01 Streaming
Word by word
Tokens land as they are spoken, not after the sentence closes.
Transcription that holds up in a loud room.
NoiseScribe streams your microphone through OpenAI Transcribe and Live Transcribe, gates the noise floor before inference, and returns speaker-labelled text in under 400 ms.
- Handles accents, crosstalk and mixed languages
- Audio is never retained after the session closes
- Word-level timestamps on every export
Recording
12:04 / 41:22
IN −4.2 dBFS
GATED −18 dB
LATENCY 312 ms
Reyes 11:22
Before. We run spectral subtraction and an RNN suppressor on the buffer, then hand a clean 16-kilohertz stream to the transcriber, so —
Live
so the noise floor drops about eighteen decibels before the model
60min
Free every month, no card
2
Engines — live as you speak, archival when you stop
6
Export formats, timestamps included
0
Recordings kept once a session closes
Everything a transcript needs
No upload step, no round trip to a separate editor.
02 Diarisation
Who said what
Up to twelve voices separated and labelled; rename them mid-session.
03 Engines
Two models, one switch
OpenAI Transcribe for archival accuracy, Live Transcribe for latency.
04 Noise
Built for bad rooms
HVAC, a projector fan, two people talking over each other — the transcript holds up.
05 Playback
Scrub locked to text
Every word carries a timestamp; click it to hear the audio behind it.
06 After
Summary, actions, chapters
Generated from the finished transcript — not from a sample of it.
Built for the way people actually record
Whatever you’re capturing, the transcript keeps up.
Export
- SRT
- VTT
- DOCX
- TXT
- JSON + word timings
Pricing
Run over? Extra hours are $2.20 each on any paid plan — no session is ever cut off mid-recording.