List of pretrained models available for download
Opened this issue · 6 comments
Deleted user commented
This issue will be used to provide a listing of pretrained models that are freely available to use (resolves #268).
Deleted user commented
This list is no longer maintained. The synthesizer models below will only work with an older version of the repo code. Use this commit and these install instructions.
English
# | Author | Type | Link | Hparams | Description |
---|---|---|---|---|---|
001 | @CorentinJ | All models | wiki/Pretrained-models | Default | Original models |
002 | @blue-fish | Vocoder | #126 (comment) | Default | Add 700k steps to the 001 vocoder, fully compatible with original models. Voice quality same but better at reducing sound artifacts? #126 (comment) |
003 | @sberryman | All models | #126 (comment) | Modified | Highly trained encoder. Sound quality is good but cloned voices do not resemble reference audio. #126 (comment) |
004 | @mbdash | Synthesizer | #449 (comment) | Default | Trained to 200k steps on LibriTTS |
005 | @mbdash | Encoder | #458 (comment) | Modified | Trained on LibriSpeech, VCTK, CommonVoice |
006 | @blue-fish | Synthesizer | #538 | Default | Fewer gaps in output compared to original synthesizer model |
Multi-language
# | Author | Type | Link | Hparams | Description |
---|---|---|---|---|---|
101 | @sberryman | Encoder | #126 (comment) | Modified | Highly trained encoder on mixed language set. Suitable for training a synthesizer or vocoder in any language? |
Chinese
# | Author | Type | Link | Hparams | Description |
---|---|---|---|---|---|
201 | @kuangdd | All models | #30 (comment) | Modified | Fork, see #30 (comment) |
Swedish
# | Author | Type | Link | Hparams | Description |
---|---|---|---|---|---|
301 | @ViktorAlm | Synthesizer Vocoder |
#257 | Default | To be used with encoder model from 001. Minor code changes are needed for this to work. |
German
# | Author | Type | Link | Hparams | Description |
---|---|---|---|---|---|
401 | @padmalcom | All models | #571 (comment) | Default | Trained on female voices in M-AILABS German dataset |
Deleted user commented
The list is still being maintained at this time. It is closed to avoid cluttering the issues list.
mennatallah644 commented
is there arabic model?
mennatallah644 commented
@blue-fish
are these the most updated pretrained models?
Deleted user commented
@mennatallah644 The list is up-to-date. There are no Arabic pretrained models at this time.
babysor commented
@mennatallah644 The list is up-to-date. There are no Arabic pretrained models at this time.
It seems the link to the Chinese model is now broken.