Ambient scribes
AI scribe products that take visit audio in and hand a draft clinical note back. echo-1 is the transcript layer.
MedSpeech.io · echo-1 — speech-to-text API
Dictation-grade ASR, built for clinical documentation.
Streaming and batch speech-to-text for English medical dictation, with word-level timestamps and a HIPAA-track architecture. One API.
wss://medspeech.io/api/v1/speech-to-text/realtime · model=echo-1partial_transcript
committed — never retractedtentative — may still change
Who it’s for
For teams that turn clinical speech into documentation.
Scribe services
Human scribe services modernizing with ASR-assisted workflows — word confidence tells reviewers where to look.
Dictation & EHR tools
Dictation and documentation vendors, and tools that live inside the EHR.
Telehealth
Telehealth platforms adding visit transcription.
Accurate on medical speech
Trained on the language of the clinic.
echo-1 is our speech recognition model, purpose-built for English medical dictation. Metoprolol, auscultation, bilaterally — the vocabulary general-purpose ASR mangles is the vocabulary it was trained on.
Live and batch
Real-time transcription, and every visit after it.
Stream a visit and watch the transcript form in real time. Or hand us the recording afterward — any length, one request, one complete transcript.
Who said what: with multichannel audio — say a doctor mic and a patient mic — each channel is transcribed independently and labeled speaker_0, speaker_1. Take per-channel transcripts, or one combined transcript merged in time order.
Streaming
Partial words arrive instantly and settle into final text as the clinician speaks. Low-latency enough to feel live.
Batch
Whole recordings, transcribed with full context. Absolute word timestamps throughout.
Word-level timestamps
Every word carries start and end times — in streaming and in batch, where timestamps are absolute positions in the file.
Word-level confidence
Every word carries a log-probability, so review UIs can flag low-confidence spans for human attention instead of asking reviewers to re-read everything.
Compliance architecture
Compliance is the architecture, not a checkbox.
Audio and transcripts exist only in memory, only for the duration of a request. They are never written to disk, never logged, and never carried in billing events — logs and usage records contain ids, durations, and model names. Nothing else.
regional boundary · e.g. eu-west-1→→
Visit audio
in memory only
echo-1
processing
Transcript
returned, then gone
PHI never leaves the region. Connect to a regional endpoint and audio is processed in that region, guaranteed.
Built for HIPAA
Designed around HIPAA requirements. BAA available for design partners — talk to us.
Regional data residency
Pin a region at connection time. Only non-PHI metadata ever crosses regions.
Scale & reliability
Built to scale with you.
Scale automatically with concurrent streams and request rates, with further quotas available on request.
| Region | AWS region | Location | Status |
|---|---|---|---|
| US | us-east-1 | N. Virginia | preview |
| EU | eu-west-1 | Ireland | available on request |
| Asia | ap-southeast-1 | Singapore | available on request |
| Australia | ap-southeast-2 | Sydney | available on request |
Integrate in an afternoon
Change the base URL. Keep your code.
The API is wire-compatible with ElevenLabs speech-to-text: same xi-api-key header, same field and event names, same response shapes. Existing client code points at our base URL and works.
curl
curl -H "xi-api-key: $KEY" \ -F file=@visit.wav -F model=echo-1 -F use_multi_channel=true \ https://medspeech.io/api/v1/speech-to-text
Python SDK · torch-free
from medasr.client import MedASRClient
client = MedASRClient("https://medspeech.io/api", api_key="...")
result = client.transcribe("visit.wav", model="echo-1")
print(result.text)
with client.realtime(on_partial=print, on_committed=print) as s:
for chunk in mic_chunks():
s.send_audio(chunk)From an API key to your first transcript in minutes. Follow the quickstart →