occasionally there is a buttressing of an argument by appeal to a comparative social model or practice (see, e.g., 70-71, 85). There are seven chapters whose titles and arrangement suggest at one level close analysis of what would have been the concrete reality for an Athenian litigant speaking before a court. So "The Challenge of Court Speech" (chap. 1), [End Page 699] "Amateur Litigants, Amateur Speakers" (chap. 2), "Terrors of the Courtroom" (chap. 4), "Performance as Evidence" (chap. 5), "Appeals to Pity and Displays of Anger" (chap. 6), "Tactics, Amateur and Professional" (chap. 7). Chapter 3, "Natural and Artificial Speech from Homer to Hyperides: A Brief Sketch," follows a different, if parallel course. Here Bers concludes that ". . . other things being equal, an overemotional style was likely to fail in Athenian courts" (35), and with that enters the recent debates about the relationship between the courts and the tragic stage (for Bers, subsequent to Antiphon, there was virtually none [37-43]), and the significance and meaning of appeals to pity and expressions of anger (taken up in chap. 6). Beyond the seven chapters, there is an appendix ("The Formality Hypothesis," for which see below), a list of works cited, an index locorum, an index of Greek words, and a subject index. The case Bers makes for large numbers of poor amateur litigants in the courts fashioning their own presentations is speculative but compelling, as he extrapolates from the work of Mogens Hansen and P. J. Rhodes, fleshing out the various types of circumstance that could land an individual in court (from public scrutinies to family feuds), along with the practical limitations for men of limited means in securing professional help (e.g., logographoi cost money). There is succinctness and clarity here on a...

Concatenative Synthesis of Persian Language Based on Word, Diphone and Triphone Databases

In this paper a Persian text-to-speech system based on concatenative speech synthesis approach is proposed. Nowadays, concatenative method is used in most modern TTS systems to produce artificial speech. Choosing an appropriate unit for creating a database is the major challenge in this method. In the proposed method, such database is created with different sizes of speech units and is used to produce speak utterances which include words, diphones and triphones. Moreover, a dictionary of 600 common Persian words is built. The smaller speech units (diphones and triphones) is used for synthesis. They are chosen to achieve unlimited vocabulary of speech, while the word units are used for synthesizing limited set of sentences. The simulation results show the effectiveness of the proposed method.

What's in a face? Visual contributions to speech segmentation

Recent research has demonstrated that adults successfully segment two interleaved artificial speech streams with incongruent statistics (i.e., streams whose combined statistics are noisier than the encapsulated statistics) only when provided with an indexical cue of speaker voice. In a series of five experiments, our study explores whether learners can utilise visual information to encapsulate statistics for each speech stream. We initially presented learners with incongruent artificial speech streams produced by the same female voice along with an accompanying visual display. Learners successfully segmented both streams when the audio stream was presented with an indexical cue of talking faces (Experiment 1). This learning cannot be attributed to the presence of the talking face display alone, as a single face paired with a single input stream did not improve segmentation (Experiment 2). Additionally, participants failed to successfully segment two streams when they were paired with a synchronised single talking face display (Experiment 3). Likewise, learners failed to successfully segment both streams when the visual indexical cue lacked audio-visual synchrony, such as changes in background screen colour (Experiment 4) or a static face display (Experiment 5). We end by discussing the possible relevance of the speaker's face in speech segmentation and bilingual language acquisition.

Colliding cues in word segmentation: The role of cue strength and general cognitive processes

The process of word segmentation is flexible, with many strategies potentially available to learners. This experiment explores how segmentation cues interact, and whether successful resolution of cue competition is related to general executive functioning. Participants listened to artificial speech streams that contained both statistical and pause-defined cues to word boundaries. When these cues ‘collide’ (indicating different locations for word boundaries), cue strength appears to dictate the predominant parsing strategy. When cues are relatively equal in strength, the ability to successfully deploy a segmentation strategy significantly correlates with stronger performance on the Simon task, a non-linguistic cognitive task typically thought to involve executive processes such as inhibitory control and selective attention. These results suggest that general information processing strategies may play a role in solving one of the early challenges for language learners.

