Publication detail

State-Space Representation Of Cepstral Vocal Tract Model For DSP Implementation

VONDRA, M., SMÉKAL, Z., VÍCH, R.

Original Title

State-Space Representation Of Cepstral Vocal Tract Model For DSP Implementation

English Title

State-Space Representation Of Cepstral Vocal Tract Model For DSP Implementation

Type

journal article - other

Language

en

Original Abstract

One of the methods of speech production in text-to-speech (TTS) synthesis is the parametric method. An excitation signal excites the vocal-tract model with time-varying parameters. A new state-space cepstral vocal-tract model is described, which approximates both the formants and the antiformants of the model frequency response for voiced and unvoiced speech sounds. It thus differs from the currently used LPC model, which approximates the formants alone. Unlike methods of the type PSOLA, this method is convenient for prosody modelling and requires less memory capacity. The cepstral speech synthesis starts from the cepstral coefficients obtained by analysing the speech signal. In the paper a new structure of parametric vocal-tract model is proposed, which is formed by combining IIR and FIR digital filters. The model is optimised with respect to implementation on a fixed-point digital signal processor with Harvard architecture.

English abstract

One of the methods of speech production in text-to-speech (TTS) synthesis is the parametric method. An excitation signal excites the vocal-tract model with time-varying parameters. A new state-space cepstral vocal-tract model is described, which approximates both the formants and the antiformants of the model frequency response for voiced and unvoiced speech sounds. It thus differs from the currently used LPC model, which approximates the formants alone. Unlike methods of the type PSOLA, this method is convenient for prosody modelling and requires less memory capacity. The cepstral speech synthesis starts from the cepstral coefficients obtained by analysing the speech signal. In the paper a new structure of parametric vocal-tract model is proposed, which is formed by combining IIR and FIR digital filters. The model is optimised with respect to implementation on a fixed-point digital signal processor with Harvard architecture.

RIV year

2002

Released

03.09.2002

ISBN

1213-161X

Book

ElectronicsLetters.com

Edition

3/9/2002

Edition number

1

Pages from

1

Pages to

6

Pages count

6

BibTex


@article{BUT40902,
  author="Martin {Vondra} and Zdeněk {Smékal} and Robert {Vích}",
  title="State-Space Representation Of Cepstral Vocal Tract Model For DSP Implementation",
  annote="One of the methods of speech production in text-to-speech (TTS) synthesis is the parametric method. An excitation signal excites the vocal-tract model with time-varying parameters. A new state-space cepstral vocal-tract model is described, which approximates both the formants and the antiformants of the model frequency response for voiced and unvoiced speech sounds. It thus differs from the currently used LPC model, which approximates the formants alone. Unlike methods of the type PSOLA, this method is convenient for prosody modelling and requires less memory capacity. The cepstral speech synthesis starts from the cepstral coefficients obtained by analysing the speech signal. In the paper a new structure of parametric vocal-tract model is proposed, which is formed by combining IIR and FIR digital filters. The model is optimised with respect to implementation on a fixed-point digital signal processor with Harvard architecture.",
  booktitle="ElectronicsLetters.com",
  chapter="40902",
  edition="3/9/2002",
  journal="ElectronicsLetters.com - http://www.electronicsletters.com",
  number="6.9.",
  volume="2002",
  year="2002",
  month="september",
  pages="1",
  type="journal article - other"
}