Pandrator

Pandrator is a local-first workspace for creating audiobooks, subtitles, and voiceovers. It combines document preparation, transcription, optional LLM correction and translation, speech generation, detailed review, and export in one browser-based interface.

The desktop installer and launcher manage Pandrator and its optional local model services. Docker and WSL are not required.

The application itself is web-only. The retired Qt application is preserved on the qt-maintenance branch; Qt remains in main only for the desktop installer and launcher.

TL;DR

If you want to…	Start with…
Create an audiobook with ready-made voices	Kokoro is the simplest lightweight starting point. Consult the language table for alternatives.
Clone a voice from a reference recording	Install Qwen3 TTS Base or another cloning model supporting your language.
Create subtitles from audio or video	Install CrispASR. Use Whisper large-v3 for broad coverage or Parakeet TDT 0.6B v3 for its supported languages.
Correct or translate text and subtitles	Configure a local or cloud LLM provider. This is optional for basic speech generation.
Convert generated speech to another trained voice	Install RVC. It runs after speech generation and is optional.

Download the launcher from GitHub Releases. You can begin with only the components you need and add others later.

Local models process content on your machine. Cloud LLM and speech providers are optional; when used, they may send content to an external service and incur charges.

What Pandrator does

Audiobooks

• Imports plain text, pasted text, PDF, EPUB, DOCX, and MOBI sources.
• Extracts structure and chapter markers, with OCR and a reviewable cleaning workflow for difficult PDF and EPUB files.
• Includes a browser PDF editor with page stacks, left/right stacks, cropping, whiteouts, and deletion.
• Applies deterministic text normalization and configurable segmentation before speech generation.
• Optionally uses an LLM to clean a complete document or optimize small batches while generating.
• Keeps generated speech as reviewable segments: edit, play as a playlist, mark, regenerate, compare takes, and select RVC variants.
• Exports WAV, MP3, Opus, FLAC, or M4B with chapters, metadata, and cover art.

Subtitles and voiceovers

• Starts from SRT subtitles or common audio and video formats.
• Transcribes media through CrispASR with word timestamps, VAD, and optional diarization controls.
• Keeps transcription, correction, translation, subtitle composition, speech generation, synchronization, and export as separate, rerunnable steps.
• Supports professional translation directly from an original transcript or from a corrected revision.
• Provides side-by-side subtitle review with timing, text, split, and merge editing.
• Creates subtitle-only exports or dubbed media with original, mixed, or dubbing-only audio and soft, burned, translated, original-language, or bilingual subtitles.

Providers and voices

• Connects to local TTS services, OpenAI, Google Gemini, and configurable custom speech endpoints.
• Connects to local OpenAI-compatible LLM servers such as LM Studio as well as supported cloud providers.
• Stores model-specific LLM temperature, reasoning, and cached/uncached token pricing defaults.
• Manages recorded and uploaded reference samples, transcripts, and persistent previews of pre-built voices in the voice library.
• Supports RVC model management and XTTS training as separate workflows.

Installation

Windows

Download PandratorInstaller.exe from Releases and run it. Choose an installation location and the local services you need. The same application is used later to launch, update, repair, or extend the installation.

Linux

Download PandratorInstaller-x86_64.AppImage, make it executable, and run it:

chmod +x PandratorInstaller-x86_64.AppImage
./PandratorInstaller-x86_64.AppImage

The default installation location is ~/Pandrator. The installer keeps Pixi environments, services, model caches, and application data under the selected workspace. It does not install system packages. Calibre is optional and is needed only for MOBI conversion.

Launching and access

The launcher starts the web application, worker, and selected speech services, waits for Pandrator to become ready, and opens it in your browser. Its Open Web UI action reopens the page if you close the tab.

Pandrator listens on 127.0.0.1 by default. The installer can enable access from other devices; non-local access requires owner authentication. Use HTTPS through a reverse proxy when exposing Pandrator beyond a trusted local network.

Closing the browser does not stop generation. Use the launcher to reopen the interface or stop the managed processes.

