Record or upload. Everything else is detected
A clinical or behavioral-health conversation in English or Arabic (Saudi, Gulf or Modern Standard) goes in. Medera BH-ASR detects the language, finds the speakers and reads who is the clinician and who the patient, then every open-weight recogniser on this Space transcribes the same speech and is scored live.
- English Β· Ψ§ΩΨΉΨ±Ψ¨ΩΨ©
- Speakers and roles detected
- Open-weight recognisers
- Never written to disk
Your recording
- The microphone records only while the button is on, and nothing leaves your browser until you press Analyze.
- The raw signal is kept: no noise suppression or automatic gain, so the recognisers hear your voice, not the browser's processing.
What happens next
- Language, for every stretch of speech
English or Arabic, or a mix of both, detected as it is spoken. Nothing to choose.
- Speakers and their roles
Voices are told apart, then what each person says, in English or in Saudi, Gulf or Modern Standard Arabic, shows who asks as the clinician and who answers as the patient.
- Every recogniser, the same speech
Medera BH-ASR first, then each open-weight recogniser on this Space, one after another.
- Scored with no transcript needed
Each recogniser against the recognisers independent of it: agreement, labelled as such. Accuracy against human transcripts is on the leaderboard.
Up to 3 minutes Β· English and Arabic Β· held in memory only, never written to disk
Record or upload a recording.
Result
Which recogniser keeps what matters clinically
Every recogniser on every test clip in its languages, one CPU runtime, one scoring method. Lower word error rate is better; alerts kept counts the safety statements that survive transcription.
Leaderboard
Method and live reproduction
Methods, models and integrity checks
How a recording is benchmarked
Recognisers and licences
Every model file checked, every example run end to end
When the Space starts, it checks every model file against the SHA-256 the engine pins, then runs every example through BH-ASR exactly as your recordings run, and scores each transcript against its script. You can run it again.
How this Space is running
Computed from the runs this Space has served since it started. Refreshed every 15 seconds.