Parameters Of Speech Signal Research Articles

Enhancement of performance parameters of speech signal using model order reduction approach

In this paper, a model order reduction approach is used with the aim to enhance the performance parameters of the speech signal. Initially, this model reduction technique is applied on the original higher order system for the reduction of the order of system while keeping all the features of the original one in it. The performance parameters such as SNR, NRMSE and SD for the reduced order system have been improved compared to the same performance parameters with original higher order one. Finally, the comparisons of these performance parameters are made to show to enhance the performance of reduced order system irrespective of the order and values of state space parameters.

Reduction of Speech Signal Deformation in Patients after Nasal Septum Surgery (Septolplasty)

Nasal blockage belongs to the most common symptoms of nasal diseases in vocal tract area. At the same frequency there appear acoustic symptoms, existing as the change of human voice color. Vocal and articulation disorders of the ear, nose ane throat are usually observed in the form of closed rhinolalia and this observation can be performed both by patients and other listeners as well. Nasal polyps and nasal septum deviation are frequent reason of nasal blockage connected in consequence with decreased nasal ventilation. One of the main principles of the surgical treatment performed in mentioned situations is the restoration of nasal patency. The evaluation of the influence of nasal surgery on intensification of acoustic symptoms depends on verification of parameters of the human speech signal, so it was necessary to apply objective methods. That allowed to combine results of acoustic analysis with patient’s subjective feeling and rhinomanometric evaluation of nasal patency. The main purpose of this research was to objectively evaluate the influence of surgical treatment improving nasal patency on deformation of the voice of operated patients.

Adoption of Neural Networks to Classified the Gender of the Speaker

In this research the neural network was adopted to classified the gender of the spoken, by creating the two dimension matrix from the parameters of the spoken speech signal which normal was snigle dimension array. The porpose algorithm in this research divided in two stage :- In the first stage the seven moment were calculated for a set of spoken signal of 50 persons , to be followed creating database depend on the seven moments .This database will be used to find the threshould value for both genders (male/female) which will be trained by neural network to classify any input tothe network. In the second stage , speech of any spoken will be selected and the same feature will be extracted , as in the first stage , to be used as input to the neural network which was traind previously for gender recognition. Back propagation neural network was achieved for recognition. The result of the applied algorithem on 10 spoken passed on 8 of them and 2 of them was failed .

Monitoring the functional state of human operators by the parameters of the fundamental tone and spectrum of speech

The possibility of distant monitoring of the functional state of a human operator by speech signal parameters is considered. Examples of this monitoring in pilots during simulated and real flights are presented.

Hidden Markov model-based packet loss concealment for voice over IP

As voice over IP proliferates, packet loss concealment (PLC) at the receiver has emerged as an important factor in determining voice quality of service. Through the use of heuristic variations of signal and parameter repetition and overlap-add interpolation to handle packet loss, conventional PLC systems largely ignore the dynamics of the statistical evolution of the speech signal, possibly leading to perceptually annoying artifacts. To address this problem, we propose the use of hidden Markov models for PLC. With a hidden Markov model (HMM) tracking the evolution of speech signal parameters, we demonstrate how PLC is performed within a statistical signal processing framework. Moreover, we show how the HMM is used to index a specially designed PLC module for the particular signal context, leading to signal-contingent PLC. Simulation examples, objective tests, and subjective listening tests are provided showing the ability of an HMM-based PLC built with a sinusoidal analysis/synthesis model to provide better loss concealment than a conventional PLC based on the same sinusoidal model for all types of speech signals, including onsets and signal transitions

Mobile station with audio signal adaptation to hearing characteristics of the user

A method and apparatus for increasing the intelligibility of speech signals received by mobile stations via adapting some of the acoustic parameters of speech in the frequency domain. The mobile station may have the capability to modify speech signal parameters to conform the speech signal to a listener's hearing profile. In an alternative embodiment, mobile station processing and battery power are conserved by the communication network having the capability to modify speech signal parameters to conform the speech signal to a listener's hearing profile. To conserve mobile station memory, the user's hearing profile may be stored on the network.