Updating

Use Update in the installer. If Pandrator or one of its services is running, the installer can stop it before continuing. User data is preserved unless removal with data purging is explicitly requested.

Before the first web-based launch, Pandrator backs up legacy metadata and imports existing sessions into its new database without rewriting the original Qt data. Keep a separate backup of important work before any major application upgrade.

Choosing local services

Service	Voice type	Typical hardware path	Notes
Kokoro 82M	Pre-built	CPU, CUDA; experimental ROCm and Apple Silicon paths	Lightweight and a good first installation.
Qwen3 TTS	Pre-built and cloning	CPU, CUDA, Vulkan, Metal	CustomVoice 1.7B supplies named voices; Base 0.6B/1.7B clones references.
XTTS v2	Cloning	CPU or CUDA	Mature multilingual cloning; GPU is much faster.
VoxCPM2	Cloning	CUDA	Large multilingual model intended for capable NVIDIA hardware.
Fish S2 Pro	Cloning	Configurable native backend and quantization	Very broad declared language coverage.
Voxtral 4B	Pre-built	WGPU-compatible accelerator	Preset voices only in the packaged local service; no CPU path.
Silero	Pre-built	CPU	Efficient regional, East European, and legacy language packs.
Chatterbox	Cloning	CPU or CUDA	English and multilingual models.
Magpie 357M	Pre-built	CPU or CUDA	Five speakers shared across nine languages.
CrispASR	Transcription	CPU, CUDA, Vulkan; Metal on Apple Silicon	Whisper and Parakeet engines with word timestamps.
RVC	Speech-to-speech	CPU or CUDA	Applies a .pth model and matching .index after generation.

Hardware requirements vary with model size, quantization, input length, and the selected compute backend. The installer shows the available variants and model licences before downloading them.

Speech generation language support

“Pre-built voices” require no reference recording. “Voice cloning” uses a sample from the voice library. You need only one compatible model from the applicable column.

Language(s)	Pre-built voices	Voice cloning
English, French, Spanish	Kokoro 82M, Qwen3 CustomVoice 1.7B, Voxtral 4B, Silero, Magpie 357M	Qwen3 Base 0.6B/1.7B, XTTS v2, VoxCPM2, Fish S2 Pro, Chatterbox
German	Qwen3 CustomVoice 1.7B, Voxtral 4B, Silero, Magpie 357M	Qwen3 Base 0.6B/1.7B, XTTS v2, VoxCPM2, Fish S2 Pro, Chatterbox
Italian	Kokoro 82M, Qwen3 CustomVoice 1.7B, Voxtral 4B, Magpie 357M	Qwen3 Base 0.6B/1.7B, XTTS v2, VoxCPM2, Fish S2 Pro, Chatterbox
Portuguese	Kokoro 82M, Qwen3 CustomVoice 1.7B, Voxtral 4B	Qwen3 Base 0.6B/1.7B, XTTS v2, VoxCPM2, Fish S2 Pro, Chatterbox
Russian	Qwen3 CustomVoice 1.7B, Silero	Qwen3 Base 0.6B/1.7B, XTTS v2, VoxCPM2, Fish S2 Pro, Chatterbox
Chinese, Japanese	Kokoro 82M, Qwen3 CustomVoice 1.7B, Magpie 357M	Qwen3 Base 0.6B/1.7B, XTTS v2, VoxCPM2, Fish S2 Pro, Chatterbox
Korean	Qwen3 CustomVoice 1.7B	Qwen3 Base 0.6B/1.7B, XTTS v2, VoxCPM2, Fish S2 Pro, Chatterbox
Hindi	Kokoro 82M, Voxtral 4B, Silero, Magpie 357M	VoxCPM2, Fish S2 Pro, Chatterbox
Arabic, Dutch	Voxtral 4B	XTTS v2, VoxCPM2, Fish S2 Pro, Chatterbox
Polish, Turkish	—	XTTS v2, VoxCPM2, Fish S2 Pro, Chatterbox
Czech, Hungarian	—	XTTS v2, Fish S2 Pro
Vietnamese	Magpie 357M	VoxCPM2, Fish S2 Pro
Danish, Finnish, Greek, Hebrew, Malay, Norwegian, Swahili, Swedish	—	VoxCPM2, Fish S2 Pro, Chatterbox
Burmese, Indonesian, Khmer, Tagalog, Thai	—	VoxCPM2, Fish S2 Pro
Lao	—	VoxCPM2
Armenian, Azerbaijani, Belarusian, Bengali, Georgian, Gujarati, Kannada, Kazakh, Malayalam, Tamil, Telugu, Ukrainian	Silero	Fish S2 Pro
Bashkir, Chuvash, Erzya, Kabardian-Cherkess, Kalmyk, Khakas, Kyrgyz, Manipuri, Moksha, Rajasthani, Tajik, Tatar, Udmurt, Uzbek, Yakut	Silero	—
Afrikaans, Albanian, Amharic, Assamese, Basque, Bosnian, Breton, Bulgarian, Catalan, Croatian, Estonian, Faroese, Galician, Haitian Creole, Icelandic, Javanese, Latin, Latvian, Lithuanian, Maori, Marathi, Mongolian, Nepali, Norwegian Nynorsk, Pashto, Persian, Punjabi, Romanian, Sanskrit, Serbian, Shona, Sindhi, Sinhala, Slovak, South Slavey, Tibetan, Urdu, Welsh, Yiddish, Yoruba