Exploiting Multiple Sources of Information in Learning an Artificial Language: Human Data and Modeling

This study investigates the joint influences of three factors on the discovery of new word-like units in a continuous artificial speech stream: the statistical structure of the ongoing input, the initial word-likeness of parts of the speech flow, and the contextual information provided by the earlier emergence of other word-like units. Results of an experiment conducted with adult participants show that these sources of information have strong and interactive influences on word discovery. The authors then examine the ability of different models of word segmentation to account for these results. PARSER (Perruchet & Vinter, 1998) is compared to the view that word segmentation relies on the exploitation of transitional probabilities between successive syllables, and with the models based on the Minimum Description Length principle, such as INCDROP. The authors submit arguments suggesting that PARSER has the advantage of accounting for the whole pattern of data without ad-hoc modifications, while relying exclusively on general-purpose learning principles. This study strengthens the growing notion that nonspecific cognitive processes, mainly based on associative learning and memory principles, are able to account for a larger part of early language acquisition than previously assumed.

Brain–computer interfaces for speech communication

This paper briefly reviews current silent speech methodologies for normal and disabled individuals. Current techniques utilizing electromyographic (EMG) recordings of vocal tract movements are useful for physically healthy individuals but fail for tetraplegic individuals who do not have accurate voluntary control over the speech articulators. Alternative methods utilizing EMG from other body parts (e.g., hand, arm, or facial muscles) or electroencephalography (EEG) can provide capable silent communication to severely paralyzed users, though current interfaces are extremely slow relative to normal conversation rates and require constant attention to a computer screen that provides visual feedback and/or cueing. We present a novel approach to the problem of silent speech via an intracortical microelectrode brain–computer interface (BCI) to predict intended speech information directly from the activity of neurons involved in speech production. The predicted speech is synthesized and acoustically fed back to the user with a delay under 50 ms. We demonstrate that the Neurotrophic Electrode used in the BCI is capable of providing useful neural recordings for over 4 years, a necessary property for BCIs that need to remain viable over the lifespan of the user. Other design considerations include neural decoding techniques based on previous research involving BCIs for computer cursor or robotic arm control via prediction of intended movement kinematics from motor cortical signals in monkeys and humans. Initial results from a study of continuous speech production with instantaneous acoustic feedback show the BCI user was able to improve his control over an artificial speech synthesizer both within and across recording sessions. The success of this initial trial validates the potential of the intracortical microelectrode-based approach for providing a speech prosthesis that can allow much more rapid communication rates.

How the human brain recognizes speech in the context of changing speakers.

We understand speech from different speakers with ease, whereas artificial speech recognition systems struggle with this task. It is unclear how the human brain solves this problem. The conventional view is that speech message recognition and speaker identification are two separate functions and that message processing takes place predominantly in the left hemisphere, whereas processing of speaker-specific information is located in the right hemisphere. Here, we distinguish the contribution of specific cortical regions, to speech recognition and speaker information processing, by controlled manipulation of task and resynthesized speaker parameters. Two functional magnetic resonance imaging studies provide evidence for a dynamic speech-processing network that questions the conventional view. We found that speech recognition regions in left posterior superior temporal gyrus/superior temporal sulcus (STG/STS) also encode speaker-related vocal tract parameters, which are reflected in the amplitude peaks of the speech spectrum, along with the speech message. Right posterior STG/STS activated specifically more to a speaker-related vocal tract parameter change during a speech recognition task compared with a voice recognition task. Left and right posterior STG/STS were functionally connected. Additionally, we found that speaker-related glottal fold parameters (e.g., pitch), which are not reflected in the amplitude peaks of the speech spectrum, are processed in areas immediately adjacent to primary auditory cortex, i.e., in areas in the auditory hierarchy earlier than STG/STS. Our results point to a network account of speech recognition, in which information about the speech message and the speaker's vocal tract are combined to solve the difficult task of understanding speech from different speakers.

