83languages, detected automatically. Nothing to configure.
17 of them come back with speaker labels, the who said what in your transcript. 18 run on our newest model, with dialect-aware transcription.
Where the supported languages are spoken
Highlighted: a supported language is spoken here
Highlighted: an Enhanced language is spoken here
- 83 languages
- 17 with speaker labels
- 18 enhanced
Supported languages
Detected automatically on every call.
Enhanced
All other languages
How well does it work in my language?
Clear audio
A headset beats a laptop microphone by more than any two languages differ from each other. Room echo and background music are what cost you words.
One speaker at a time
Crosstalk is the hardest thing to transcribe in any language. Two people talking over each other costs more accuracy than an unfamiliar accent.
Its own script
Your transcript comes back in the script the language is written in, the way the atlas above prints each name, not transliterated into Latin letters.
Accuracy details
Word error rate is the share of words a person would correct. These are the bands the transcription provider publishes, across the full list.
- Enhanced18 languagesNo published WER
- The newest model we run, with dialect-aware transcription. No published word error rate yet.
- Excellent5 languages10% WER or better
- About one word in ten or fewer needs a correction.
- Strong18 languages10 to 25% WER
- Corrections cluster on names and jargon. The thread of the call stays clear.
- Good14 languages25 to 50% WER
- Readable notes from clean audio. Worth a skim before you share the transcript.
- Supported28 languagesAbove 50% WER
- Accuracy varies with the audio. Best with a headset and one speaker at a time.
Ready to start recording?
NotesBot automatically detects your language. Just start recording. Setup takes less than a minute.