Finite register length effects in a hidden Markov model speech recognizer

Autor: Wu-Ji Yang, Hsiao-Chuan Wang
Rok vydání: 1990
Předmět:
Zdroj: Speech Communication. 9:239-245
ISSN: 0167-6393
DOI: 10.1016/0167-6393(90)90060-m
Popis: This paper presents a study on finite-register-length effects in a Hidden Markov Model speech recognizer. A statistic model is utilized to approximate the distribution of the score differences. The range of recognition rate due to quantization noise on HMM parameters is calculated by using the statistic model. Then the relation between the recognition rate and the quantization noise is derived. This provides the information for determining the register length in the hardware realization of a HMM speech recognizer.
Databáze: OpenAIRE