Acoustic analysis and recognition of whispered speech
Tóm tắt
The acoustic properties and a recognition method of whispered speech are discussed. A whispered speech database that consists of whispered speech, normal speech and the corresponding facial video images of more than 6,000 sentences from 100 speakers was prepared. The comparison between whispered and normal utterances show that: 1) the cepstrum distance between them is 4 dB for voiced and 2 dB for unvoiced phonemes; 2) the spectral tilt of whispered speech is less sloped than for normal speech; 3) the frequency of the lower formants (below 1.5 kHz) is lower than that of normal speech. Acoustic models (HMM) trained by the whispered speech database attain an accuracy of 60% in syllable recognition experiments. This accuracy can be improved to 63% when MLLR (maximum likelihood linear regression) adaptation is applied, while the normal speech HMMs adapted with whispered speech attain only 56% syllable accuracy.
Từ khóa
#Speech analysis #Speech recognition #Speech processing #Hidden Markov models #Image databases #Maximum likelihood linear regression #Video recording #Loudspeakers #Cepstrum #FrequencyTài liệu tham khảo
eklund, 1996, Comparative study of male and female whispered and phonated versions of the long vowels of Swedish, 54, 1
10.1121/1.2020610
kawahara, 1999, Japanese Dictation Toolkit: Plug-and-play Framework For Speech Recognition R&D, Proc IEEE Workshop Automatic Speech Recognition and Understanding (ASRU), 393
10.1044/jshr.2702.251
leggetter, 1995, Flexible speaker adaptation using maximum likelihood linear regression, Proc ARPA Workshop on Spoken Language Technology
itou, 1998, Design and development of Japanese speech corpus for large vocabulary continuous speech recognition, Proc of Oriental COCOSDA
10.1121/1.1911712
10.1121/1.1908631