The following information was submitted:
Transactions: INTERNATIONAL JOURNAL of CIRCUITS, SYSTEMS and SIGNAL PROCESSING
Transactions ID Number: 20-610
Full Name: Ahsanul Kabir
Position: Researcher
Age: ON
Sex: Male
Address: Department of Telecommunications, Technical University of Cluj-Napoca, 26 Baritiu Street, 400027 Cluj-Napoca
Country: ROMANIA
Tel: +40264401807
Tel prefix:
Fax:
E-mail address: kabirahsanul@hotmail.com
Other E-mails:
Title of the Paper: Modelling Human Speech Perception in Noise
Authors as they appear in the Paper: Ahsanul Kabir, Mircea Giurgiu
Email addresses of all the authors: ahsanul.kabir@com.utcluj.ro, mircea.giurgiu@com.utcluj.ro
Number of paper pages: 8
Abstract: Human auditory system of speech perception tries to find out by applying computational technique how human perceive speech. The difference between the current state of art automatic speech recognition (ASR) and human speech perception (HSP) is the prior knowledge about a given speaker such as speaking style, gestures, eye movements and so on. Therefore if an ASR is feed by the knowledge of a given speaker, then it could be said as HSP system. This paper presents the preliminary research in order to develop a HSP system in Romanian with a view to make it language independent. Acoustic analysis and speech glimpsing are investigated in order to do so. The principal findings are machine tends to recognize noisy speech with a more or less constant recognition rate, but still with a poor recognition rate in compare to their human counterparts, and acoustic parameters have less influence in recognizing noisy speech. In addition, a Romanian speech corpus which we named as !
RO-GRID is collected in ordered to use as the common material in speech perception and automatic speech recognition. Utterances are simple, syntactically identical phrases such as "muta bronz cu p 2 agale." The corpus is annotated at the phoneme, syllable and word level and is available on the website for research use.
Keywords: Romanian Speech Corpus, Hidden Markov Models, Speech Intelligibility, Speaker Intelligibility, Vocal Tract Length Normalization, Glimpsing Speech
EXTENSION of the file: .pdf
Special (Invited) Session: A Romanian Corpus for Speech Perception and Automatic Speech Recognition
Organizer of the Session: 650-582
How Did you learn about congress:
IP ADDRESS: 193.226.5.148
