One-click local PDF/TXT/DOCX/EPUB to audiobook studio for Windows + RTX 3050 4 GB laptops.

The goal of v4 is simple: after you create a default reference voice one time, you can drop in a PDF and press Start / Resume Audiobook. The app automatically chooses safe defaults:

• Smart PDF text extraction with OCR fallback per page.
• OCR cache, so OCR is not repeated when you resume.
• Safe XTTS chunks below the English 250-character warning.
• Resumable chunk generation. If Windows sleeps or the GPU resets, run the same job again.
• Failed chunk skip/retry so one bad line does not destroy a 1000-page job.
• FLAC part export instead of one giant file.
• Natural room-tone gaps, ACX-style mastering target, and 44.1 kHz MP3 delivery support.
• Gradio one-click interface.

cd D:\download\voicebook_studio_pro_v4_ideal\voicebook_studio_pro
& "D:\download\elevenlabs_style_tts_lab\elevenlabs_style_tts_lab\.venv\Scripts\Activate.ps1"
python -m pip install -e .
voicebook doctor
voicebook ui

Open the URL shown by Gradio, normally:

http://127.0.0.1:7860

If you already have a reference at data/ui_reference.wav, you can skip this. Otherwise create one:

voicebook reference-slice "D:\download\my_voice.mp3" --out data\ui_reference.wav --start 20 --duration 15

Then test:

voicebook speak --text "This is a studio narration test." --out outputs\test.wav --emotion studio --master

voicebook oneclick "D:\download\book.pdf"

This uses the saved reference voice automatically and writes output to:

outputs/<book_name>_audiobook/

voicebook audiobook "D:\download\book.pdf" --out outputs\book_audiobook --emotion studio --format flac --ocr auto --chunk-chars 220 --part-segments 80 --fail-policy skip

Use the defaults. The job is resumable. If stopped, run the same command again with the same output folder.

Files created:

outputs/book_audiobook/chunks/chunk_000001.wav
outputs/book_audiobook/parts/part_0001.flac
outputs/book_audiobook/state.json
outputs/book_audiobook/manifest.json
outputs/book_audiobook/ocr_cache/

Readable PDFs do not need OCR. Scanned/image PDFs need Tesseract OCR installed.

Try:

winget install -e --id UB-Mannheim.TesseractOCR

If WinGet gives 403, install Tesseract manually from the UB Mannheim Windows build page, then restart PowerShell/VS Code.

Use studio for the default audiobook style. It is intentionally faster and less draggy than old story/warm presets.

Recommended outputs:

• flac: best practical audiobook master.
• wav: editing/training master.
• mp3: delivery/share version, 44.1 kHz mono 192 kbps CBR.

Only clone your own voice or voices where you have permission. Check model and book licenses before publishing or selling generated audiobooks.