Skip to content

Add scribe stream command for live microphone transcription #1

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
javiertoledo wants to merge 13 commits into
base: main
Choose a base branch
from
Draft

Add scribe stream command for live microphone transcription #1

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
13 commits
Select commit Hold shift + click to select a range
161c508
Add scribe stream command for live microphone transcription
javiertoledo Mar 30, 2026
4ad5e71
Fix stream latency, speaker label, and dedup
javiertoledo Mar 30, 2026
84af5b0
Fix stream: use library streaming config, show volatile + confirmed u…
javiertoledo Mar 30, 2026
e08ff0b
Switch stream to Nemotron 560ms — true streaming ASR engine
javiertoledo Mar 30, 2026
2d95652
Default stream to multilingual engine, Nemotron as opt-in for English
javiertoledo Mar 30, 2026
8f9acdb
Add model download retry, progress logging, and streaming docs
javiertoledo Mar 30, 2026
3c93c3d
Fix Nemotron streaming: emit delta text, not full accumulated transcript
javiertoledo Mar 30, 2026
771dcc6
Add --audio-file streaming test harness with eval baseline
javiertoledo Apr 9, 2026
4bcf7e8
Fix streaming content loss: use finish() result + 1120ms variant
javiertoledo Apr 9, 2026
ac2fc01
Re-run batch eval baseline — verify no regression after Stream.swift …
javiertoledo Apr 9, 2026
11dc099
Unify mic and file streaming paths through one AsyncStream pipeline
javiertoledo Apr 9, 2026
a7c3967
Warm up CoreML engine before starting mic capture
javiertoledo Apr 10, 2026
40903c6
Flowing text output + save audio for batch re-transcription
javiertoledo Apr 10, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 17 additions & 2 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -72,6 +72,19 @@ scribe transcribe meeting.wav --language es # Spanish
scribe transcribe meeting.wav --language fr # French
```

### Live streaming from microphone

```bash
scribe stream # multilingual (25 languages, ~11s latency)
scribe stream --engine nemotron # English-only, low latency (~560ms), with punctuation
scribe stream --format jsonl # JSONL output (one JSON object per line)
scribe stream --output meeting.txt # save to file while streaming
```

> **Note on streaming engines:**
> - **Default** (Parakeet TDT v3): Supports all 25 languages including Spanish. Higher latency (~11s for confirmed text) because it uses a batch model in sliding windows. Live preview appears on screen while speaking.
> - **Nemotron** (`--engine nemotron`): English-only but much faster (~560ms latency). Includes punctuation and capitalization. Recommended for English-only meetings.

### Pre-download models

```bash
Expand Down Expand Up @@ -143,8 +156,10 @@ Tested on Apple Silicon (M-series):
|------|-------|---------|
| Transcription only | ~130x real-time | 4-min file in 1.7s |
| Transcription + diarization | ~30x real-time | 4-min file in 7.5s |
| Live streaming (Nemotron) | ~560ms latency | English, with punctuation |
| Live streaming (default) | ~11s latency | 25 languages |

Models are downloaded automatically on first use (~600MB for ASR, ~50MB for diarization).
Models are downloaded automatically on first use (~600MB per model). First run may take a minute.

## Requirements

Expand All @@ -159,7 +174,7 @@ Bulgarian, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, G

scribe is built on the shoulders of excellent open-source projects:

- **[NVIDIA Parakeet](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3)** (CC-BY-4.0) — The speech recognition model that powers transcription
- **[NVIDIA Parakeet](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3)** (CC-BY-4.0) — Speech recognition models (TDT v3 for batch, Nemotron for streaming)
- **[FluidAudio](https://github.com/FluidInference/FluidAudio)** (Apache 2.0) by FluidInference — CoreML speech processing SDK for Apple Silicon
- **[pyannote.audio](https://github.com/pyannote/pyannote-audio)** (MIT) by Herve Bredin — The diarization model architecture
- **[swift-argument-parser](https://github.com/apple/swift-argument-parser)** (Apache 2.0) by Apple — CLI argument parsing
Expand Down
Loading