Pipeline Mode

Audio version — Estimated duration: 2 min 32 sec

Batch-process an entire directory of markdown files.

python tts_pipeline.py --dir ./path/to/docs

How it works

1. Scan — Recursively walks the directory, skipping hidden dirs and voice-notes/
2. Order — index.md first, then pipeline/, then everything else alphabetically
3. LLM rewrite — Sends each file to Cerebras or llama.cpp for conversational rewrite
4. TTS synthesis — Generates audio via edge-tts or Piper
5. Embed — Injects <audio> tag with duration into the source .md file
6. Save — Writes voice-friendly markdown to voice-notes/ (mirrored tree)

File ordering

Priority	Folder	Notes
1	. (root)	index.md is always first
2	pipeline/	Logical tutorial flow
3	Everything else	Alphabetical

The index file gets a special prompt: ~200 words and a suggestion to continue reading the next file.

Rate limiting

Online LLM mode enforces 5 requests per 60 seconds (sliding window). The pipeline pauses automatically — you will see [RATE] messages.

Arguments

See CLI Reference for all flags.

python tts_pipeline.py --dir ./docs --llm local --tts online --voice en-GB-SoniaNeural