|
Voice conversion based on eigenvoices| old_uid | 7322 |
|---|
| title | Voice conversion based on eigenvoices |
|---|
| start_date | 2009/07/21 |
|---|
| schedule | 13h30 |
|---|
| online | no |
|---|
| summary | Voice conversion (VC) technique allows us to convert the source speaker's voice into the target speaker's voice without changing linguistic information. A conversion method based on a Gaussian mixture model (GMM) is used widely. In this method, a GMM of joint probability density of source and target acoustic features is previously trained with a parallel data set consisting of utterance pairs of source and target voices. Although this method works well, this training framework is less flexible and causes many limitations of VC applications.
To solve these problems, eigenvoice conversion (EVC) has been proposed. In this framework, eigenvoice GMM (EV-GMM) is trained in advance with multiple parallel data sets consisting of a reference speaker and many pre-stored speakers. The conversion model can be constructed using only a few utterances of the adapted speaker in an unsupervised manner. EVC brings new frameworks, i.e., one-to-many VC, many-to-one VC and many-to-many VC. |
|---|
| responsibles | Loevenbruck, Welby |
|---|
| |
|