README.md

metadata

title: AudioBook Forge
emoji: 🎧
colorFrom: indigo
colorTo: blue
sdk: gradio
sdk_version: 6.13.0
python_version: '3.12'
app_file: app.py
pinned: false
license: apache-2.0
short_description: AI audiobook generator with character voices via Qwen3-TTS

AudioBook Forge

Model-agnostic, high-fidelity audiobook generator powered by Qwen3-TTS. Create audiobooks where every character speaks with their own unique voice.

Features

• 📁 File Upload — Import EPUB, PDF, TXT, or HTML directly
• 📖 Chapter Detection — Auto-detects chapters/sections for selective generation
• 🎙️ Character Voice Mapping — Auto-extract characters and assign unique voices
• 🎭 Three Voice Modes for Every Voice
  • Preset — 9 premium built-in speakers (English, Chinese, Japanese, Korean, dialects)
  • Clone — Upload a 3–10 second voice sample to clone any real voice
  • Design — Describe a voice in text (e.g., "a raspy old man with a warm chuckle") and the AI creates it
• ⚡ Quick Generate — One-click audiobook from the Story tab with full voice customization (preset, clone, or design)
• 🎚️ Speed & Temperature Control — Adjust playback speed per voice (0.5x–2.0x) and generation temperature
• 📦 Multi-format Export — MP3, WAV, or ZIP of individual segments
• 💾 Save/Load Projects — Export and restore your voice configurations
• 🌐 10 Languages — English, Chinese, Japanese, Korean, German, French, Spanish, Italian, Portuguese, Russian
• ⚡ ZeroGPU — Runs on Hugging Face ZeroGPU (free A100/H200 compute)
• 🔧 Model Agnostic — Backend is swappable; upgrade to future SOTA TTS models without changing the UI

How to Use

1. Paste or upload your story in the 📖 Story tab.
2. Quick Generate (optional) — Generate a full audiobook immediately with a customized narrator voice.
3. Extract characters with the 🔍 button (AI enhancement is on by default for richer voice descriptions).
4. Configure voices in the 🎭 Voice Cast tab:
   • Set the Narrator voice (preset, cloned, or AI-designed)
   • Assign a voice to each Character (all default to AI-designed voices)
5. Generate in the ⚡ Generate tab and download your MP3, WAV, or ZIP audiobook.
6. Save your project in the 💾 Project tab to preserve voice configs for later.

Architecture

• app.py — Gradio frontend with dark-themed custom UI
• backend.py — Model-agnostic TTS engine, dialogue parser, and audio stitcher
• TTS Backend: Qwen3-TTS 1.7B (CustomVoice + Base + VoiceDesign)
• Text Processing: Paragraph-aware chunking, sentence-boundary splitting, quote detection
• Audio Pipeline: Per-segment synthesis → crossfade stitching → peak normalization → MP3 export

License

The application code is Apache 2.0. The underlying Qwen3-TTS models are also Apache 2.0, making this stack fully commercially usable.