Closed AmitMY closed 8 months ago
The current spoken language identification - implemented in https://github.com/sign/translate/issues/10 - has limitations, specifically, it struggles with short texts https://github.com/google/cld3/issues/76#issuecomment-1625233427
The MediaPipe Language identification does not suffer from the same limitations. It is also smaller in size (315kb compared to 1.1mb).
We can replace the current spoken language identification, or better yet, allow for options in the settings.
Keep the current identification. For spoken language change recommendation, don't show the suggestion until the text is "long enough"
fixed in https://github.com/sign/translate/commit/69641a41cc73d0eea05db8b035b4a732cdbe0d78
Problem
The current spoken language identification - implemented in https://github.com/sign/translate/issues/10 - has limitations, specifically, it struggles with short texts https://github.com/google/cld3/issues/76#issuecomment-1625233427
Description
The MediaPipe Language identification does not suffer from the same limitations. It is also smaller in size (315kb compared to 1.1mb).
We can replace the current spoken language identification, or better yet, allow for options in the settings.
Alternatives
Keep the current identification. For spoken language change recommendation, don't show the suggestion until the text is "long enough"