Seamless Expressive with manual translation

facebookresearch / seamless_communication

Foundational Models for State-of-the-Art Speech and Text Translation

Other

10.82k stars 1.05k forks source link

Seamless Expressive with manual translation #253

Open ra100 opened 10 months ago

ra100 commented 10 months ago

Is it possible to modify the resulting translated text before it's transformed to audio?

I want to S2ST with expressions, but if the translation doesn't quite match the context or whatever, to "force" the resulting translation with one or two words changed.

darkzbaron commented 10 months ago

Same here. I tried to use the translator object to get the phonetic units but with no success.

text_supersede, unit_output_supersede = translator.predict( example["fbank"], "S2ST", tgt_lang=target_language_code, src_lang=target_language_code, text_generation_opts=text_generation_opts, unit_generation_ngram_filtering=False, duration_factor=1.0, prosody_encoder_input=prosody_encoder_input, src_text='voici un test', # for mintox check )

Thomcle commented 1 week ago

Same

https://huggingface.co/facebook/seamless-expressive/discussions/7