Additional notes:

• Qwen3 CustomVoice provides its named pre-built voices. Qwen3 Base performs voice cloning.
• VoxCPM2 also supports several Chinese dialects.
• Silero’s legacy Indic model supports Hindi, Malayalam, Manipuri, Bengali, Rajasthani, Tamil, Telugu, Gujarati, and Kannada. It expects ISO-romanized input.
• A declared language indicates backend support, not equal quality across every model or voice.
• Model licences differ. The installer displays the applicable licence and usage conditions before installation.
• Custom and commercial endpoints may provide languages not listed here.

Transcription language support

Both transcription engines run through CrispASR and produce word-level timestamps. Model choice, quantization, compute backend, language, and VAD settings are configurable.

Model	Available variants	Coverage
Whisper large-v3	FP16 or Q5_0	100 languages; the broadest option
Parakeet TDT 0.6B v3	FP16, Q8_0, Q5_0, or Q4_K	25 primarily European languages

English, Chinese, German, Spanish, Russian, Korean, French, Japanese, Portuguese, Turkish, Polish, Catalan, Dutch, Arabic, Swedish, Italian, Indonesian, Hindi, Finnish, Vietnamese, Hebrew, Ukrainian, Greek, Malay, Czech, Romanian, Danish, Hungarian, Tamil, Norwegian, Thai, Urdu, Croatian, Bulgarian, Lithuanian, Latin, Maori, Malayalam, Welsh, Slovak, Telugu, Persian, Latvian, Bengali, Serbian, Azerbaijani, Slovenian, Kannada, Estonian, Macedonian, Breton, Basque, Icelandic, Armenian, Nepali, Mongolian, Bosnian, Kazakh, Albanian, Swahili, Galician, Marathi, Punjabi, Sinhala, Khmer, Shona, Yoruba, Somali, Afrikaans, Occitan, Georgian, Belarusian, Tajik, Sindhi, Gujarati, Amharic, Yiddish, Lao, Uzbek, Faroese, Haitian Creole, Pashto, Turkmen, Nynorsk, Maltese, Sanskrit, Luxembourgish, Myanmar, Tibetan, Tagalog, Malagasy, Assamese, Tatar, Hawaiian, Lingala, Hausa, Bashkir, Javanese, Sundanese, and Cantonese.

Parakeet TDT 0.6B v3 supports Bulgarian, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, German, Greek, Hungarian, Italian, Latvian, Lithuanian, Maltese, Polish, Portuguese, Romanian, Russian, Slovak, Slovenian, Spanish, Swedish, and Ukrainian.

Input and output formats

Category	Formats
Documents	TXT, PDF, EPUB, DOCX, MOBI, or pasted text
Subtitles	SRT
Audio sources	AAC, AIFF, FLAC, M4A/MKA, MP3,