Accurate 'sentence' timings, is it now possible with new word-level timestamps update?

[erturkdotgg]

There has been a long-standing timing issue in Whisper.cpp. In a normal video transcription for example(for regular subtitles), there is no gap between two lines. In the transcribed text, when the previous subtitle line ends, next line starts appearing on the screen before the speaker actually begins to speak the second line. However, this issue does not exist in other Whisper ports, e.g., WhisperX. Is it possible to solve this issue? Or perhaps with a custom alignment model?

Example:
Whisper.cpp Timing:
00:00:00 - 00:00:20: Hello everyone
(no gap, silence)
00:00:20 - 00:00:33: How are you?

Proper timing:
00:00:15 - 00:00:20: Hello everyone
(proper gap, silence)
00:00:27 - 00:00:33: How are you?