Segmentation Of Speech Signal Research Articles

Segmentation of specific speech signals from multi-dialog environment using SVM and wavelet

In this paper, a novel multi-speaker segmentation technique is presented. This technique makes use of wavelets and support vector machines (SVMs) to segment specific speakers’ speech signals from multi-dialog environments. The proposed method first applies wavelets to determine the acoustical features such as subband power and pitch information from a given multi-dialog speech data. Then the multi-speaker segmentation of the given multi-dialog speech data can be accomplished by the use of a bottom–up SVM over these acoustical features and additional parameters, such as frequency cepstral coefficients. A public audio database, Aurora-2, is used to evaluate the performances of the proposed method. Experimental results show that the accuracy of multi-speaker segmentation is 100% achieved in the combination of two speakers. And the segmental accuracy can achieve at least 94.12% and 85.93% for 4-speaker and 8-speaker conditions, respectively.

Isolated-word segmentation based on the dynamics of the parameters of short correlation functions

The theoretical and experimental results obtained in an investigation of the dynamic behavior of short-correlation-function parameters and applied to the problem of speech-signal segmentation are described. The side maxima of the correlation functions of various quasi-periodic signals are investigated theoretically. Their dependences on the characteristics that determine the degree of signal quasi-periodicity are presented. These dependences are used to construct instrumental estimates of the degree of such parameters as the quasiperiodicity, the duration of quasi-periodicity segments, and the fundamental-tone period. The obtained estimates are interpreted for the case of voice signals, and it is shown that many important procedures used in speech technologies can be implemented on the basis of these estimates.

Alaryngeal Speech Enhancement Using Pattern Recognition Techniques

An alaryngeal speech enhancement system is proposed to improve the intelligibility and quality of speech signals generated by an artificial larynx transducer (ALT). Proposed system identifies the voiced segments of alaryngeal speech signal, by using pattern recognition methods, and replaces these by their equivalent voiced segments of normal speech. Evaluation results show that proposed system provides a fairly good improvement of the quality and intelligibility of ALT generated speech.

Evaluation of formant-like features on an automatic vowel classification task.

Numerous attempts have been made to find low-dimensional, formant-related representations of speech signals that are suitable for automatic speech recognition. However, it is often not known how these features behave in comparison with true formants. The purpose of this study was to compare two sets of automatically extracted formant-like features, i.e., robust formants and HMM2 features, to hand-labeled formants. The robust formant features were derived by means of the split Levinson algorithm while the HMM2 features correspond to the frequency segmentation of speech signals obtained by two-dimensional hidden Markov models. Mel-frequency cepstral coefficients (MFCCs) were also included in the investigation as an example of state-of-the-art automatic speech recognition features. The feature sets were compared in terms of their performance on a vowel classification task. The speech data and hand-labeled formants that were used in this study are a subset of the American English vowels database presented in Hillenbrand et al. [J. Acoust. Soc. Am. 97, 3099-3111 (1995)]. Classification performance was measured on the original, clean data and in noisy acoustic conditions. When using clean data, the classification performance of the formant-like features compared very well to the performance of the hand-labeled formants in a gender-dependent experiment, but was inferior to the hand-labeled formants in a gender-independent experiment. The results that were obtained in noisy acoustic conditions indicated that the formant-like features used in this study are not inherently noise robust. For clean and noisy data as well as for the gender-dependent and gender-independent experiments the MFCCs achieved the same or superior results as the formant features, but at the price of a much higher feature dimensionality.

Recognition of speech from temporal patterns

Spectrum of short segments of speech signal is accepted as one of the prime acoustic correlates of phonetic quality and it forms the basis of most current feature representations of speech. However, the short-term spectral vectors are affected by coarticulation. The coarticulation can span time intervals of several phonemes. In spite of that, human speech perception is able to derive the identity of the individual underlying phoneme. This process does not seem to require successful decoding of neigboring phonemes [Allen, IEEE Trans. AU-2(4), 567–577 (1994)], but it may require information from a relatively large temporal span of the speech signal [Kozhevnikov and Chistovich, ‘‘Speech: Articulation and Perception,’’ Dept. of Commerce (1967)]. To allow for effective use of a syllable-level temporal information, a short-term spectrum-based vector is replaced as input to the classifier by rather long (about 1 s) vectors representing the time trajectory of critical-band spectral energy. Each temporal vector then spans many more than one phoneme. This technique forms the basis of a complete ASR system which is competitive with the current state-of-the-art speech recognition technology. [Work supported by the Department of Defense and by the National Science Foundation.]

