Quest Journals
Journal of Electronics and Communication Engineering Research
Volume 7 ~ Issue 10 (2021) pp: 01-05
ISSN(Online) : 2321-5941
www.questjournals.org
*Corresponding Author: Rubén Hernández 1 | Page
Research Paper
Affine-Projection Adaptive Filter Speech Enhancement
System
Rubén Hernández
1
, Julio C. Sosa
2
, Víctor H. García
2
1
Unidad Profesional Interdisciplinaria en Ingeniería y Tecnologías Avanzadas del Instituto Politécnico
Nacional
Av. Instituto Politécnico Nacional No. 2580,Col. Barrio la Laguna Ticomán.
CP 07340 Gustavo A. Madero, D. F., México.
2
Escuela Superior de Cómputo del Instituto Politécnico Nacional
Av. Juan de Dios Batiz s/n, Col. Lindavista, CP 07738. Gustavo. A. Madero, D.F. México.
ABSTRACT
Evaluation of a system for speech enhancement is presented. A two stage dual channel structure of adaptive
filters allready proposed [1] is used introducing the implemenation of Affine Projection Algorithm at second
stage. Clasical Widrow noise canceller concept is used under reallistic input considerations, where signal of
interest and noise are present at both inputs. Simple Root Mean Square Error (RMSE), Itakura-Saito Distance
(ISD) and subjective evaluation tests are made in order to ease comparisons against single stage adaptive filter
and two stage LMS structure. Using a simple adaptive filter, it is not enough to reduce background noise while
maintaining the intelligibility of enhanced speech. Therefore, a noise emphasizing channel is used in order to
achieve a better noise canceling.
Received 20 September, 2021; Revised: 03 October, 2021; Accepted 05 October, 2021 © The
author(s) 2021. Published with open access at www.questjournals.org
I. INTRODUCTION
Quality in a signal coming from a microphone can be poor if enviroment is noisy, this is common for
most of personal devices such as hearing aids. Many noise reduction schemes have been proposed over the last
decades [Kompis 1998, Moisa 2011, Ngo 2011, Davis 2002]. They can be divided into several categories and
using many algorithms see [Davis 2002] for a very good review. Out of a noise cotrolled place, there is ambient
noise, where close to people babbling is the main noise, and this is the case of more interest when a hearing aid
is used. Esporadic ambient noise such that from the street can represent a similar problem to normal people as
hearing aid users, nevertheless babbling noise in crowded places affect more to hearing aid users.
The hearing loss is related to the intelligibility of the speech embedded in noise and the SNR is useful
in this case for cuantifying of both phenomena, so it is necessary a high value of SNR to get an acceptable
comprehension of the speech, a person with hearing loss has a double trouble due to a loss in the intelligibility.
Degradation in hearing loss of 10 dB is accompanied with degradation in intelligibility within 1 to 1.5 dB
[Davis 2002]. This means that a person with hearing loss has more difficulties for getting a good comprehension
than a normal person when the conversation is done in a noisy environment.
This has made necessary to design speech enhancement systems for improving the intelligibility and
quality of speech. Many noise reduction schemes have been proposed over the last decades [Kompis 1998,
Moisa 2011, Ngo 2011, Davis 2002]. They can be divided into several categories and using many algorithms
see [Davis 2002] for a very good review.
We are interested on hearing aids because there is a need for small devices with few resources and a
very good noise cancelling with great intelligibility.
Some studies have shown that single-channel speech enhancement systems are still unable to improve
intelligibility, even if they can now at least enhance signal quality without reducing intelligibility [Valente
1996]. In contrast, multiple-microphone noise reduction schemes have been shown repeatedly to increase
speech intelligibility even if there remain some theoretical and practical issues to be solved [Kompis 1998,
Benesty 2009]. The performances of multiple channel speech enhancement algorithms improve with an