|
Inclusion of conversational elements in text-to-speech systems: challenges and opportunities for the synthesis of disfluences| old_uid | 12044 |
|---|
| title | Inclusion of conversational elements in text-to-speech systems: challenges and opportunities for the synthesis of disfluences |
|---|
| start_date | 2016/09/29 |
|---|
| schedule | 14h |
|---|
| online | no |
|---|
| summary | This talk defends the need of including conversational elements of spontaneous speech, such as disfluences, in text-to-speech systems in order to improve its naturalness on the way to move from conventional reading machines to expressive talking machines. But including disfluences in nowadays text to speech technology is a challenging task due to the lack of prediction models and specific corpora. The speaker presents how the use of heuristics trained from corpora permits to successfully solve these limitations. Furthermore the state of the art in this promising field of research will be surveilled including alternative approaches for predicting the position of disfluences from text and its automatic modelling from corpus.
The speaker will profit the occasion to present his current projects on automatic prosodic labeling and on computer assisted pronunciation training. He is doing a research stay in GIPSA-Lab from March to June 2017 so it will be an excellent occasion for identifying common fields of interest. |
|---|
| responsibles | Hueber |
|---|
| |
|