More reliable imports for phone audio and irregular timestamps

Some phone recordings and variable-bitrate MP3 files do not store duration in the place a typical audio file does. The audio can play normally while a basic duration check still reports nothing useful.

Pepys now checks the audio track first, then the wider file information. If neither contains a usable duration, Pepys measures the decoded audio as a final bounded check before deciding that the recording cannot be read.

A video container can be perfectly valid and still contain no audio track. Pepys now distinguishes that case from an unreadable duration and stops before trying to create a transcript from silence.

The result is a clearer correction path: choose a version with an audio track, or export the recording's sound and upload that file instead.

Word timestamps are estimates, and adjacent words can occasionally arrive a fraction out of order even when the text and segment timing are usable. Pepys now restores those valid words to chronological order and keeps them attached to the transcript.

Slight timing drift at the edge of a recording is also brought back to the measured audio boundary. The readable transcript survives instead of failing because an optional word marker landed just beyond the final moment.

These checks recover valid media with unusual metadata; they do not pretend that every file contains usable audio. Empty recordings, corrupt data, and impossible transcript timing still stop with an error rather than producing a misleading result.

If a file still fails, confirm that it plays from beginning to end and export it once as MP3, M4A, or WAV before trying again. Re-exporting rebuilds the audio information that damaged downloads often lose.