Title | ||
---|---|---|
Automatic segmentation combining an HMM-based approach and spectral boundary correction |
Abstract | ||
---|---|---|
Currently, AT&T Labs' Natural Voices multilingual TTS system produces high-quality synthetic speech with a large- scale speech corpus (1). In the development of such sys- tems, automatic segmentation constitutes a major compo- nent technology. The prevalent approach for automatic segmentation in speech synthesis is Hidden Markov Model (HMM) - based. Even though an HMM-based approach is the most auto- matic and reliable, there are still several limitations, such as mismatches between hand-labeled transcriptions and HMM alignment labels which can lead to discontinuities in the synthetic speech, or the need for hand-labeled bootstrap data in HMM initialization. This paper introduces a new ap- proach to automatic segmentation which aims both to min- imize human intervention and to achieve a higher segmen- tal quality of synthetic speech in unit-concatenative speech synthesis, by combining a conventional HMM-based ap- proach and spectral boundary correction. A preference test demonstrates the proposed method is effective in reducing discontinuities in synthetic speech. |
Year | Venue | Keywords |
---|---|---|
2002 | INTERSPEECH | hidden markov model,speech synthesis |
Field | DocType | Citations |
Scale-space segmentation,Pattern recognition,Computer science,Segmentation,Segmentation-based object categorization,Speech recognition,Artificial intelligence,Hidden Markov model | Conference | 30 |
PageRank | References | Authors |
3.17 | 3 | 2 |
Name | Order | Citations | PageRank |
---|---|---|---|
Yeon-Jun Kim | 1 | 52 | 9.52 |
Alistair Conkie | 2 | 264 | 38.03 |