A soft voice activity detector based on a laplacian-gaussian model

A new voice activity detector (VAD) is developed in this paper. The VAD is derived by applying a Bayesian hypothesis test on decorrelated speech samples. The signal is first decorrelated using an orthogonal transformation, e.g., discrete cosine transform (DCT) or the adaptive Karhunen-Loeve transform (KLT). The distributions of clean speech and noise signals are assumed to be Laplacian and Gaussian, respectively, as investigated recently. In addition, a hidden Markov model (HMM) is employed with two states representing silence and speech. The proposed soft VAD estimates the probability of voice being active (VBA), recursively. To this end, first the a priori probability of VBA is estimated/predicted based on feedback information from the previous time instance. Then the predicted probability is combined/updated with the new observed signal to calculate the probability of VBA at the current time instance. The required parameters of both speech and noise signals are estimated, adaptively, by the maximum likelihood (ML) approach. The simulation results show that the proposed soft VAD that uses a Laplacian distribution model for speech signals outperforms the previous VAD that uses a Gaussian model.

Comparison between physical and auditory parametrization of speech corpus for the unit selection in text-to-speech synthesis

It is important to design an appropriate cost function to improve the quality of speech produced by a corpus-based text-to-speech (TTS) synthesis. Although the final product of the TTS system is evaluated perceptually, the definition of cost functions has to be based on the physical parameters of speech signal. And a cost function will become inappropriate if the discrepancy between the physical and the perceptual measures becomes critically large. The discrepancy could be caused by two main factors, i.e., (1) inappropriate selection of parameters, and (2) inappropriate scaling. One may judge that the selection of physical parameters is inappropriate if changes in a single physical parameter result in changes in multiple perceptual attributes. One may also judge that the scaling of physical dimension is inappropriate if the mapping function of the physical space to the perceptual space appears to be critically nonlinear. Aiming at a better definition of cost functions, a signal analysis based on the standard linear prediction coding (LPC) method and one based on the models of the peripheral auditory system (DSAM-AIM) are compared. The results of principal component analysis for Japanese speech databases of a corpus-based TTS system, CHATR, are presented.

Neurobiology of speaking and stuttering

A multi-level model of speech motor control including the word, syllable and phonem level is proposed, which takes into account linguistic, technical, biological, physiological and developmental aspects. Crucial point is the double function of the vowel onset, which serves to control vowel length and also to synchronize the serialization of words into temporal strings of syllables. In (long) stressed syllables this is done via auditory feedback (re-afferent control), while the utterance of unstressed syllables is automatized using internal signals (efferent control). Young children's problem when crossing from infantil to adultlike speech is to learn to inhibit re-afferent and to apply efferent control of unstressed syllables. If this task is learned incompletely, the model predicts stuttering, e.g. speech blocks, or repetitions and prolongations of syllables or phonems. Because all these control and learning processes refer to low level motor control, the model also contributes to unburden persons who stutterer and their parents from personal responsibility for the occurence of this communication disorder. FACTS, THEORIES AND BELIEVES CONCERNING STUTTERING Basic symptoms of stuttering are repetitions of words or syllables or parts of them, prolongations of speech sounds (phonems), and blockades. All of young children aged between two and five years use to have normal dysfluencies (Yairi 1981, 1997, Froschels 1952), but about 5% of the children show the basic symptoms of stuttering mentioned above. Many of the children lose these dysfluencies until puberty. In the end, about 1% of the adult population exhibits, regardless of culture and language, persistent speech dysfluencies, now three times more male than female people. Stuttering often is detrimental for personal achievement, social contacts, carreer, and self-esteeme, therefore, almost all people who stutter participated in at least one therapy, whereby, however, stuttering often cannot be cured completely, or re-appears later on. Explanations of the symptoms of stuttering usually start from investigations referring to sensory feedback of self produced speech, especially to the effects of delayed auditory feedback (Lee 1951). Usually, nonstuttering persons exhibit repetitions of syllables and words under this experimental manipulation, whereby delay times between 100 and 300 ms are most effective. Paradoxically, symptoms diminish in stutterers under delayed auditory feedback, at least at the beginning of application (Bergmann 1987). Occasionally, also auditory masking by loud noise applied through headphones facilitates fluent in stutterers (Bergmann 1987). Fluency enhancement in stutterers is also reported when using frequency shifted auditory feedback (Howell et al 1987, Natke & Kalveram, in press). Because of such studies, it is concluded (e.g. Van Riper 1982, Perkins et al. 1976, Adams 1974) that during speech development the auditory feedback channel gradually loses influence on speech control, until, in adults, only tactile-proprioceptive feedback is used for purposes of control. Speech dysfluencies, then, originate from auditory feedback not sufficiently being suppressed. This may lead to interferences and/or discoordinations between different feedback channels, which disturbes the auto-regulation of speaking (Fiedler & Standop 1992). However, experiments of Bauer et al. (1997) do not support interference models of stuttering. In these experiments, irregular tactile and proprioceptive feedback was artificially generated through mechanical disturbations of the jaw movement during speaking. These disturbations neither led to significant changes of speech relevant parameters of the acoustical speech signal, nor differentiated between stutterers and non-stutterers. Furthermore, experiments with manipulated auditory feedback (see also the chapter concerning audiophonatory coupling) show that the auditory feedback channel is not at all switched off during speaking, neither in stutterers nor in fluent speakers. Therefore, it can be argued that auditory, not proprioceptive, feedback is of particular importance for speech flow control, 1 Supported by grants Ka417/13-1 to -3 from Deutsche Forschungsgemeinschaft (DFG)

