The recording is transcribed on hardware we own, using a speech model that runs locally, not a third-party cloud API — and the original audio is destroyed on the job's own retention window, not kept.
Why this matters for some recordings
Most transcription services, free ones especially, keep a copy of your audio on their servers as part of how the product works. Fine for a podcast. Not fine for a deposition, an HR interview, or a clinical recording.
The actual numbers
Transcription runs on a speech model measured at roughly 3% word-error rate on clear English speech — tested against real recordings, not a vendor's marketing claim — and every job comes back with a completion report and a destruction certificate carrying the SHA-256 of the file you received.
Where this breaks down — said plainly
- Heavy accents, cross-talk, and poor audio quality all raise the error rate. True of any transcription system; disclosed rather than smoothed over.
- Tested on English read speech. Other languages and spontaneous conversational speech haven't been benchmarked to the same number.
Related guides
- Transcription that labels who's speaking
- Caption a podcast without uploading it anywhere public
- How file encryption during upload actually works
Where this fits on the site: this is the Transcribe a recording to text service under Audio transcription & captions pricing — published tiers starting at $199 for up to 2 hours, no quote, no sales call. See Services & Pricing or go straight to Choose your job.