Evaluation of Extremely Small Sound Source Signals Used in Speaking-Aid System with Statistical Voice Conversion

We have so far proposed a speaking-aid system for laryngectomees using a statistical voice conversion technique. In the proposed system, artificial speech articulated with extremely small sound source signals is detected with a Non-Audible Murmur (NAM) microphone, and then, the detected artificial speech is converted into more natural voice in a probabilistic manner. Although this system basically allows laryngectomees to speak while keeping the external source signals silent, it is still questionable how much these new sound source signals affect the converted speech quality. In this paper, we investigate the impact of various sound source signals on voice conversion accuracy. Various small sound source signals are designed by changing the spectral envelope and the waveform power independently. We conduct objective and subjective evaluations. The results of these experimental evaluations demonstrate that voice conversion accepts 1) various sound source signals with different spectral envelopes and 2) large degree of power of the sound source signals unless the power of speaking parts is almost equal to that of silence parts. Moreover, we also investigate the effectiveness of enhancing auditory feedback during speaking with the extremely small sound source signals.

Talking ultrasound

Near real-time speech regeneration for voice-loss patients using ultrasonic speech for phoneme detection and classification is under development by researchers at Nanyang Technological University in Singapore. Their research is important for a range of applications including: prosthesis for voice-loss patients; improving the performance of artificial speech recognition systems, especially in the presence of audible acoustic noise; and for the more private use of mobile telephones in public.

Recognizing Sequences of Sequences

The brain's decoding of fast sensory streams is currently impossible to emulate, even approximately, with artificial agents. For example, robust speech recognition is relatively easy for humans but exceptionally difficult for artificial speech recognition systems. In this paper, we propose that recognition can be simplified with an internal model of how sensory input is generated, when formulated in a Bayesian framework. We show that a plausible candidate for an internal or generative model is a hierarchy of ‘stable heteroclinic channels’. This model describes continuous dynamics in the environment as a hierarchy of sequences, where slower sequences cause faster sequences. Under this model, online recognition corresponds to the dynamic decoding of causal sequences, giving a representation of the environment with predictive power on several timescales. We illustrate the ensuing decoding or recognition scheme using synthetic sequences of syllables, where syllables are sequences of phonemes and phonemes are sequences of sound-wave modulations. By presenting anomalous stimuli, we find that the resulting recognition dynamics disclose inference at multiple time scales and are reminiscent of neuronal dynamics seen in the real brain.

The role of perceptual salience during the segmentation of connected speech

Statistical learning allows listeners to track transitional probabilities among syllable sequences and use these probabilities for subsequent speech segmentation. Recent studies have shown that other sources of information, such as rhythmic cues, can modulate the dependencies extracted via statistical computation. In this study, we explored how syllables made salient by a pitch rise affect the segmentation of trisyllabic words from an artificial speech stream by native speakers of three different languages (Spanish, English, and French). Results showed that, whereas performance of French participants did not significantly vary across stress positions (likely due to language-specific rhythmic characteristics), the segmentation performance of Spanish and English listeners was unaltered when syllables in word-initial and word-final positions were salient, but it dropped to chance level when salience was on the medial syllable. We argue that pitch rise in word-medial syllables draws attentional resources away from word boundaries, thus decreasing segmentation effectiveness.

Perception and hierarchical dynamics

In this paper, we suggest that perception could be modeled by assuming that sensory input is generated by a hierarchy of attractors in a dynamic system. We describe a mathematical model which exploits the temporal structure of rapid sensory dynamics to track the slower trajectories of their underlying causes. This model establishes a proof of concept that slowly changing neuronal states can encode the trajectories of faster sensory signals. We link this hierarchical account to recent developments in the perception of human action; in particular artificial speech recognition. We argue that these hierarchical models of dynamical systems are a plausible starting point to develop robust recognition schemes, because they capture critical temporal dependencies induced by deep hierarchical structure. We conclude by suggesting that a fruitful computational neuroscience approach may emerge from modeling perception as non-autonomous recognition dynamics enslaved by autonomous hierarchical dynamics in the sensorium.

A role for backward transitional probabilities in word segmentation?

A number of studies have shown that people exploit transitional probabilities between successive syllables to segment a stream of artificial continuous speech into words. It is often assumed that what is actually exploited are the forward transitional probabilities (given XY, the probability that X will be followed by Y), even though the backward transitional probabilities (the probability that Y has been preceded by X) were equally informative about word structure in the languages involved in those studies. In two experiments, we showed that participants were able to learn the words from an artificial speech stream when the only available cues were the backward transitional probabilities. Learning is as good under those conditions as when the only available cues are the forward transitional probabilities. Implications for some current models of word segmentation, particularly the simple recurrent networks and PARSER models, are discussed.

Evaluation of an Artificial Speech Bandwidth Extension Method in Three Languages

Quality and intelligibility of narrowband telephone speech can be improved by artificial bandwidth extension (ABE), which extends the speech bandwidth using only information available in the narrowband speech signal. This paper reports a three-language evaluation of an ABE method that has recently been launched in several of Nokia's mobile telephone models. The method extends the speech bandwidth to frequencies above the telephone band by first utilizing spectral folding and then modifying the magnitude spectrum of the extension band with spline curves. The performance of the method was evaluated by formal listening tests in American English, Russian, and Mandarin Chinese. The results of the listening tests indicate that ABE processing improved the subjective quality of coded narrowband speech in all these languages. Differences between bandwidth-extended American English test sentences and their original wideband counterparts were also evaluated using both an objective distance measure that simulates the characteristics of human hearing and a conventional spectral distortion measure. The average objective error was calculated for different categories of speech sounds. The error was found to be smallest in nasals and semivowels and largest in fricative sounds.

The Role of Pause Cues in Language Learning: The Emergence of Event-related Potentials Related to Sequence Processing

Humans can derive sequential dependencies from unfamiliar artificial speech within several of minutes of exposure. However, there is an ongoing debate about the nature of the underlying learning processes. In a widely discussed study Peña et al. [Peña, M., Bonatti, L. L., Nespor, M., & Mehler, J. Signal-driven computations in speech processing. Science, 298, 604-607, 2002] argued for the importance of subtle acoustic cues in the signal, such as pauses, in order to switch between two computational mechanisms, which are conceptualized as rule-based versus statistical. The present study was aimed to approach this problem by recording event-related potentials in response to correct and incorrect phrases consisting of bisyllabics after short exposure to either rule-based or random artificial speech streams. Rule-based streams contained dependencies of the form AXC, whereby A elements reliably predicted the C elements and X elements were variable. Participants were exposed to four input and test phases. Two of the input streams were rule-based and contained either only probabilistic information related to the distribution of the AXC stimuli or an additional acoustic cue indicating the boundaries of relevant units. The other two streams were random variations of the rule-based streams. During the test phase in the condition with pause cues, an early negativity and a later positivity emerged for correct and incorrect items in comparison to their acoustically identical counterparts, which were presented after the random control condition. In the noncued condition, only negativities were seen. The timing and the scalp distribution of the negativities were different for correct and incorrect sequences in both the cued and the noncued conditions. The results are interpreted in support of a view of grammatical learning in which both distributional and acoustic cues may contribute to different aspects of syntactic learning.

Temporal Compression Of Speech: An Evaluation

Efficient browsing of speech recordings is problematic. The linear nature of speech, coupled with the lack of abstraction that the medium affords, means that listeners have to listen to long segments of a recording to locate points of interest. We explore temporal compression algorithms that attempt to reduce the amount of time users require to listen to speech recordings, while retaining the important content. This paper implements two main approaches to temporal compression: artificial speech rate alteration (speed-up) and unimportant segment removal (excision). We evaluate the effectiveness of these approaches by having listeners rate comprehension and listening effort for different types of temporal compression. For different compression levels, we compare performance of various implementations of speed-up and excision as well as techniques based on semantic features and acoustic features. Our results indicate that listeners prefer low compression levels, excision over speed-up, and algorithms based on semantic rather than acoustic features. Finally, listeners were negative about hybrid algorithms that used speed-up to indicate missing regions in an excised recording.

