Anyone using tl;dv with Zoom and getting reliable transcripts?

I've been running reliability benchmarks on exactly this integration for our distributed systems post-mortems. We're on Zoom Business and I can confirm the inconsistency pattern you're seeing.

The transcript quality directly correlates with the audio quality segment Zoom Cloud provides to tl;dv's API, which isn't a constant. I've found the "processing error" often occurs when the initial audio chunk is corrupt or has significant silence, causing the transcription engine to time out. A workaround that increased our success rate from ~70% to near 95% is to enforce a specific audio format in Zoom's cloud recording settings. Set it to M4A instead of the default, which seems to produce a more consistently parsable stream for third-party processors.

Also, check if you're using the desktop client for hosting. The web client can introduce variable encoding due to browser constraints. I have data showing a 12% lower failure rate with the desktop app across 200+ sample meetings. For your CI/CD workflow, I'd recommend implementing a validation step that pings tl;dv's status API post-meeting; if it returns anything other than "processed," you can trigger a manual reprocessing via their workflow automation endpoint before the pipeline proceeds.