60db.ai speech-to-text evaluation

Diarization on · language auto-detected · results saved on the server

Limits to know before testing

Max audio length: 10 min per file / per live session. Longer recordings must be split. The live session stops itself at 10:00.
Max file size: 10 MB (API reference). The features page says 25 MB, so the docs disagree; this page enforces 10 MB.
An uncompressed WAV hits 10 MB fast: 16 kHz mono ≈ 5.2 min, 44.1 kHz stereo ≈ 59 s. Keep Compress to fit on to convert to 16 kHz mono FLAC (lossless for speech) on the server.
Formats: WAV, MP3, FLAC, OGG, M4A, WebM, MP4 audio.
Concurrency: 8 requests per user on REST; live sessions 2 on the Free plan (REST and live count together).
Live: one turn is cut at 30 s of continuous speech; the socket closes after 50 s with no audio.
Languages: the upload API docs say these return unsupported_language: Urdu, Japanese, Korean, Chinese, Thai, Vietnamese, Indonesian, Tagalog, Swahili, Turkish, Persian, Hebrew. Yet the live socket advertises 45 languages, including Japanese, Korean, Chinese, Turkish and Hebrew. Worth testing both if you need them.

Bypass limit checks (send anyway, to find the real API limits)

Label Actual speakers Min speakers hint Max speakers hint

Leave the hints empty to measure how many speakers 60db finds on its own. Fill in Actual speakers so every result is scored against the truth.

00:00 / 10:00

Idle.

Compress to fit 10 MB (16 kHz mono FLAC)

No file selected.