Verified project record

denizsafak/abogen

Abogen converts ePub, PDF, text, markdown, and subtitle files into spoken audio and synchronized subtitles using the Kokoro-82M text-to-speech model. The application supports local execution on Windows, Linux, and macOS, with the ability to operate completely offline after models are downloaded.

Project overview

The application provides synchronized subtitle generation alongside audio conversion, supports custom voice mixing with adjustable weights, and enables completely offline operation after an initial model download phase.

Project type: Audio & Speech
Use cases: Translation & Subtitles · Audio & Speech
Deployment: Refer to project documentation
License: MIT

Best for

Creators and general users on Windows, Linux, or macOS needing to convert text-based documents into audio with synchronized subtitles, including environments requiring offline operation after initial model downloads.
Users who require custom voice mixing profiles or batch processing queues for managing multiple documents.

Key capabilities

Converts ePub, PDF, plain text, markdown, and subtitle files into audio using the Kokoro-82M model.
Generates subtitles synchronized with the audio output in various styles including sentence, word, or highlighting.
Allows users to create voices by mixing different voice models with adjustable weights and saving them as profiles.
Supports adding multiple files to a processing queue to convert several files in one batch with individual settings.
Optionally uses spaCy to detect sentence boundaries more accurately instead of relying solely on punctuation.
Allows downloading all required models and voices and disabling internet access for completely offline use during inference.
Planned feature to add OCR scan capability for PDF files using docling or tesseract.

Limitations and risks

The Web UI and Desktop GUI have different feature sets. The Web UI is more feature-rich but remains under active development.
AMD GPU acceleration via ROCm is not available on Windows for this application.

Getting started

Installation is rated as easy. Users can install via pip, uv, or an automated Windows script that does not require a separate Python installation. PyPI is used as an external service during installation, and HuggingFace Hub is used for model downloads. After installation, users can run abogen or abogen-web, drag and drop files, and start.

Evidence and sources

Open original GitHub project