Well, it uses Google's speech recognition, so it might improve. I think it (used to?) chop up the sound into chunks to send to Google without checking if it was in the middle of a sentence and place the timestamps just based on which chunk it was in. The Cloud Speech API supports files longer than a minute only if they're uploaded to Google's own cloud storage and can return word-by-word timestamps. I wish I had a week to do some coding
It seems it's very sensitive to the quality of the source audio. I tested with a scene from RCT-152, one of the bukkake announcer series, and got a translation where at least the majority of the sentences make sense. Of course, given the results I've had when just google translating movie titles, porn dialogue might not work well in general.
It seems it's very sensitive to the quality of the source audio. I tested with a scene from RCT-152, one of the bukkake announcer series, and got a translation where at least the majority of the sentences make sense. Of course, given the results I've had when just google translating movie titles, porn dialogue might not work well in general.