On robust Kalman filtering with forgetting factor for sequential speech analysis

We propose a robust Kalman filter with forgetting factor to estimate the time-varying parameters of speech signals. The proposed robust Kalman filter is based on a modified least-squares criterion with forgetting factor. The input signal is assumed to have a heavy-tailed non-Gaussian nature with outliers due to spiky excitation. To alleviate the effects of outliers, this algorithm extends the concept of Huber's min-max approach, named M-estimation, to the Kalman filtering. The introduction of forgetting factor enables the time-varying speech parameters to be estimated, giving more weight on the most recent portion of the data. Experimental results show that the proposed algorithm achieves more accurate estimation and provides improved tracking performance.

Wideband speech and audio coding

Typical parameters of wideband speech and audio signals, including digitized versions of each, potential applications, and available transmission media, are described. Facts about human auditory perception that are exploited in audio coding and quality measures that play an important role in coder evaluations and designs are reviewed. Techniques for efficient coding of wideband speech and audio signals, with an emphasis on existing standards, are discussed. The audio coding standard developed by the Moving Pictures Expert Group within the International Organization for standardization (ISO/MPEG) is covered in some detail, since it will be used in many application areas, including digital storage, transmission, and broadcasting of audio-only signals and audiovisual applications such as videotelephony, videoconferencing, and TV broadcasting. Ongoing research and standardization work is outlined. >

Timing deficits in apraxia of speech.

This paper deals with a particular aspect of speech motor control in patients suffering from apraxia of speech. Three experiments are reported concerning the phase relations between individual speech gestures. These include the timing of laryngeal, velar and labial movements relative to lingual gestures. A total of 8 patients and 12 normal controls were examined using speech material which was designed according to appropriate phonetic paradigms. Evaluation was performed on the basis of speech signal parameters referring to the kinematics of inter-articulatory phasing. Deviations of the patient group were found in all three experiments. This suggests that disturbed phase relations of individual speech movements are a general feature of apraxic speech. It is further hypothesized that the described motor symptoms are the origin of a variety of phonemic errors. Support for this view is provided by appropriate examples which refer to the examined paradigms. By this argument, much of the disturbed phonemic structure of apraxic speech may be accounted for by timing deficits.

A Lattice Structured Algorithm for Recursive Solution of the Extended Yule-Walker Equations

Abstract Estimation of LPC (AR) parameters of speech signals in additive noise can be approached as the estimation of the AR parameters of an ARMA process. A recursive algorithm is presented that estimates the AR parameters without noise bias. This is based on the Toeplitz extended Yule-Walker equations. Using the Trench algorithm, a lattice structure is developed, leading to a gradient-type algorithm which may be steepest ascent or steepest descent.