Speech in noise and the medial olivo‐cochlear efferent system

A central question in auditory scene analysis is how we are able to follow speech against a background of interfering noise. The question is particularly acute for artificial speech recognition algorithms and the hearing impaired. The medial efferent system has been suggested as one contributor to our ability to hear speech in noise. We have recently added efferent suppression to our model of the auditory periphery and evaluated it against physiological observations at the level of the basilar membrane and auditory nerve. We have also replicated a study using compound action potentials where a tone in a noise background became more salient when the efferent system was artificially activated. Visual representations of the computed auditory nerve response to speech in noise show considerable improvement when the efferent system is activated. Tests using the auditory model as a front end to a connected‐word recognition algorithm also showed improved performance in the presence of noise when efferent effects were included. The benefits of efferent suppression include reduced compression and extension of the dynamic range of individual auditory nerve fibers.

Robust Speech Recognition by Model Adaptation and Normalization Using Pre-Observed Noise

Users require speech recognition systems that offer rapid response and high accuracy concurrently. Speech recognition accuracy is degraded by additive noise, imposed by ambient noise, and convolutional noise, created by space transfer characteristics, especially in distant talking situations. Against each type of noise, existing model adaptation techniques achieve robustness by using HMM-composition and CMN (cepstral mean normalization). Since they need an additive noise sample as well as a user speech sample to generate the models required, they can not achieve rapid response, though it may be possible to catch just the additive noise in a previous step. In the previous step, the technique proposed herein uses just the additive noise to generate an adapted and normalized model against both types of noise. When the user's speech sample is captured, only online-CMN need be performed to start the recognition processing, so the technique offers rapid response. In addition, to cover the unpredictable S/N values possible in real applications, the technique creates several S/N HMMs. Simulations using artificial speech data show that the proposed technique increased the character correct rate by 11.62% compared to CMN.

Combined Gesture-Speech Recognition and Synthesis Using Neural Networks

Sign languages such as Spanish Sign Language (LSE) are the primary communication way among members of the Deaf community. However this language is not widely known outside of this community. The techniques for automatic recognizing hand signs proposed in this paper allow creating systems which can help deaf people to communicate with others, by providing them with computer tools for assisted communication or potentially with a fully automatic portable sign-language-to-speech translation system. The aim of this research it to implement a self-organizing neural network based technique for hand sign recognition, using self organizing map (SOM) and learning vector quantization (LVQ) based algorithms. The two classifiers are then combined to make the final decision. SOM and LVQ classifiers are training for the 26 signs of the manual alphabet and the speech synthesizer concatenate the different phones, diphones and triphones stored in the database to generate the right words in artificial speech. Finally, the system is to contribute to the implementation of meaningful human machine interactions in a workstation, mainly for welfare applications.

Methods and apparatus for rapid acoustic unit selection from a large speech corpus

A speech synthesis system can select recorded speech fragments, or acoustic units, from a very large database of acoustic units to produce artificial speech. The selected acoustic units are chosen to minimize a combination of target and concatenation costs for a given sentence. However, as concatenation costs, which are measures of the mismatch between sequential pairs of acoustic units, are expensive to compute, processing can be greatly reduced by pre-computing and caching the concatenation costs. Unfortunately, the number of possible sequential pairs of acoustic units makes such caching prohibitive. However, statistical experiments reveal that while about 85% of the acoustic units are typically used in common speech, less than 1% of the possible sequential pairs of acoustic units occur in practice. A method for constructing an efficient concatenation cost database is provided by synthesizing a large body of speech, identifying the acoustic unit sequential pairs generated and their respective concatenation costs, and storing those concatenation costs likely to occur. By constructing a concatenation cost database in this fashion, the processing power required at run-time is greatly reduced with negligible effect on speech quality.