Unvoiced Components Research Articles

Adaptive Refinements of Pitch Tracking and HNR Estimation within a Vocoder for Statistical Parametric Speech Synthesis

Recent studies in text-to-speech synthesis have shown the benefit of using a continuous pitch estimate; one that interpolates fundamental frequency (F0) even when voicing is not present. However, continuous F0 is still sensitive to additive noise in speech signals and suffers from short-term errors (when it changes rather quickly over time). To alleviate these issues, three adaptive techniques have been developed in this article for achieving a robust and accurate F0: (1) we weight the pitch estimates with state noise covariance using adaptive Kalman-filter framework, (2) we iteratively apply a time axis warping on the input frame signal, (3) we optimize all F0 candidates using an instantaneous-frequency-based approach. Additionally, the second goal of this study is to introduce an extension of a novel continuous-based speech synthesis system (i.e., in which all parameters are continuous). We propose adding a new excitation parameter named Harmonic-to-Noise Ratio (HNR) to the voiced and unvoiced components to indicate the degree of voicing in the excitation and to reduce the influence of buzziness caused by the vocoder. Results based on objective and perceptual tests demonstrate that the voice built with the proposed framework gives state-of-the-art speech synthesis performance while outperforming the previous baseline.

Classification of Sex based Speech Differentiation in Healthy Human Beings based on Voiced and Unvoiced Components

Background/Objective: The objective of the present study is to classify a given speech signal by using energy as a differentiating parameter into voiced and unvoiced components due to the fact that the voiced components have a higher energy than their unvoiced counterparts. Method/Statistical Analysis: This is accomplished by dividing the speech signal into frames and by computing the short time energy of each frame. The recorded speech signal is segmented and then the energy component of these frames are obtained and then classified into voiced and unvoiced components. The current protocol involves 44 subjects, both males and females of no known vocal pathology. Predefined set of words, both in Kannada and English were recorded in a noise proof environment which was then separated into voiced and unvoiced components using MATLAB tool. Findings: The results proved a successful discrimination of the speech signal into voiced and unvoiced components based on the statistical parameters calculated for voiced as well as unvoiced components thereby providing a definite cue towards an automated approach to differentiate the speech into voiced and unvoiced components using statistical parameters. Application/Improvements: Such an approach can further be useful in various speech processing as well as speech recognition applications.

Segregation of voiced and unvoiced components from residual of speech signal

In conventional source-filter models, voiced and unvoiced components were considered independently. However, in practice it was difficult to separate the source into two parts. An actual source consists of a mixture of two sources and the ratio varies according to the content or the intention of speaker. It had been investigated to separate the voiced and unvoiced components for different source models. Source signals were modeled based on the residual signal measured from inverse filtering. Three different source models were assumed. The parameters of each model were optimized for the original speech signal using a genetic algorithm. The resulting parameters were compared in terms of the mel-cepstral distance to the original signal, the spectrogram and the spectral envelope from the synthesized signal. The optimization method achieves an improvement of 15% for the Klatt model, but there is little improvement in the modified residual case.

Improving signal quality of a speech codec using hybrid perceptual-parametric algorithm

A hybrid parametric-perceptual speech codec architecture is presented. The basic CELP parametric codec structure is enhanced using the perceptual coding method. The objective of the codec hybridisation is obtaining significant improvement in the perceived signal quality. Two hybrid architectures are proposed. The first one encodes voiced parts perceptually in the CELP residual signal. The second one divides the signal into voiced signal components that are encoded using the perceptual algorithm, unvoiced components that are encoded parametrically and transients remaining unencoded. The signal quality achieved using the hybrid codec is compared to the quality of some standard speech codecs.

