Skip to content

v0.2.0 — Multilingual support + interactive speaker labeling

Choose a tag to compare

View all tags
@pretyflaco pretyflaco released this 14 Mar 00:12
· 144 commits to main since this release
v0.2.0
494c53a

When v0.1.0 went up, a lot of people responded. 500+ likes, 600+ bookmarks on X, 172 GitHub stars. The thing that kept coming up in replies was simple: people are buying RTX cards for local AI and asking what they should actually do with them. meetscribe is one answer — a meeting recorder where the audio never leaves your machine, no subscription, no cloud API, no training data.

v0.2.0 builds on that with two features people asked for most.

What's new

Multilingual support

Whisper large-v3-turbo already handles 99 languages — meetscribe now plumbs that through the entire pipeline. Auto-detection is the default. You can also pin a language:

meet run --language de    # German
meet run --language tr    # Turkish
meet run --language fr    # French
meet run --language es    # Spanish
meet run --language fa    # Farsi (RTL)

The summary prompt adapts to the detected language, the PDF uses DejaVu Sans for full Unicode coverage, and Farsi gets right-to-left rendering via Noto Naskh Arabic + arabic-reshaper.

Interactive speaker labeling

meetscribe auto-assigns YOU / REMOTE_1 / REMOTE_2 from the dual-channel signal. Now you can replace those with real names:

meet label ~/meet-recordings/meeting-20260313-214133

For each speaker: shows a table, plays an audio clip from that person's channel, prompts for a name. All outputs (txt, srt, json, summary, pdf) are regenerated. The GUI also shows a labeling dialog automatically when 2+ speakers are detected.

Changelog

  • feat: multilingual transcription, alignment, summary, and PDF (99 languages)
  • feat: Farsi RTL support in PDF (pip install "meetscribe-offline[rtl]")
  • feat: meet label CLI command — interactive post-hoc speaker naming with audio playback
  • feat: GUI speaker labeling dialog — appears after transcription with 2+ speakers
  • refactor: meet/audio.py, meet/languages.py, meet/utils.py, meet/label.py extracted as shared modules
  • refactor: post_process() centralises all output generation; GUI and CLI share the same code path
  • build: PyPI package renamed to meetscribe-offline
  • test: 81-test suite added

Install

pip install meetscribe-offline

Requires Linux, NVIDIA GPU (8 GB+ VRAM), ffmpeg, and a free HuggingFace token for diarization.

Assets 2
Loading