Automatic segmentation and labeling of speech
ICASSP '91 Proceedings of the Acoustics, Speech, and Signal Processing, 1991. ICASSP-91., 1991 International Conference
Real-time discrimination of broadcast speech/music
ICASSP '96 Proceedings of the Acoustics, Speech, and Signal Processing, 1996. on Conference Proceedings., 1996 IEEE International Conference - Volume 02
| Hi-index | 0.02 |
In speech recognition, the procurement of accurate patterns that describe an input signal is a crucial task. Frequency-domain processing provides with rich information for such signal descriptions. However a first interpretation of the time-domain characteristics of the speech utterances may be enough for obtaining important information contained in the signal in a faster way. This paper shows that segmentation and labelling of speech may be performed using only time-domain information in an exact and accurate way. The method obtains syllable and phoneme level segmentation in two stages. The first identifies sonority decrease intervals for estimating transitions between syllables. The second, refines the placement of boundaries using a set of fuzzy-rules that compared current time-marks with previously computed syllable-transition values. The system was tested using an Italian language digit database. The reported results show that the accuracy of the inter-syllabic boundary placements get improved when using the fuzzy-correction method.