Robust pitch estimation method and device for telephone speech

The present invention provides a pitch estimating method and device for accurately estimating the pitch of digitized speech signals, in spite of the presence of contaminants and distortions in telephone speech signals by (1) determining a set of pitch candidates to estimate a pitch of the digitized speech signal at each of a plurality of time instants, wherein series of these time instants define segments of the digitized speech signal; (2) constructing a pitch contour using a pitch candidate selected from each of the sets of pitch candidates determined in the first step; and (3) selecting a representative pitch estimate for the digitized speech signal segment from the set of pitch candidates comprising the pitch contour.

Selective application of speech coding techniques to input signal segments

A speech coding method and apparatus which selectively applies speech coding techniques to time segments of speech information signals, such as, e.g., pitch cycle waveforms is disclosed. A speech information signal comprising N signal segments is coded with a first speech coder to provide a first coded representation for each of the N signal segments. A second speech information signal reflecting speech information not coded by the first coder is determined for each of one or more of the N signal segments. In addition to coding the N first speech information signal segments with the first speech coder, M of the second speech information signals are coded with a second speech coder, where 1≦M≦N-1. The selective coding of M of the second speech information signals is done responsive a coding criterion. By selective use of the second speech coder, the number of bits needed to represent speech information may be reduced, or alternatively, better performance may be obtained without an increase in bit rate. The first and second speech coders may be any of those known in the art.

Speech analysis method and device for supplying data to synthesize speech with diminished spectral distortion at the time of pitch change

A method for speech analysis applicable to a speech analysis/synthesis system employed for producing a synthetic speech. Voiced and unvoiced segments of input speech signals X(n) are discriminated. An amplitude information A(ω) and a phase information P X (ω) are extracted from the voiced segments of the input speech signals. A pitch period is detected from the voiced segments of the input speech signals. A pulse train S(n) as a sound source information is generated so that its period corresponds on the time scale to the detected pitch period of the input speech signals. A phase information P S (ω) is extracted from the pulse train S(n). A difference P(ω) between the phase information P S (ω) of the pulse train S(n) and the phase information P X (ω) of the input speech signal is found and is supplied as the phase information of the desired one-pitch period within the input speech signals.

Sequential speech segmentation based on the spectral ARMA transition measure

Sequential segmentation algorithms based on the AR model tend to produce false alarms or to omit the change for sequences that corresponds to the ARMA model. In this paper a new sequential segmentation algorithm based on the ARMA model is presented. The ARMA model is estimated over the relatively short sequence, which has called for the implementation of the estimation algorithm with appropriately initialized starting values. The proposed algorithm adopts the MGLR concept of the sliding reference and test windows, which allow the process of decision making to be separated from the evaluation of the discrimination function. This has enabled the new triangular decision rule to be proposed; this is based on the expected shape of the discrimination function at the time of the model change. Two possible discrimination functions have been suggested. One of them is optimal in the statistical sense; the other has the better asymptotic behavior. Natural speech signal segmentation is also discussed, and an appropriate pitch-synchronous signal prearrangement has been suggested. This not only enhances the segmentation algorithm but also increases its speed, as the time can be increased by a step equal to the pitch period. The segmentation algorithm is verified on test signals as well as on the natural speech signal. The experimental results also include a comparison of the sequential AR and ARMA model-based segmentation.

Speech-recognition circuitry employing nonlinear processing, speech element modeling, and phoneme estimation

A phoneme estimator in a speech-recognition system includes energy detect circuitry for detecting the segments of a speech signal that should be analyzed for phoneme content. Speech-element processors then process the speech signal segments, calculating nonlinear (powers and products) representations of the segments. The nonlinear representation data is applied to speech-element modeling circuitry which reduces the data through speech element specific modeling. The reduced data are then subjected to further nonlinear processing. The results of the further nonlinear processing are again applied to speech-element modeling circuitry, producing phoneme isotype estimates. The phoneme isotype estimates are rearranged and consolidated, that is, the estimates are uniformly labeled and duplicate estimates are consolidated, forming estimates of words or phrases containing minimal numbers of phonemes. The estimates may then be compared with stored words or phrases to determine what was spoken.

