Speech Spectral Components Research Articles

Visual speech display for the deaf

In this paper, the results of the preliminary development of a visual speech display as a speech training aid for the hearing impaired will be presented. The training aid consists of analog electronics for extracting speech parameters and a microprocessor‐based system for converting the speech parameters to color parameters and for controlling the display parameters. The speech parameters consist of six spectral shape factors, similar to speech spectral principal components and a pitch signal proportional to voice fundamental frequency. The spectral shape factors are computed by appropriately combining the low‐pass‐filtered logarithmically scaled outputs of a bank of 16 bandpass filters. The spectral shape factors will be used to control the color of the display, and the pitch signal will be used to control “texture” in the display. The results of a vowel discrimination experiment for steady‐state vowels will be presented. [Work supported by the Whitaker Foundation.]

Optimal vector quantization of speech spectral principal components

The vector quantization method described by Linde et al. [IEEE Trans. Commun. 28, 84–95 (1980)] was used to optimally quantize speech spectral principal components. The spectral principal components, obtained from linear combinations of spectral band energies, are an optimal uncorrelated parameter set containing a very large amount of spectral variance with a relatively small number of dimensions. For a given percentage of quantization distortion, the vector quantizer required fewer bits per frame than individual quantization of principal components, thus indicating a clustering of the speech data in the uncorrelated parameter space. Using the principal components parameters to measure both gain and spectral shape and using eight bits per frame, the mean square vector quantization error was about 2.5% in a four‐dimensional space, and about 5% in a seven‐dimensional space. Speech synthesized from the vector quantized principal components using nine bits per frame was somewhat inferior in quality to speech synthesized with full precision principal components, but quite intelligible. [Work supported by NSF.]

The speech noise and its spectrum

By recording on one magnetic tape the simultaneous utterances of several talkers through several channels, we have obtained a noise that is constant in time and continuous in frequency, and renders the average spectral components of speech more faithfully than other methods hitherto applied. Such speech noises are open to analysis, and their spectra can thus be determined. The advantages of the method are: 1) the integration needed for determining the longtime spectrum can be reduced to a few seconds; 2) the noise can be treated as a statistical noise constant in time; 3) the noise does not contain any line spectrum elements; 4) it improves the signal-to-noise ratio even under a normal force of utterance. It was shown that the speech spectrum of a certain language may be looked upon as being representative of the language in question. Long-time speech spectra obtained by this method were used to compute the average loudness level of a single person. The method can be used to find the characteristic spectra of certain speech sounds and of individual pronunciation.