Upload README.md
Browse files
README.md
CHANGED
|
@@ -37,6 +37,10 @@ Voices are listed in [VOICES.md](https://huggingface.co/hexgrad/Kokoro-82M/blob/
|
|
| 37 |
|
| 38 |
Support for non-English languages may be absent or thin due to weak G2P and/or lack of training data. Some languages are only represented by a small handful or even just one voice (French).
|
| 39 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 40 |
### Usage
|
| 41 |
|
| 42 |
The following can be run in a single cell on [Google Colab](https://colab.research.google.com/).
|
|
|
|
| 37 |
|
| 38 |
Support for non-English languages may be absent or thin due to weak G2P and/or lack of training data. Some languages are only represented by a small handful or even just one voice (French).
|
| 39 |
|
| 40 |
+
Most voices perform best on a "goldilocks range" of say, 50-150 tokens out of ~500 possible. Voices may perform worse at the extremes:
|
| 41 |
+
- **Weakness** on short utterances (especially less than 10-20 tokens). Root cause could be lack of short-utterance training data and/or model architecture. One possible inference mitigation is to bundle shorter utterances together.
|
| 42 |
+
- **Rushing** on long utterances (over 400 tokens). You can chunk down to shorter utterances or can adjust the `speed` parameter to mitigate this.
|
| 43 |
+
|
| 44 |
### Usage
|
| 45 |
|
| 46 |
The following can be run in a single cell on [Google Colab](https://colab.research.google.com/).
|