@overcastfm Congrats!

Transcription for Portuguese works, but sentence breaks are nonexistant (likely not your fault). It appears to force a break after 60–100 words — is that your doing, or the model’s? If it’s you, a few suggestions:

• The algorithm appears to add breaks immediately after the letter sequence ends, even there’s punctuation afterwards — would be better to break at whitespace instead
• In fact, why not scan for a recent sentence break (period, exclamation or question mark followed by whitespace) rather than breaking mid-sentence? To avoid excessively short lines, it could limit to the second half of the excessively-long line (e.g. most recent 30–50 words)
• The extra effort solution: if you have timings for individual words, find the longest pause after the end of a sentence, and break there (or, if you have voice change information, break when the speaker changes)

@overcastfm
Great! The transcript highlighting and synch with the podcast is smooth and looks perfect to me.

The Apple transcript quality is very good, and its music markers are icing on the cake! 🍰

@overcastfm Congrats on the release. The feature is awesome!

Suggestion: add an option to locally re-transcribe episodes (ignoring your datacenter transcription).

Sometimes the transcript is unsynced (usually because a misdetected dynamic ad) and it would be a useful plan B.