You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hello @kauterry , I tested the example you provided in the Seamless Tutorial (https://github.com/facebookresearch/seamless_communication/blob/main/Seamless_Tutorial.ipynb) about building a SeamlessStreaming S2ST + S2TT agent. In the example you provided, the input audio is in English, the target language is Spanish ("spa"), and the source_segment_size=320ms. However, when I changed the target language to Chinese ("cmn"), I found that the translated results for both S2ST and S2TT were still in English. Only when I set the source_segment_size to around 700ms did the output result in Chinese. Could you help explain this?
More important, how can I force the final translation to be in Chinese (even if the result is poor), regardless of how small the source_segment_size is set?
The text was updated successfully, but these errors were encountered:
Hello @kauterry , I tested the example you provided in the Seamless Tutorial (https://github.com/facebookresearch/seamless_communication/blob/main/Seamless_Tutorial.ipynb) about building a SeamlessStreaming S2ST + S2TT agent. In the example you provided, the input audio is in English, the target language is Spanish ("spa"), and the source_segment_size=320ms. However, when I changed the target language to Chinese ("cmn"), I found that the translated results for both S2ST and S2TT were still in English. Only when I set the source_segment_size to around 700ms did the output result in Chinese. Could you help explain this?
More important, how can I force the final translation to be in Chinese (even if the result is poor), regardless of how small the source_segment_size is set?
The text was updated successfully, but these errors were encountered: