lukaszliniewicz/Pandrator

Turn PDFs and EPUBs into audiobooks; subtitles or videos into dubbed videos (including translation), and more. For free. Pandrator uses local models, including voice-cloning (instant, RVC-enhanced, XTTS fine-tuning) and LLM processing. It aspires to be a user-friendly app with a GUI, an installer and all-in-one packages.

Pandrator – Local‑first audiobook, subtitle & voice‑over studio

What it is – Pandrator is a desktop‑style web application that lets you turn text (books, PDFs, webpages, etc.) or existing media (videos, audio files) into audiobooks, subtitles, or dubbed voice‑overs. It bundles the whole workflow – document import, optional OCR, text cleaning, transcription, AI‑assisted correction/translation, text‑to‑speech generation, review, and export – into a single browser UI.

Why it matters – All heavy‑weight speech and transcription models can run locally, so your source documents and generated audio never have to leave your computer unless you explicitly enable a cloud provider. This gives privacy‑focused creators a way to produce high‑quality spoken media without a permanent internet connection.

Core capabilities

Area	What Pandrator does
Audiobook creation	Import TXT, PDF, EPUB, DOCX, MOBI or paste text; auto‑detect chapters; split for synthesis; generate segment‑by‑segment speech; optionally run voice‑conversion (RVC); export to WAV/MP3/Opus/FLAC/M4B with chapters and cover art.
Subtitle & dubbing	Load SRT (or WebVTT/ASS/SSA) or raw audio/video; transcribe with timestamps & diarisation; edit/translate subtitles side‑by‑side; generate synchronized dubbed speech; export subtitles alone or as a dubbed video with selectable audio/subtitle tracks.
Speech generation	Choose from local TTS models (e.g., Kokoro, Qwen3‑TTS, XTTS‑v2, VoxCPM2, Silero, etc.) or cloud services (OpenAI, Google Gemini, ElevenLabs, custom endpoints). Supports voice cloning, multilingual voices, and optional RVC voice‑to‑voice conversion.
AI‑assisted text work	Optional LLM providers can be used for subtitle correction, glossary‑aware translation, document cleaning, and pronunciation optimisation. Each AI step creates a separate, reviewable revision so nothing is overwritten silently.
Pronunciation library	Deterministic, user‑maintained “search‑and‑replace” rules that are applied only to the payload sent to the TTS engine, leaving the displayed text unchanged.
Model management	Pandrator Manager installs, updates, and removes speech/transcription components. It shows compute requirements, licences, and download sizes before any change.
Remote / headless use	The service listens on `127.0.0.1` by default but can be exposed via a reverse‑proxy (HTTPS) for LAN, VPN, or cloud‑pod deployments.
Agent integration	A side‑car (`pandrator‑mcp`) lets an MCP‑compatible AI agent inspect installations, run workflow plans, and perform limited recovery actions with credential‑store integration.

Getting started (quick‑start)

Download the Manager – Windows: PandratorManager‑0.9.17‑windows‑x86_64.exe; Linux: PandratorManager‑0.9.17‑x86_64.AppImage.
Run the Manager, pick a parent folder for the workspace, and let it install Pandrator.
Open the browser UI that the Manager launches.
In Providers & services, install the local models you need (e.g., Kokoro for a lightweight voice, CrispASR for transcription). Cloud providers are optional.
Create a new project, import your source document or media, and follow the step‑by‑step panels (clean → segment → generate → review → export).

Technical stack (as described in the README)

Python 3.11 – core backend, worker processes, CLI.
Node.js (locked via the web-build Pixi environment) – builds the browser UI.
Pixi – package manager that handles both Python and Node dependencies and provides reproducible environments.
AppImage / native Windows exe – bundled runtime, no external OS packages required.
Optional external tools – yt-dlp for URL imports, Calibre for MOBI conversion, system credential stores (Windows Credential Manager, macOS Keychain, Linux Secret Service) via the credential‑stores extra.

Typical use cases

Self‑publishing authors who want to generate an audiobook from a manuscript without uploading the text to a cloud service.
Educators / podcasters creating narrated versions of lecture slides or PDFs.
Video creators needing accurate subtitles and optional dubbed tracks in multiple languages.
Privacy‑conscious users who prefer local speech models (e.g., Whisper‑based CrispASR, XTTS) but still want the convenience of a unified UI.

Limitations & caveats

Model size vs. hardware – Larger voice‑cloning models (VoxCPM2, Fish S2 Pro) require a CUDA‑capable GPU; smaller models can run on CPU but will be slower.
Multilingual inline code‑switching – Not fully supported; you must split foreign language segments manually.
Remote deployment – Possible but requires manual reverse‑proxy configuration and explicit enabling of non‑loopback listeners.
ElevenLabs integration – Native API only; no validation without an API key.
Single‑owner design – Not built for multi‑user SaaS scenarios.

License & contribution

The core Pandrator code is released under the MIT License. Third‑party models and services retain their own licences.
Contributions are welcomed via GitHub Issues and pull requests; the project asks for focused changes with tests where practical.

Bottom line – Pandrator is a genuine, open‑source, locally‑runnable platform that brings together transcription, AI‑assisted text processing, and modern TTS/voice‑cloning models into a single, easy‑to‑use web interface. It is well‑suited for creators who need privacy‑preserving audio production without stitching together separate tools.