Spatio‐temporal patterns of speech evoked activity in the auditory nerve

We examine the speech evoked responses of large populations of auditory‐nerve fibers in the cat [M. Miller and M. Sachs, J. Acoust. Soc. Am. Suppl. 1 70, S9 (1981)]. Three‐dimensional pictures of the spatio‐temporal patterns of activity are constructed by spatially ordering the display of the fibers' PST response histograms according to the CF of each fiber. The distribution of the relative phases of synchronized activity emerges as an important feature in the response patterns. Specifically, each strong low‐order harmonic component of the stimulus tends to synchronize the activity of a relatively broad segment of fibers in a pattern which mirrors the underlying traveling wave. The sphere of influence of each such harmonic can be roughly subdivided into two regions: (1) a region basal to the point of resonance where the fiber PST histograms accumulate only small delays (or phase shifts) relative to each other reflecting the fast speed of propagation of the traveling wave. (2) A region at or very near the point of resonance where the responses exhibit drastic relative phase shifts owing to the sudden slow down of the traveling wave and the consequent rapid acumulation of phase shifts. These rapid phase shifts thus appear to disrupt an otherwise spatially uniform response pattern at approximately the CFs corresponding to the low harmonic components (< 1.5–2 kHz) and thus may play a helpful role in the processing and extraction of the speech signal parameters by the central auditory system.

A Signal Processor for a Multiple-electrode Hearing Prosthesis

A 22-electrode implantable hearing prosthesis uses a wearable speech processor which estimates three speech signal parameters. These are voice pitch, second formant frequency and flattened spectrum amplitude. The signal is monitored continuously for periodicity in the range 80-400 Hz and, if this is present, stimulation occurs at the same rate. Otherwise, as in the case of unvoiced sounds, it occurs at the random rate of fluctuation of the signal envelope. The second formant is obtained by filtering to extract the dominant peak in the midband region and by continuous measurement of the zero crossing rate. The amplitude measured is that of the whole speech spectrum, pre-emphasized by differentiation. The values that are presented to the patient are the parameter estimates immediately prior to the stimulation pulse. Second formant frequency is coded by selection of an appropriate electrode in the cochlea and amplitude by a suitably controlled current. Automatic gain control is used to keep the dynamic range of the amplitude estimate within the 30 dB range of the circuitry.

Auditory recognition of intonation patterns

Intonation is considered to be represented by a pitch pattern. This pattern, however, is not obtainable in a traditional psychophysical experiment. It can be obtained only in an acoustico-linguistic experiment, since physical parameters of speech signals are perceived in a special manner. Two methods can be considered for detecting the features used in the auditory recognition of intonation patterns: one is to collect all minimal contrasts in the language and find the smallest number of patterns necessary to distinguish all the contrasts; the other is to systematically transform the experimentally obtained patterns until a sufficiently large number of contrastive (not necessarily minimal) intonations are all distinguished. In such signals as intonation, which are extremely difficult to classify into minimal contrasts, the second method seems to be the only alternative. In this report, such a method of determining the auditorily perceived pattern of intonation is described. This study was conducted by simulating the auditory recognition of intonational signals by a computer. The simulated analysis included the extraction, normalization, and segmentation of the physical parameter, F0, and transformation and classification of processed F0 into intonation patterns.

Procedure for time-domain segmentation of connected speech

The segmentation procedure subdivides the speech signal into discrete successive sections whose sequence permits localization of most of the spoken phonemes in connected speech. In two preliminary steps, the pauses and the voiceless sections are separated from the rest of the signal. Subsequently, the main segmentation procedure tries to describe the speed of the vocal-tract movement. Since, for an individual phoneme, the vocal tract attempts to realize the accompanying “target position” and to keep it for a certain time before approaching the next phoneme rather than moving in a constant way, sections with little change in vocal-tract position (“stationary” segments) can be separated from ones with major, faster changes (“dynamic” segments). A small number of crude speech signal parameters (such as level) have proved sufficient to run this algorithm. By analyzing the direction of change within the dynamic segments, these segments can be further subdivided into “monotonously transient” sections. Except for a slight adjustment of the parameter change thresholds used when defining a “major” change, the algorithm doesn't need an adaptation to an individual speaker.

Parameter estimation in speech: A lesson in unorthodoxy

The accurate estimation of both discrete and continuous parameters of speech signals has played a central role in speech processing and research. Interestingly, the most successful estimation procedures have often relied on intuition based on knowledge of speech signals and their production in the human vocal apparatus rather than routine applications of well-established theoretical methods. This experience in speech processing should serve as a reminder that a thorough understanding of the signal is paramount for successful analyses of real-world processes.