audiobook generator.stub

---

id: audiobook-generator name: Audiobook Generator repo_path: ~/Documents/sw30labs/repos/audiobook_generator remote_url: https://github.com/sw30labs/audiobook_generator.git primary_language: Python framework: LangGraph + Qwen3-TTS ingested: 2026-04-11 last_commit_date: 2026-04-11 category: local-inference-mlx stacks: [langgraph, agentic]

Audiobook Generator

Purpose

Convert books to audiobooks using Qwen3-TTS and LangGraph. Multi-format support (RTF, TXT, MD, HTML, DOCX), smart chapter detection, semantic chunking, TTS tag preprocessing, quality verification via faster-whisper STT, checkpoint-based resumption, parallel generation.

Supported Formats

• RTF
• TXT
• Markdown
• HTML
• DOCX

Architecture

LangGraph StateGraph with workflow nodes:

1. Convert: Format-specific parsing (RTF → striprtf, DOCX → python-docx, HTML → BeautifulSoup)
2. Conversion QA: Verify parsed content via Ollama QA agent
3. Split: Detect chapters via regex and semantic markers
4. Chunk: Semantic chunking for paragraph grouping
5. TTS Preprocess: Clean markup, normalize whitespace, format tags
6. TTS Generate: Qwen3-TTS with parallel synthesis
7. Audio QA: Verify audio quality via faster-whisper STT roundtrip
8. Merge: Combine audio chunks, normalize levels, add metadata

Key Features

• Smart chapter detection
• Semantic text chunking
• TTS tag preprocessing
• Quality verification (whisper-based STT)
• Checkpoint-based resumption
• Parallel generation

Dependencies

langchain, langgraph, langchain-ollama, striprtf, markdownify, python-docx, beautifulsoup4, soundfile, librosa, numpy, scipy, faster-whisper, typer, rich, tqdm, pydantic, qwen-tts, torch

Tags

python, langgraph, mlx, apple-silicon, local-inference, tts, stt, audio

Cross-references

• tars-ai — shared TTS domain
• mlx-youtubescribe — shared audio processing