Electrolaryngeal speech enhancement for telephony

A technique for separating an acoustic signal into a voiced (V) component corresponding to an electrolaryngeal source and an unvoiced (U) component corresponding to a turbulence source. The technique can be used to improve the quality of electrolaryngeal speech, and may be adapted for use in a special purpose telephone. A method according to the invention extracts a segment of consecutive values from the original stream of numerical values, and performs a discrete Fourier transform on the this first group of values. Next, a second group of values is extracted from components of the discrete Fourier transform result which correspond to an electrolaryngeal fixed repetition rate, F0, and harmonics thereof. An inverse-Fourier transform is applied to the second group of values, to produce a representation of a segment of the V component. Multiple V component segments are then concatenated to form a V component sample stream. Finally, the U component is determined by subtracting the V component sample stream from the original stream of numerical values.

Multi-feature speech/music discrimination system

A speech/music discriminator employs data from multiple features of an audio signal (10) as input to a classifier (16). Some of the feature data is determined from individual frames of the audio signal, and other input data is based upon variations of a feature over several frames, to distinguish the changes in voiced and unvoiced components of speech from the more constant characteristics of music. Several different types of classifiers for labeling test points on the basis of the feature data are disclosed. A preferred set of classifiers is based upon variations of a nearest-neighbor approach, including a K-d tree spatial partitioning technique.

Wavelet-Based Speech Enhancement Using Time-Frequency Adaptation

Wavelet denoising is commonly used for speech enhancement because of the simplicity of its implementation. However, the conventional methods generate the presence of musical residual noise while thresholding the background noise. The unvoiced components of speech are often eliminated from this method. In this paper, a novel algorithm of wavelet coefficient threshold (WCT) based on time-frequency adaptation is proposed. In addition, an unvoiced speech enhancement algorithm is also integrated into the system to improve the intelligibility of speech. The wavelet coefficient threshold (WCT) of each subband is first temporally adjusted according to the value of a posterior signal-to-noise ratio (SNR). To prevent the degradation of unvoiced sounds during noise, the algorithm utilizes a simple speech/noise detector (SND) and further divides speech signal into unvoiced and voiced sounds. Then, we apply appropriate wavelet thresholding according to voiced/unvoiced (V/U) decision. Based on the masking properties of human auditory system, a perceptual gain factor is adopted into wavelet thresholding for suppressing musical residual noise. Simulation results show that the proposed method is capable of reducing noise with little speech degradation and the overall performance is superior to several competitive methods.

Digital-formant synthesizer for speech-synthesis studies.

Recent work on speech synthesis by rule has led to the design and evaluation of a new digital-formant synthesizer. The synthesizer was simulated on a digital computer at a 20-kHz sampling frequency. The improvements over previous synthesizers include: an excitation source for the unvoiced component of voiced fricatives; a method for producing a voice bar for the closure period of voiced stops; and a simple design for the higher pole correction network that exploits the properties of sampled data systems. Spectrographic examples of synthetic speech are shown in order to illustrate these improvements. A brief discussion of some advantages and disadvantages of serial and parallel synthesizers is presented.

Digital Formant Synthesizer for Speech Synthesis Studies

Recent work on speech synthesis by rule has led to the development and evalation of a novel digital formant synthesizer. The synthesizer was simulated on a computer at a 20-kHz sampling frequency. Among the novel points are an excitation source for the unvoiced component of voiced fricatives; a voicebar for the stop gap period of voiced stops; a novel source filter providing the voiced excitation for the synthesizer and the elimination of the higher pole correction network for voiced sounds by exploiting the properties of sampled data systems. A comparison (in terms of the frequency response for vowels) between digital and analog synthesizers will be made. A brief discussion of the advantages and disadvantages of cascade and parallel synthesizers will be presented. Examples of synthetic speech will be presented.

Some New Methods for Digital Encoding of Voice Signals and for Voice Code Translation

Recently developed techniques are described for improving the speech quality of voice signals that are first digitally encoded, placed in random access storage, and on demand are then translated into normal speech in an audio response unit under the control of a host processor. The development is an extension and modification of the channel vocoder principle. Speech quality is enhanced by hardware and software features for treatment of unvoiced components of the coded speech signal in particular by separating harmonics from the excitation function digital signal before smoothing. A new program of bit selection is used to assure that the aggregation function digital signal carries maximum information. In addition, an efficient method of storage assignment is shown for the excitation function and the aggregate function registers in the voice code translator.