This documentation explains how to install, use, operate, and understand Pandrator. Choose the path that matches the result you want; you do not need to read it in order.
Read the Pandrator 0.10.1 release notes for the latest transcription and export fixes. Previous notes: 0.10.0.
| Goal | Guide |
|---|---|
| Install or update Pandrator | Installation |
| Make an audiobook | Your first audiobook |
| Transcribe, correct, or translate subtitles | Your first subtitles |
| Transcribe a file or microphone recording without a session | Quick transcription |
| Create a synchronized voiceover | Your first voiceover |
| Choose a model, provider, or voice | Providers and voices |
| Decide how correction or translation should run | Correction and translation |
| Use the model already running in an MCP host | Passive dispatch |
| Take a local file through an agent-run workflow and return deliverables | End-to-end agent workflows |
| Connect Codex, OpenCode, Claude Code, or Antigravity | Agent connections |
| Fix names or prepare text specifically for speech | Pronunciation and speech text |
| Diagnose a problem | Troubleshooting |
- Updates, data, repair, and removal
- Remote and headless deployments
- Agent connections and MCP transports
- Troubleshooting
- Privacy and security
Exact Manager commands, component recipes, recovery behavior, and automation interfaces live in the Pandrator Manager guide.
- Supported formats and exports
- Document ingestion and narration pipeline
- Subtitle-to-speech pipeline and parameters
- Speech directions, dialogue, and character voices
- Speech-text optimization and dispatch
- Voiceover repair history and undo
- Local model groups, Qwen transcription, and audio preprocessing
- Qwen forced alignment for multilingual captions
- Contextual performance planning and provider controls
The document reference covers upload lineage, PDF layout/OCR, EPUB structure, cleanup, narration preparation, and generation segments. The subtitle reference explains the durable distinction between timed words, display cues, LLM batches, speech blocks, generation takes, and alignment. The speech reference compares standalone, generation-time, and passive optimization and documents their batching and validation contracts.
The Pandrator MCP guide is the canonical reference for installing the sidecar, enrolling targets, selecting scopes, generating host configuration, diagnostics, protocol compatibility, and app-down Manager recovery. Public workflow docs explain why and when to use MCP; the component guide owns exact commands and security contracts.
The files under pandrator_mcp/guides/ are packaged, versioned instructions
served to agents by the MCP server. They are not a second public documentation
site and should not be moved or copied into this directory.
- Run Pandrator from source
- Contribute code or documentation
- Lint, types, and code quality policy
- CJK speech and subtitle pipeline contracts
- Frozen-tail video export paths
This directory contains durable, public product documentation. To prevent several almost-identical sources from drifting:
- the root README is the product landing page;
- this directory owns task-oriented and conceptual product guidance;
- component READMEs own exact Manager and MCP operational contracts;
- GitHub Releases own downloads, checksums, versions, and release notes; and
- experiments, qualification records, incident notes, and implementation reviews belong in issues, pull requests, release records, or other internal working material—not in the public documentation tree.
Documentation should prefer stable names and the latest-release page over hard-coded version numbers and filenames. When behavior is version-specific, say so in that version's release notes.