Speech Signal Synthesis Research Articles

An Adaptive Method for Robust Detection of Vowels in Noisy Environment

Automatic detection of vowels plays a significant role in the analysis and synthesis of speech signal. Detecting vowels within a speech utterance in noisy environment and varied contexts is a very challenging task. In this work, a robust technique based on non-local means (NLM) estimation is proposed for the detection of vowels in noisy speech signals. In the NLM algorithm, the signal value at each sample point is estimated as the weighted sum of signal values at other sample points within a search neighborhood. The weight value is computed by finding square of the difference between the signal values belonging to two different segments. During the estimation, one segment is kept as fixed, while other segment is slid over the search neighborhood. For any particular sample point, the sum of those weight values is significantly less when the segments under consideration are higher in magnitude. In a given speech signal, the vowels are regions of high energy. This will be true even under noisy conditions. In this work, the sum of weight values (SWV), computed at each time instant is used as a discriminating feature for detecting the vowels in a given speech signal. In the proposed approach, the regions where the SWV exhibits significant transitions and attain lower values for a considerable duration of time compared to the preceding and succeeding regions are hypothesized as the vowels. This hypothesis is statistically validated for detecting vowels under clean as well as noisy test conditions. For proper comparison, a three-class statistical classifier (vowel, non-vowel and silence) is developed for detecting the vowels in a given speech signal. For developing the said classifier, the mel-frequency cepstral coefficients are used as the acoustic feature vectors, while deep neural network (DNN)-hidden Markov model (HMM) is employed for acoustic modeling. The proposed vowel detection method is observed to outperform the DNN-HMM-based statistical classifiers as well as existing signal processing approaches under both clean and noisy test conditions.

Автоматичне розпізнавання, розуміння та синтез мовленнєвих сигналів в Україні

Автоматичне розпізнавання, розуміння та синтез мовленнєвих сигналів в Україні

Modeling of Phoneme Transitions for Natural Synthesis of Speech Signals

Modeling of Phoneme Transitions for Natural Synthesis of Speech Signals

A survey on speech signal synthesis system

A survey on speech signal synthesis system

Study of local EEG specificities in children with mental development disorders using independent component analysis

Baseline EEGs in the frequency range of 3–13 Hz in children with mental disorders of perinatal origin during wakefulness with the eyes open were analyzed using independent component analysis. In cases of severe mental retardation, a significant increase in the power density spectra of the θ band was revealed in the left-sided frontotemporal and right-sided temporal cortices, which allows us to consider these regions to be putative sources of slow activity and markers for a lesion or immaturity in the fronto-thalamic system, as well as for the temporal areas responsible for the auditory analysis and synthesis of speech signals and the integration of audio-visual information.

Extraction of formant bandwidths using properties of group delay functions

Formant frequencies represent resonances of vocal tract system during the production of speech signals. Bandwidths associated with the formant frequencies are important parameters in analysis and synthesis of speech signals. In this paper, a method is proposed to extract the bandwidths associated with formant frequencies, by analysing short segments (2–3ms) of speech signal. The method is based on two important properties of group delay function (GDF): (a) The GDF exhibits prominent peaks at resonant frequencies and (b) the influence of one resonant frequency on other resonances is negligible in GDF. The accuracy of the method is demonstrated for synthetic signals generated using all-pole filters. The method is evaluated by extracting bandwidths of synthetic signals in closed phase and open phase regions within a pitch period. The accuracy of the proposed method is also compared with that of two other methods, one based on linear prediction analysis of speech signals, and another based on filterbank arrays for obtaining amplitude envelopes and instantaneous frequency signals. Results indicate that the method based on the properties of GDF is suitable for accurate extraction of formant bandwidths, even from short segments of speech signal within a pitch period.

Integrating speaker-nonspecific timbre transformation to an HNM-based speech synthesis scheme

In this article, a speech signal synthesis scheme based on harmonic-plus-noise model (HNM) is extended to provide the function of speaker-nonspecific timbre transformation. To transform synthetic speech's timbre, we have developed a formant-based frequency mapping method called piece-wise linear frequency mapping (PLFM). Additionally, a commonly adopted method is frequency axis scaling (FAS). Both methods have been integrated into the HNM-based signal synthesis scheme, and a real-time speech synthesis system has been implemented according to this scheme. Using the speech files synthesized, perception tests are conducted. The results show that the proposed scheme can indeed transform the source timbre of a female adult into the timbre of a male adult, boy, or girl. In addition, the method PLFM is shown to be better than FAS in obtaining more masculine timbre.

A Three-Level Speech, Music, and Mixture Classifier

Due to the new techniques of analysis and synthesis of speech signals, the musical signal processing has gained particular weight, and therefore, the classical sound analysis techniques are used in processing music signals. In this paper, a general three-level algorithm for discriminating three general types of signals: speech, music, and mixture of both signals is proposed. The proposed classifier is based on the well-known zero-crossing rate (ZCR) [1] and short-time energy (STE) [2,3] algorithms, with the positive derivative and voting procedures. Some theoretical analysis and simulation has been applied to a library of different types of musical and speech signals to show the performance of the proposed algorithm.

Identification and Synthesis of Speech Signal Using Multi-Mother Wavelet.

In production and management of various plants, it is important to provide an effective human-machine interface for communication between human and machine system. Of all communication approaches human speech signal is the most natural and effective. To compose the human-machine interface of the speech, it is necessary to develop a method with high accuracy and high quality for identification and synthesis of the speech. In this paper, a new approach for speech identification and synthesis using multi-mother wavelet is presented. The optimum mother wavelet can be searched by using multi-mother wavelet method for the speech identification and synthesis. Through the case study, it is concluded that the efficient speech identification and synthesis can be realized by the proposed approach. In future the presented speech identification and synthesis technology is expected to be applied to various cases such as plant production process and control.

Method of speech synthesis by means of concentration and partial overlapping of waveforms

Method for speech signal synthesis by means of time concatenation of waveforms representing elementary units of speech signal, in which: at least the waveforms associated to voiced sounds are subdivided into a plurality of intervals, corresponding to the responses of the vocal duct to a series of excitation impulses of the vocal cords, synchronous with the fundamental frequency of the signal; each interval is subjected to a weighting; the signals resulting from the weighting are replaced with a replica thereof shifted in time by an amount that depends on a prosodic information; and the synthesis is carried out by overlapping and adding the shifted signals. In each interval of original signal to be reproduced in synthesis, an unchanging part is identified, which contains the fundamental information and which is reproduced unaltered in the synthesized signal, and the operations of weighting, overlapping and adding involve only the remaining part of the interval.

Pitch-synchronous wavelet representations of speech and music signals

A new wavelet representation is explored. The transform is based on a pitch-synchronous vector representation and it adapts to the oscillatory or aperiodic characteristics of signals. Pseudo-periodic signals are represented in terms of an asymptotically periodic trend and aperiodic fluctuations at several scales. The transform reverts to the ordinary wavelet transform over totally aperiodic signal segments. The pitch-synchronous wavelet transform is particularly suitable to the analysis, rate-reduction coding and synthesis of speech signals and it may serve as a preprocessing block in automatic speech recognition systems. Feature extraction such as separation of voice from noise in voiced consonants is easily performed by means of partial wavelet expansions. A stochastic model of aperiodic fluctuations is proposed. >

Adults and infants show a “prototype effect” for speech sounds

Dennis Klatt provided valuable assistance to many speech perception researchers. His advice on the synthesis of speech signals was particularly helpful, and investigators relied on his expertise when problems arose in the development of stimuli designed to approach a problem in a new way. This was the case in the preparation of stimuli to test whether there are “prototypes” for vowel sounds. The experimental question was whether or not adults and infants responded differently in a within‐category vowel discrimination task when the “standard” stimulus was an exceptionally good instance of the vowel /i/—a prototype of the category—as opposed to a nonprototypic /i/ vowel. The results showed that when the prototype of the category served as the standard stimulus, it was more difficult to hear differences between it and novel /i/ vowels than it was to hear differences between the nonprototypic stimulus and novel /i/ vowels. In other words, the prototype was perceived to be more similar to new members of the category than was the nonprototypic stimulus. The effect was observed both in adults and in 6‐month‐old infants. Described first will be the hypothesis underlying the prototype test, the method used to construct the stimuli (the test stimuli for the prototype and the nonprototype were scaled psychophysically using the “mel” scale), and the results on English adults and infants. Described next will be the second phase of the research program, which entails cross‐language tests on Swedish adults and infants. The cross‐language tests are designed to assess two different explanations for the prototype effect: (1) that particular vowel stimuli (/a,i,u/) are inherently more resistent to the effects of articulatory/acoustic change (Stevens' quantal theory) and (2) that the effects observed in American adults and infants are attributable to experience in listening to English, even in the first 6 months of life. Careful preparation of the English and Swedish stimuli was critical in designing these experiments. Dennis Klatt was always ready to provide advice on such matters; his assistance is gratefully acknowledged and sorely missed. [Work supported by NIH.]

Method and apparatus for the analysis and synthesis of speech signals

Synthesized speech is produced by a vocal tract model corresponding functionally to the human vocal tract and constructed of a linear digital filter. The parameter of the synthesis vocal tract model are determined by an analysis operation on an original speech signal using an identical vocal tract model, which may be the same model as used for synthesis. The analysis vocal tract model has its parameters adjusted according to a comparison between the original speech signal and the output signal of the analysis model so as to minimize the deviation between these two signals. Those parameters for which the deviation falls below a threshold value are used directly as parameters of the synthesis vocal tract model. The adjustments of the parameters are determined by a parameter computer working on the results of the aforementioned comparison and which itself may contain the vocal tract model.

Vocoders: Analysis and synthesis of speech

Techniques for analysis and synthesis of speech signals are reviewed with emphasis on vocoders and related devices for more efficient transmission and storage of speech. Selected applications of speech coding methods as sensory aids to the handicapped are described.