Closed rbychn closed 1 year ago
Tentatively done, although I'm not sure if it's working perfectly
Tried it but something broke
Traceback (most recent call last): File "C:\ProgramData\Anaconda3\envs\tts-fast\lib\site-packages\streamlit\runtime\scriptrunner\script_runner.py", line 565, in _run_script exec(code, module.__dict__) File "app.py", line 13, in <module> from scripts.inference import run_and_save_tts, infer_on_texts, split_and_recombine_text File "C:\Users\Administrator\Desktop\tortoise-tts-fast\scripts\inference.py", line 13, in <module> def parse_voice_str(voice_str: str, all_voices: list[str]): TypeError: 'type' object is not subscriptable 2023-02-15 15:23:33.478 Uncaught app exception Traceback (most recent call last): File "C:\ProgramData\Anaconda3\envs\tts-fast\lib\site-packages\streamlit\runtime\scriptrunner\script_runner.py", line 565, in _run_script exec(code, module.__dict__) File "C:\Users\Administrator\Desktop\tortoise-tts-fast\app.py", line 13, in <module> from scripts.inference import run_and_save_tts, infer_on_texts, split_and_recombine_text File "C:\Users\Administrator\Desktop\tortoise-tts-fast\scripts\inference.py", line 13, in <module> def parse_voice_str(voice_str: str, all_voices: list[str]): TypeError: 'type' object is not subscriptable
4b03ddafbf5976e5bfe82e0c4fe11e50ea1bb95d works fine
@rbychn looks like a python 3.8 incompatibility bug. I'll edit it to be backwards compat.
Thank you, pulled and tried it; seems to work but this time only 1 output was shown on the webpage but in the CLI it shows it generated 3 (as selected in advanced settings) but there's also a warning on the page that it'll only choose 1 candidate, so it's a bit confusing. So not sure what's up with that, it did take a lot of time though.
Todo:
⚠️candidates != 1 while splitting text; only choosing the first candidate for each text fragment!
The audio generated after this repeats a certain piece of text twice.
Example: tortoise tts fast. is cool Output: tortoise tts fast fast is cool
~fat-fingered and closed, sorry~
could you give an example string and seed for me to debug with
could you give an example string and seed for me to debug with
Example String: "The expressiveness of autoregressive transformers is literally nuts! I absolutely adore them. Master the foundations of data science and spin up awesome data science projects anytime, anywhere! Expert-led tutorials. Cloud-hosted notebooks. Take your work to the cloud today."
Seed probably doesn't matter because I've checked multiple voices and it does this for all similarly sized inputs.
Voice: Freeman, Preset: High quality Advanced: Low VRAM (unchecked), Sampler: P
1 output was shown on the webpage but in the CLI it shows it generated 3 (as selected in advanced settings) but there's also a warning on the page that it'll only choose 1 candidate, so it's a bit confusing. So not sure what's up with that, it did take a lot of time though.
I have changed the functionality such that it displays all 3 samples generated.
Seed probably doesn't matter because I've checked multiple voices and it does this for all similarly sized inputs.
It matters a lot on my end. Some seeds cause the problem, others don't. Do you know if this happens with the mrq fork? It might have to do with the duration of each chunk.
I have changed the functionality such that it displays all 3 samples generated.
Thank you!
It matters a lot on my end. Some seeds cause the problem, others don't. Do you know if this happens with the mrq fork? It might have to do with the duration of each chunk.
Will keep that in mind and try to add the seed value, but hope it's sorted now so not need. Will test it!
And no, I've not come across this issue with that fork so far.
regarding everything that was still unaddressed here:
I know this is very new but would be great to have support for large form text input from a file in the UI