Third International Conference on Spoken Language Processing (ICSLP 94)

Yokohama, Japan
September 18-22, 1994

A Neural Network for Phonetically Decoding the Speech Trace

Victoria Rodellar, Victor Nieto, Pedro Gomez, Daniel Martinez, M. Mercedes Perez

Departamento de Arquitectura y Tecnologia de Sistemas Informaticos, Universidad Politecnica de Madrid, Madrid, Spain

A Time-Delay Neural Network to decode Speech Traces into Phoneme Codes in Real time is presented. Such network associates traces of LPC Spectral Parameters with strings of 8-bit Phonetic Codes from a Phonetic Encoding Scheme, which is also presented. The LPC Adaptive Parameter Extractor and the Time-Delay Neural Network may be easily integrated under a common recursive loop in the Time Domain, allowing a fast implementation in VLSI with small memory requirements, which convert this hybrid structure in a powerful Phoneme Spotter, to be used in Speech Recognition or in the Assesment of the Speech Trace. A set of experiments is presented to show the ability of the Neural Network to determine the most significant elements in the LPC vectors from several speakers.

Full Paper

Bibliographic reference.  Rodellar, Victoria / Nieto, Victor / Gomez, Pedro / Martinez, Daniel / Perez, M. Mercedes (1994): "A neural network for phonetically decoding the speech trace", In ICSLP-1994, 1575-1578.