README.md

metadata

title: Haseeb's TTS
emoji: 🚀
colorFrom: indigo
colorTo: purple
sdk: streamlit
sdk_version: 1.54.0
python_version: '3.10'
app_file: app.py
pinned: false
license: apache-2.0
thumbnail: >-
  https://cdn-uploads.huggingface.co/production/uploads/652ac2e92aa5b27c77cba196/6Y7vGO0SQfVaCj9CYXzzf.png

🎧 Haseeb's TTS (Audiobook MP3 Generator)

Generate audiobook-style narration using Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice with a Streamlit UI built for long chapters.

Why qwen-tts instead of transformers.pipeline()?

The model uses the qwen3_tts architecture. Some Transformers builds in hosted environments may not recognize it. This Space uses Qwen’s official qwen-tts package which supports:

• generate_custom_voice(text, language, speaker, instruct, ...)
• get_supported_speakers() / get_supported_languages()

(As shown in Qwen’s official Qwen3-TTS repo docs.) :contentReference[oaicite:1]{index=1}

Features

• ✅ MP3 output (no ffmpeg needed)
• ✅ Batch mode: upload multiple .txt files → get multiple MP3s + ZIP download
• ✅ Long chapters (10,000+ chars) via chunking + stitching
• ✅ Language Support (dropdown; auto-populated from the model when possible)
• ✅ Voices / Speakers (auto-populated from the model when possible)
• ✅ Instruction Control (style/emotion/pacing)

How to use

Single chapter

1. Paste text (or upload a single .txt)
2. Choose language, speaker, instruction
3. Click Generate MP3

Batch mode

1. Switch to Batch mode
2. Upload multiple .txt files (each file = one chapter)
3. Click Generate MP3s (Batch)
4. Download the ZIP containing all MP3 outputs

Tips for audiobooks

• Chunk size: 1200–1800 chars is usually stable for long narration.
• Silence between chunks: 200–350 ms reduces audible joins.
• If memory is tight, reduce:
  • chunk size
  • max_new_tokens

Files

• app.py — Streamlit UI + batch mode + MP3 encoding + chunking/stitching
• requirements.txt — dependencies