Přístupnostní navigace
E-application
Search Search Close
Publication detail
KOULA, I. ESPOSITO, A.
Original Title
Noise cancellation algorithms for speech signal distorted in telecommunication networks.
Type
conference paper
Language
English
Original Abstract
This paper aims to provide an evaluation of the effectiveness of three different speech noise power spectrum estimation algorithms The evaluation of their efficiency was based on the hit rate recognition obtained at the output of an HMM phoneme based speech recognizer. Noisy speech consisted of 100 speech sentences randomly extracted from the NTIMIT database. The best speech noise power spectrum estimator proved to be a procedure based on the arithmetic average of the power spectrums obtained from signal frames where no speech activity was detected. The noise spectrum estimate provide by either a four layer MLP neural network, or an Adaptive Neural Fuzzy Inference System (ANFIS) proved to give lower performance than the average noise spectrum estimator, even though both of them are able to detect some of the noise features and the ANFIS performance are better than those obtained from the MLP neural network.
Keywords
spectral subtraction, thresholdig, neural network, ANFIS, speech recognizer
Authors
KOULA, I.; ESPOSITO, A.
RIV year
2006
Released
1. 1. 2006
Publisher
Ústav radiotechniky a elektroniky, Akademie věd České republiky.
Location
česká republika, Praha
ISBN
86269-15-9
Book
16th Czech-German Workshop on speech processing
Pages from
1
Pages to
7
Pages count
BibTex
@inproceedings{BUT24891, author="Ivan {Koula} and Anna {Esposito}", title="Noise cancellation algorithms for speech signal distorted in telecommunication networks.", booktitle="16th Czech-German Workshop on speech processing", year="2006", pages="7", publisher="Ústav radiotechniky a elektroniky, Akademie věd České republiky.", address="česká republika, Praha", isbn="86269-15-9" }