You need to agree to share your Hugging Face information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Use of this model is prohibited for the following purposes:

  • commercial purposes, including but not limited to the following activities when carried out for commercial gain: advertising, voicing audiobooks and e-books; films, videos and multimedia; creating and producing audio content, including marketing and training materials; podcasts. Commercial use is permitted only with prior written consent.
  • immoral or unethical purposes;
  • political purposes, including military use.

For more information or to request permission for commercial use, please contact ping@tartunlp.ai.


Mudeli kasutamine on keelatud:

  • ärilistel eesmärkidel, mis hõlmab sealhulgas, kuid mitte ainult, järgmisi tegevusi, tingimusel, et tegevusega taotletakse ärilisi eesmärke: reklaam, audioraamatute ja e-raamatute helindamine; filmid, videod, multimeedia; helisisu loomine ja tootmine, sh turundus- ja koolitusmaterjalid; podcastid. Äriline kasutus on lubatud üksnes eelneval kirjalikul nõusolekul.
  • ebamoraalsetel või ebaeetilistel eesmärkidel;
  • poliitilistel eesmärkidel, sealhulgas sõjalises valdkonnas.

Lisateabe saamiseks või ärilise kasutuse loa taotlemiseks võtke ühendust aadressil ping@tartunlp.ai.

Log in or Sign Up to review the conditions and access this model content.

Multi-Style FastPitch Model for Estonian

An experimental multi-style Estonian text-to-speech model developed by the University of Tartu using the FastPitch implementation of the TransformerTTS library (version v1.2.0-beta.1).

This is the first version and still somewhat rough. Inference can be used via the code in Koodivaramu. The model supports 10 speakers and 3 styles, however, it was trained with data from more than 700 speakers to help with pronunciation robustness and versatility. The main benefit of this model is more natural prosody with longer conversational texts. For production use, we still recommend using the regular multi-speaker model for reliability and wait for a more stable release.

When using the model, please cite to the following publication:

@inproceedings{ratsep24_interspeech,
  title     = {{Enabling Conversational Speech Synthesis using Noisy Spontaneous Data}},
  author    = {Liisa Rätsep and Rasmus Lellep and Mark Fishel},
  year      = {2024},
  booktitle = {{Interspeech 2024}},
  pages     = {4923--4927},
  doi       = {10.21437/Interspeech.2024-2403},
  issn      = {2958-1796},
}

The speakers and styles are mapped to ID-s as follows:

speakers:
1: albert
2: indrek
3: kalev
4: kylli
5: liivika
6: mari
7: meelis
8: peeter
9: tambet
10: vesta

styles:
0: news
1: fiction
2: conversational

Terms of Use

Use of this model is prohibited for the following purposes:

  • commercial purposes, including but not limited to the following activities when carried out for commercial gain: advertising, voicing audiobooks and e-books; films, videos and multimedia; creating and producing audio content, including marketing and training materials; podcasts. Commercial use is permitted only with prior written consent.
  • immoral or unethical purposes;
  • political purposes, including military use.

For more information or to request permission for commercial use, please contact ping@tartunlp.ai.

Downloads last month
-
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support