Version 1.2

Updated Estonian TTS models for the same speakers with even better corpora preprocessing to improve speech quality in the beginning and end of sentences. For best quality, we recommend using the autosegment.pth model file.

Preset file: presets/eesti_konekorpus.json

PS: It is recommended using the same version of Deep Voice 3 code with the attached model, as other versions may not be compatible.

Version 1.1

Updated Estonian TTS model with:

• Improved data preprocessing to improve speech quality in the beginning and end of sentences.
• Reduced vocabulary size

Preset file: presets/eesti_konekorpus.json

PS: It is recommended using the same version of Deep Voice 3 code with the attached model, as other versions may not be compatible.

Version 1.0

This release contains a public model and can be used with our TTS API. The model has been trained on an Estonian news corpus and two literature corpora. Six speakers are supported:

0. Mari (news)
1. Kalev (news)
2. Albert (news)
3. Vesta (news)
4. Külli (literature)
5. Meelis (literature)

PS: It is recommended using the same version of Deep Voice 3 code with the attached model, as other versions may not be compatible.