how it works · timestamps
Podcast audio changes after it's published. Our timestamps don't drift with it.
Most transcripts are cut against the audio file that existed on the day of transcription. The file you stream a month later is often a different one. Podmenti notices, and re-aligns automatically. It is included on every plan, Free too.
The problem nobody tells you about
Nearly every large podcast is served through dynamic ad insertion. When you press play, the host (Megaphone, Art19, Acast, Omny and the rest) assembles the episode on the fly: the show's audio, plus whichever ads are booked for you, right now, in your country. Next week a different campaign runs, a mid-roll gets longer or shorter, and the file is a few seconds different from the one your transcript was cut against.
A few seconds does not sound like much. It is enough to land a quote on the wrong sentence, put a subtitle a beat late, and make a "jump to 41:07" link open on the ad instead of the answer. We measured real episodes that had drifted between 10 and 42 seconds within weeks of publication. Every ad break in front of a moment adds its own offset, so drift grows the deeper into an episode you go.
What we do about it
- 01 We fingerprint the audio at transcription time.
Every transcript stores a content hash of the exact audio file it was cut against, ads included.
- 02 We keep checking the file the host is serving now.
A tiny request, no download, tells us whether the current file still matches that hash. Any change to any ad, pre-roll, mid-roll or post-roll, changes the hash. Active shows are checked daily; quiet back-catalogue backs off to a fortnightly check.
- 03 When it changed, we re-transcribe only what moved.
Rather than redoing a two-hour episode, we re-run recognition on short windows around the affected sections, match the words we already have, and shift the timing to where those words now sit. It is around five hundred times cheaper than a full re-transcription, which is why we can afford to do it for every episode, forever.
- 04 Every timestamp updates together.
Segment times, word-level times, the .srt and .vtt exports, the API and the "listen from here" links all read the same corrected timing. Nothing you already exported has to be re-bought: exports are permanent and always serve the current alignment.
Why not just use the ad-free file?
Because it would be dishonest to the show. Creators earn their living from those ad slots, and the ad-free origin file is not what any listener actually hears. We transcribe the episode as it is really served, credit the host for every stream, and do the extra work on our side to keep the words and the clock in agreement.
What this means for you
- Deep links land where they should. A timestamp from AI Search or an export points at the moment, not at whatever ad happens to be running.
- Subtitles stay in step. .srt and .vtt files cut against the live audio, not a stale copy.
- It costs you nothing extra. Re-alignment is part of what an unlocked episode is here, on every plan, including Free.
- The words never change. Re-alignment moves timing only. The transcript text you read, searched or exported is stable.