Differences

This shows you the differences between two versions of the page.

Link to this comparison view

analysis_of_the_influence_of_acoustical_adverse_conditions_on_speaker_gender_identification [2014/03/15 12:41] (current)
bziolko created
Line 1: Line 1:
 + Tomasz Mąka, Piotr Dziurzański (Poland)
  
 +**Analysis of the influence of acoustical adverse conditions on speaker gender identification**
 +
 +Speaker gender as a biometric feature plays an important role in numerous voice-based services. In the tasks like audio surveillance, human-computer interaction and access control systems such information can be used to improve their accuracy and capabilities. Since gender recognition provides biometric properties of the user, it may be exploited in voice tasks adaptation process to user needs. In this work we perform accuracy analysis of gender recognition system in different acoustical environments (several indoor and outdoor auditory scenes). For this task we prepare a large set of acoustic and prosodic features at the feature extraction stage, where for each background noise the feature selection (using several filter-type feature selection procedures) process has been exploited. A feature discriminatory procedure has been performed using one sentence spoken by 630 speakers (438 male and 192 female speakers) from the TIMT database. At the evaluation stage, each sentence has been mixed with several types of background noises using various SNR (Signal to Noise Ratio) levels. In order to parametrize sentences, we have used features based on filter banks, vocal-tract properties and long-term descriptors. The obtained feature trajectories were fused together and post-processed to maximize recognition accuracy at the classification stage, where the SVM classifier (Support Vector Machines) was employed. We have made an attempt to determine the audio feature sets dedicated to different acquisition conditions of speech signal. In the result, the proposed approach can be used to compensate an environmental influence on speaker gender recognition systems.
Copyright © XXII PVC Organizing Committee 2013. All Rights Reserved.