Evaluation of the effectiveness of parameters of the global description of words in simple automatic speech recognition systems

Downloads

Authors

  • Cz. BASZTURA Institute of Telecommunication and Acoustics Technical University of Wrocław, Poland
  • W. MAJEWSKI Institute of Telecommunication and Acoustics Technical University of Wrocław, Poland
  • W. BARYCKI Institute of Telecommunication and Acoustics Technical University of Wrocław, Poland

Abstract

In this paper, oriented at the so-called simple systems of speech recognition, the effectiveness of 6 sets of parameters in the global description of words in a given vocabulary was analysed. The usability of simple parametrization methods and such parameters as: density of zero-crossings, distribution of intervals between zero-crossings, parameters of two so-called phase planes, spectral parameters in octave and tertiary bands, was investigated and analysed on the basis of sound material for one operator, mainly and a vocabulary from 5 to 40 words. It was proved that such parameters as the density of zero-crossings and the distribution of time intervals between zero-crossings can be applied in simple systems with a vocabulary preferably not exceeding 10 words, unless it would be possible to select certain words from the vocabulary. Parameters of the first phase plane and spectral parameters exhibited positively weak discrimination ability, especially in octave bands. Also the usability of the NN algorithm and the Camberra distance which standarizes parameters in such ASR systems was confirmed.

References

[1] Cz. BASZTURA, Acoustic sources, signals and images (in Polish) WKik 1988.

[2] Cz. BASZTURA, Automatic speaker recognition by zero-crossing analysis of the speech signal, in: Speech Analysis and Synthesis, [Ed.] W. Jassem, 5, PWN, Warszawa 1980, pp. 5-40.

[3] Cz. BASZTURA, J. ZUK, Similarity functions of images in non-parametric voice identification algorithms, Archives of Acoustics, 16, 2, (1991).

[4] Cz. Basztura, T. SAWCZYN, Automatic segmentatation of a speech signal with the application of selected parametrization methods, submitted for publication in Archives of Acoustics.