feat: added deepgram for transcription

[masterchief164]

This PR fixes #96.

[adamjonas]

In testing it out, whisper is much more accurate than the default nova model. I'd suggest leaving the current whisper code as is and adding a flag to use deepgram but be able to pass it various models.

One of the problems is that the chapters breaking up the timestamps. Instead of breaking up the file, I'd suggest adding the headers by locating the timestamp and keeping the file intact. That might be difficult without granular timestamps like deepgram provides, but it should be possible even with the whisper srt format. If we keep the file intact, we also should get summarization via deepgram which can close #38.

Additionally, the default should be with the speaker diarization off and turn it on via flag and/or adding multiple speakers to the speakers metadata.

[masterchief164]

One of the problems is that the chapters breaking up the timestamps. Instead of breaking up the file, I'd suggest adding the headers by locating the timestamp and keeping the file intact. That might be difficult without granular timestamps like deepgram provides, but it should be possible even with the whisper srt format. If we keep the file intact, we also should get summarization via deepgram which can close #38.

Could you please provide an example of the first part.

[adamjonas]

https://github.com/adamjonas/bitcointranscripts/pull/113/files

[adamjonas]

testing the chapters

tstbtc '1Z7GjXgdUy4' test-dir -p -D -C produces https://github.com/adamjonas/bitcointranscripts/pull/146/files and https://github.com/adamjonas/bitcointranscripts/pull/145/files, which both cut off before transcribing the last chapter of the video.

Otherwise this is ready to merge.

EDIT: 025e519 should fix it

[masterchief164]

This PR also fixes #38