|
|
--- |
|
|
license: mit |
|
|
datasets: |
|
|
- upai-inc/saspeech |
|
|
- sleeping-ai/Fish-Hebrew |
|
|
language: |
|
|
- he |
|
|
base_model: |
|
|
- fishaudio/fish-speech-1.5 |
|
|
--- |
|
|
|
|
|
Hebrew is fundamentally a hard language to work in the field of Natural language processing and it is also one of the underrepresented language in the field of Speech-Speech and Text-to-Speech models. Mainly boils down to limited availability of data. To explore Speech-Speech **(Voice Cloning)**, I used [Dataset](https://www.isca-archive.org/interspeech_2023/sharoni23_interspeech.pdf) to fine-tune Fish-speech 1.5 on roughly 2.5 hours of Hebrew audio on their Gold-standard subset. |
|
|
|
|
|
I have also fixed a few bugs on Fish's fine-tuning code and created a [pull-request](https://github.com/fishaudio/fish-speech/pull/973) |
