Interspeech'2005 - Eurospeech

Lisbon, Portugal
September 4-8, 2005

A Performance Investigation of Noisy Voice Recognition Over IP Telephony Networks

Gang Chen, Douglas O'Shaughnessy, Hesham Tolba

Université du Québec, Canada

A performance analysis of noisy automatic speech recognition (ASR) based on Internet protocol (IP) telephony networks is presented in this paper. The present public telephone voice communication networks, which utilize digital technology via circuit switching, provide satisfactory quality service. In contrast, the Internet telephony switching network, which is a packet-switched network, does not provide a guarantee of quality service at all, since such networks have packet losses. Moreover, the performance of ASR systems deteriorates when such ASR systems are used in adverse environments. For measuring the influence of missing speech packets on the ASR system performance, we use a Soekris net 4501 IP simulator in order to control packet loss rate. Also, to examine how additive acoustic noise influences the speech recognition performance, eleven different types of noise sources are chosen for use in our experiments. In these experiments, the results show that the level of packet loss has a detrimental impact on the speech recognition performance. Furthermore, the speech recognition rate also degrades as the level of additive noise increases under the different kinds of noise conditions.

Full Paper

Bibliographic reference.  Chen, Gang / O'Shaughnessy, Douglas / Tolba, Hesham (2005): "A performance investigation of noisy voice recognition over IP telephony networks", In INTERSPEECH-2005, 2681-2684.