SPEECH RECOGNITION FOR WEB BASED TELEPHONY
Independent thesis Advanced level (degree of Master (Two Years))Student thesis
Web based telephony purges the need of explicit downloading and installing a VoIP client software. Calls in web based telephony can be made directly from the browser. The combination of web technologies and traditional telephony makes it possible to introduce new exciting services. One such new service is introduced as a result of this thesis work. The voicemails received are automatically transcribed and converted into text; the text is then saved to an inbox. The performance of the introduced service is good and gives a better recognition rate in the current configuration. The speech recognition covers a continuous speech of English and a maximum vocabulary of 64 thousand words. Adobe Flash 10 has a proprietary protocol for the streaming of audio over internet. Red5 server is an open source server that has support for RTMP plug in. Red5Phone is an open source SIP phone containing a flash based client. The new service introduced is added to the existing Red5Phone solution. Speech recognition for web based telephony was investigated, developed, implemented, and tested. Sphinx-4 is an open source state-of-the art ASR system. It is capable of keeping up with the requirement of large vocabulary transcription. Sphinx-4 was configured and integrated with the developed service for the transcription of voicemails. The performance of Sphinx-4 was rigorously evaluated before its configuration.
Place, publisher, year, edition, pages
2010. , 63 p.
Speech Recognition, VoIP, Web Telephony, RTMP, SIP, Red5Phone, Sphinx-4, Voicemail
Signal Processing Telecommunications
IdentifiersURN: urn:nbn:se:bth-3426Local ID: oai:bth.se:arkivex7083C58C4987F9CAC125770B0041C141OAI: oai:DiVA.org:bth-3426DiVA: diva2:830732