Use fsspec and torch for embedding file IO #1581
Conversation
There was a problem hiding this comment.
I think that will be necessary to update the docs and run examples on the compute embeddings script (now it is not .json anymore)
There was a problem hiding this comment.
We also need to do changes in the bin/compute_embedding.py file, like here for exemple.
TTS/TTS/bin/compute_embeddings.py
Line 82 in a790df4
|
|
TTS/tts/utils/managers.py
Outdated
|
|
||
|
|
||
| def save_file(obj: Any, path:str): | ||
| with fsspec.open(path, "rb") as f: |
There was a problem hiding this comment.
you meant "wb" here right @erogol ?
4c9b101
to
41d86e0
Compare
| @@ -22,11 +23,13 @@ | |||
| parser.add_argument("config_dataset_path", type=str, help="Path to dataset config file.") | |||
| parser.add_argument("--output_path", type=str, help="Path for output `pth` or `json` file.", default="speakers.pth") | |||
| parser.add_argument("--old_file", type=str, help="Previous embedding file to only compute new audios.", default=None) | |||
| parser.add_argument("--use_cuda", type=bool, help="flag to set cuda. Default False", default=False) | |||
| parser.add_argument("--disable_cuda", type=bool, help="Flag to disable cuda.", default=False) | |||
There was a problem hiding this comment.
To me, this is counter-intuitive but can live with it :)
There was a problem hiding this comment.
I just spent 3h computing phonemes before realizing that it wasn't using the GPU ^^' That's why I changed it to that. Having use_cuda to true by default and then throwing errors if you had no GPU seemed bad...
No description provided.