R2T2 Real Real-Time Transcribe View source ↗
An abstract copper and blue sound signal flowing through darkness
Open source Code + weights live

R2T2: An open-source, high-performance streaming ASR model.

Real real-time transcription model, built for public use. Explore the open code, model weights, and hosted demo now.

Try demo ↗ Get the model ↗
ReleaseCode + weights
FocusStreaming ASR
StatusAvailable now
Video
01 Video comparison

Streaming ASR models,
side by side.

GPT-Live-Transcribe and R2T2 processing the same audio, shown together in real time.

Loading video…
02 Live replay

Hear speech
become text.

Replay short English and Mandarin clips with synchronized audio, spectrum, and committed transcripts.

Original recorded audio Synchronized transcript Committed updates
Explore more demonstrations
Recorded replay Real-time transcription 00:00.0
English · committed replay Original audio Committed output
Audio spectrum Source waveform
Committed transcriptSample 01

Start the replay and watch R2T2 turn speech into text in real time.

Replay progress
00:00.0 / 03:25.0
03 Open source
Open sourceLive now

Build with R2T2

Ready
when you are.

The repository, model weights, and hosted demo are live. Start with the resource that fits your workflow. The full technical report will be released separately.

Open codeGitHub ↗ Model weightsHugging Face ↗ Model weightsModelScope ↗ Hosted demoTry it ↗