zoq
commented
10 months ago If I understand you correctly, you would delay the playback by e.g. 10 seconds and use the 10 seconds as context, instead of doing it in a live setting directly from the beginning? Similar to caching a video, so the playback is continuous.
We haven't thought about it, but I agree in certain scenarios this would be helpful and improve the user experience. We are currently working on a list of features, we will add it to list.
Let us know if this is what you head in mind.
kirawi
fallenangel3k
What I mean is that the transcriptions up to e.g. 10 seconds of where the video is currently playing are computed. This is helpful for languages such as Japanese (and presumably Chinese) where context determines which word it is (e.g. the variations of かえる or まく). I presume that this behavior is why the transcription is constantly being modified rather than solely appended to. It is very jarring. Transcribing ahead of time should give enough context to significantly reduce how often this happens. https://github.com/vantezzen/skip-silence does something similar I believe to debounce how often videos get sped up based on spans of silence.
I'd also suggest erasing past transcriptions after a period of silence, but I presume this might already be implemented?