The periodicity bias

The evidence presented in the target articles suggests that infants' phonetic development is driven by the acquisition of words as holistic units; and that segmental representations, previously considered to become language-specific only late in the first year of life, actually arise throughout this first year, with language-specific representations of vowels arising earlier than of consonants. We propose that these circumstances are unified by a bias towards attention to periodic sounds, present at birth. This equips the child to exploit linguistic rhythm to achieve initial segmentation of continuous speech signals to extract word units; the use of linguistic rhythm for segmentation is still present in adult processing. It also has the effect of making vowels achieve effective prototypical representations earlier than consonants.

A new statistical approach for the automatic segmentation of continuous speech signals

A statistical approach for the segmentation of a continuous speech signal to detect acoustic events is presented. Experiments are carried out to test the segmentation algorithms. Reasonable results are obtained with speech signals, although these are not exactly piecewise stationary. A comparison between the experimental results of automatic and handmade segmentations, demonstrates the potential acoustic-phonetic classification capability of the proposed algorithms.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Hearing aid method and apparatus

Method and apparatus for aiding hearing are disclosed comprising an automatic gain control unit which includes a first section for increasing the amplitude of input signal segments that are below a threshold level, and a second section for reducing the amplitude of input signal segments that are above said threshold level. Both sections have short attack and long release time constants. A long attack short release time constant noise suppressor unit responsive to the output from the second section and having a threshold level of operation below that of the threshold level of the automatic gain control units passes speech signals and squelches background noise signals between speech signal segments for improved understandability of speech signals.

Automatic segmentation of continuous speech signals

A statistical approach of the automatic segmentation of the speech signal is discussed. The purpose is to detect acoustic events which reveal articulatory changes as voice or frication onset and termination, closure, release... and formantic variations. The main idea is to model the signal by a statistical model (AR, ARMA) and to use test statistics (generalized likelihood, statistics of cumulative sum type) to detect sequentially abrupt changes in the parameters of the model. In the three segmentations which are presented here, the identification and testing procedures are sequential and monitored after every sample to obtain a better precision of change time estimations. The results obtained by each one are similar and speaker-independent. The detected acoustic events define interesting infra-phonemic units.

Segmentation of speech signals by application of methods of hierarchical classification

A multi‐level segmentation of a one‐dimensional signal may be induced by hierarchical ordering of subsets in a corresponding parameter space. This concept has been introduced to design a segmentation algorithm that creates a special three‐level segmentation for the speech signal. For preclassification it uses the parameters short time prediction gain and short time variance to form the second level of a segment hierarchy containing the classes “pause,” “fricative,” “vocal,” and “nasal oriented.” By merging the segment classes “pause” and “fricative” as well as “vocal” and “nasal oriented” the first level is formed. Since the vocal parts comprise more than 50% of speech, a clustering procedure has been added to create a third level containing four classes that roughly correspond to four different vowel classes. The parameter vector for the clustering algorithm is the sampled LPC‐generated log power spectrum together with the L2 distance. Five samples of speech, with a duration of 1 min each, have been processed. The resulting segmentation served as a basis for a number of segment length statistics which suggest applications in speaker verification and speech coding. [Work supported by VW Foundation, at the University of Hannover, Germany.]

Speech sounds and speech signals--the perspective of a new phonometry (author's transl)

Descriptive phonetics and instrumental phonetics have two entirely different empirical domains as their subject matters. On this basis, ZWIRNER's central problem of the theoretical relationship between sound category and sound realisation is re-formulated and discussed. It is shown how the traditional descriptive procedures of segmentation and decomposition can be connected with the segments of digital speech signals and the features analysed from them.

Segmentation of speech utilizing the time‐domain properties of speech signals

One of the most difficult problems in speech analysis is the segmentation of speech signals. In this paper, we describe a pitch‐detection algorithm which mainly utilizes the time‐domain properties of speech signals and a segmentation method based on the pitch‐detection algorithm. The segmentation is carried out in three phrases. First, by using the zero‐crossing rate and the log energy of speech signals, the voiced‐unvoiced‐silence detection is approximately made. Second, the pitch period of the candidate for the voiced section can be determined by the pitch‐detection algorithm. A computational algorithm for estimating pitch periods untilizes the time‐domain properties of speech signals. The acoustic boundary is determined by utilizing the time variation of the similarity used in the pitch detector. Finally, according to the amount of speech energy, the remaining sections are successively classified into transitional regions and regions of voiced speech. In order to evaluate our method, we compare its performance with that of L. Rabiner's pitch‐detection algorithm. Quantitative results and several examples illustrating the performance of these methods are included.