How accurate are the transcripts?

What “95%+ on clear audio” means

We measure this monthly against human reference transcripts: see the live benchmark for the current numbers by recording condition.

On a recording with clear speech and a decent microphone, expect 95 or more of every 100 words to be right, with speaker labels and per-word timestamps. We do not quote a higher number, because the audio itself decides the outcome more than the software does.

What pulls accuracy down

• Background noise: traffic, music, an echoing room.
• Overlapping speakers: people talking over each other is hard for any transcription system.
• Heavy accents and mumbling: the further speech is from the microphone, the more this costs.
• Compressed phone audio: voicemails and call recordings transcribe well, but below studio quality.

We would rather tell you this upfront than promise perfection. If your audio is difficult, the free preview will show you exactly how it handles it, before you pay.

Finding and fixing mistakes

The transcript view highlights words the engine was less sure about, so your review time goes where it matters instead of into rereading everything. The built-in editor lets you correct any word and hide filler words, and rename speakers on any plan. Edits carry into every export.

Getting the best result from your side

• Record close to the speakers when you can.
• Upload the original file rather than a re-compressed copy.
• Use the free 10-minute preview on your worst audio, not your best. If it reads well, you have a good picture of how the rest of the file will come out.

And if the quality still lets you down?

We refund your purchase in full, no questions asked. How refunds work.