Dhamma Transcriber & Extractor

A local pipeline that converts MP3 Dhamma talks into Markdown transcripts using MLX Whisper, then extracts core Dhamma points using an LLM.

Requires Apple Silicon (M1/M2/M3/M4). MLX Whisper does not run on Intel Macs or Linux.

---

Setup

/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
brew install ffmpeg
curl -LsSf https://astral.sh/uv/install.sh | sh
uv sync
./setup_folders.sh

Fill in .env (created by setup_folders.sh):

• API keys — GEMINI_API_KEY, OPENROUTER_API_KEY, or DEEPSEEK_API_KEY; antigravity-cli uses local CLI auth instead
• Provider/model routing is configured in tools/ai_models.json, not .env. Requests follow the cross-provider default_models fallback order there. To restrict routing to a single provider, set PROVIDER=<name> on the command line for that run (e.g. PROVIDER=agy uv run ...); a PROVIDER value in .env is ignored.
• Bio links: set BIO_EN and BIO_RU to append a speaker bio to YouTube descriptions. Leave empty (default) for no bio.

YouTube OAuth (first-time only): docs/upload-youtube.md

---

Transcription

Drop audio files (MP3, WAV, M4A, QTA, etc.) into input/sangha/ (or any other subfolder in input/) and run:

./scripts/cl/transcribe sangha

Output: output/transcribed/sangha/. Pass --correct to also run Pali correction, writing to output/corrected_pali/sangha/.

Full reference: docs/transcription.md

---

YouTube Pipeline

Drop MP3 or MP4 files into input/ and run:

./yt_run.sh --name "Tissa Thero"

--name sets the speaker name used for metadata context and artist metadata. Custom names are appended to generated review titles, except configured no-suffix names such as Ariyadhammika Bhikkhu.

The pipeline pauses three times:

1. Optionally add chapter names to the review file. Press Enter to continue.
2. Open reviews/english_review.md — fill in recording dates (DD-MM-YYYY), review titles/descriptions, set Approved: yes. Press Enter.
3. Review generated thumbnails in output/thumbnails/. Press Enter (or r to regenerate).

Done — MP4 videos are created and uploaded to YouTube.

Add --gdrive to also upload MP4/MP3 backups to Google Drive. Video files go under video/; audio files go under audio/. Selected Playlist is the only field that can add a subfolder inside those base folders. If multiple playlists are selected, the Drive upload asks which single subfolder to use. If Selected Playlist is blank, no extra subfolder is used.

Full reference: docs/pipeline-youtube.md

---

Other Pipelines

Pipeline	Doc
Dhamma extraction	docs/pipeline-dhamma-extraction.md
Google Drive upload	docs/upload-gdrive.md
Enhance control	docs/enhance-control.md
OpenAI Batch pipeline	docs/batch-pipeline.md