search

facebookresearch / seamless_communication

Foundational Models for State-of-the-Art Speech and Text Translation
Other
10.81k stars 1.05k forks source link

I want to just t2t service #145

Closed juud79 closed 1 year ago

juud79 commented 1 year ago

I want to just t2t service (text to text translation)

do i need to load vocoder_name_or_card ??

i want to reduce resource of gpu and improve translation time

when i load no vocoder_name_or_card there is an error TypeError: Translator.init() missing 1 required positional argument: 'vocoder_name_or_card'

translator = Translator( model_name_or_card="seamlessM4T_large", vocoder_name_or_card="vocoder_36langs", device=device2, sample_rate=AUDIO_SAMPLE_RATE, )

cndn commented 1 year ago

Hey @juud79 - Good point! Right if you just need text output, vocoder is not needed. The Translator is a model wrapper that shows one way to use our model (for all tasks) - but feel free to come up with your own wrapper.