Capability matrix

Match language and voice requirements to an available processing path.

The catalogue below describes declared support, typical hardware paths, and model licence constraints. It is a selection aid, not a benchmark.

Find a compatible local starting point

This filters declared language and hardware paths. It is not a performance benchmark or a guarantee that every voice has equal quality.

Voice requirement Language Hardware path

Compatible catalogue entries

Kokoro 82M

A lightweight starting point with built-in voices.

CPU and CUDA are established paths; ROCm and Apple Silicon support are experimental upstream paths.

Apache-2.0

Qwen3 TTS

Built-in speakers with CustomVoice, or reference cloning with Base models.

CPU, CUDA, Vulkan, and Metal variants are available; model size and quantization remain selectable.

Apache-2.0

Silero

Efficient built-in voices, including East European and regional packs.

Runs on CPU on Windows and Linux.

MIT or CC BY-NC-SA 4.0, depending on the voice pack

Magpie 357M

Five built-in speakers shared across nine supported languages.

CPU and NVIDIA CUDA variants are available; the first NeMo installation is comparatively large.

NVIDIA Open Model License

The installer remains authoritative for downloadable variants, model size, host compatibility, and the licence shown before installation.

Local speech services

Model size, quantization, input length, and concurrent services all affect practical requirements.

Service	Voice type	Typical hardware path	Declared languages	Model licence
Kokoro 82M A lightweight starting point with built-in voices.	Ready-made	CPU only · NVIDIA GPU · AMD / Intel / Vulkan-capable GPU · Apple Silicon	Chinese, English, French, Hindi, Italian, Japanese, Portuguese, Spanish	Apache-2.0 Commercial use is permitted under the model terms.
Qwen3 TTS Built-in speakers with CustomVoice, or reference cloning with Base models.	Ready-made · Cloning	CPU only · NVIDIA GPU · AMD / Intel / Vulkan-capable GPU · Apple Silicon	Chinese, English, French, German, Italian, Japanese, Korean, Portuguese, Russian +1 languages Spanish	Apache-2.0 Commercial use is permitted under the model terms.
XTTS v2 A mature multilingual cloning model using a short reference recording.	Cloning	CPU only · NVIDIA GPU	Arabic, Chinese, Czech, Dutch, English, French, German, Hungarian, Italian +7 languages Japanese, Korean, Polish, Portuguese, Russian, Spanish, Turkish	Coqui Public Model License 1.0.0 The model and its outputs are restricted to non-commercial use.
VoxCPM2 A comparatively large multilingual model conditioned by a reference voice.	Cloning	NVIDIA GPU	Arabic, Burmese, Chinese, Danish, Dutch, English, Finnish, French, German +21 languages Greek, Hebrew, Hindi, Indonesian, Italian, Japanese, Khmer, Korean, Lao, Malay, Norwegian, Polish, Portuguese, Russian, Spanish, Swahili, Swedish, Tagalog, Thai, Turkish, Vietnamese	Apache-2.0 Commercial use is permitted under the model terms.
Fish S2 Pro Broad declared language coverage with selectable native backends and quantization.	Cloning	CPU only · NVIDIA GPU · AMD / Intel / Vulkan-capable GPU · Apple Silicon	Afrikaans, Albanian, Amharic, Arabic, Armenian, Assamese, Azerbaijani, Basque, Belarusian +74 languages Bengali, Bosnian, Breton, Bulgarian, Burmese, Catalan, Chinese, Croatian, Czech, Danish, Dutch, English, Estonian, Faroese, Finnish, French, Galician, Georgian, German, Greek, Gujarati, Haitian Creole, Hebrew, Hindi, Hungarian, Icelandic, Indonesian, Italian, Japanese, Javanese, Kannada, Kazakh, Khmer, Korean, Latin, Latvian, Lithuanian, Malay, Malayalam, Maori, Marathi, Mongolian, Nepali, Norwegian, Norwegian Nynorsk, Pashto, Persian, Polish, Portuguese, Punjabi, Romanian, Russian, Sanskrit, Serbian, Shona, Sindhi, Sinhala, Slovak, South Slavey, Spanish, Swahili, Swedish, Tagalog, Tamil, Telugu, Thai, Tibetan, Turkish, Ukrainian, Urdu, Vietnamese, Welsh, Yiddish, Yoruba	Fish Audio Research License Research and non-commercial use; commercial use requires a separate licence.
Voxtral 4B Preset voices through a WGPU-compatible accelerator; no packaged CPU path.	Ready-made	NVIDIA GPU · AMD / Intel / Vulkan-capable GPU	Arabic, Dutch, English, French, German, Hindi, Italian, Portuguese, Spanish	CC BY-NC 4.0 Non-commercial use only under the model terms.
Silero Efficient built-in voices, including East European and regional packs.	Ready-made	CPU only	Armenian, Azerbaijani, Bashkir, Belarusian, Bengali, Chuvash, English, Erzya, French +24 languages Georgian, German, Gujarati, Hindi, Kannada, Kabardian-Cherkess, Kalmyk, Kazakh, Khakas, Kyrgyz, Malayalam, Manipuri, Moksha, Rajasthani, Russian, Spanish, Tajik, Tamil, Tatar, Telugu, Udmurt, Ukrainian, Uzbek, Yakut	MIT or CC BY-NC-SA 4.0, depending on the voice pack The installer shows the applicable pack licence before installation.
Chatterbox Expressive cloning with English and multilingual model choices.	Cloning	CPU only · NVIDIA GPU	Arabic, Chinese, Danish, Dutch, English, Finnish, French, German, Greek +14 languages Hebrew, Hindi, Italian, Japanese, Korean, Malay, Norwegian, Polish, Portuguese, Russian, Spanish, Swahili, Swedish, Turkish	MIT Commercial use is permitted under the model terms.
Magpie 357M Five built-in speakers shared across nine supported languages.	Ready-made	CPU only · NVIDIA GPU	Chinese, English, French, German, Hindi, Italian, Japanese, Spanish, Vietnamese	NVIDIA Open Model License The model card marks the checkpoint ready for commercial use under NVIDIA terms.

Speech language overview

You need only one compatible service. Custom and commercial endpoints may add languages not listed here.

Declared languages	Ready-made	Cloning
English, French, Spanish	Kokoro, Qwen3 CustomVoice, Voxtral, Silero, Magpie	Qwen3 Base, XTTS, VoxCPM2, Fish S2 Pro, Chatterbox
German	Qwen3 CustomVoice, Voxtral, Silero, Magpie	Qwen3 Base, XTTS, VoxCPM2, Fish S2 Pro, Chatterbox
Italian	Kokoro, Qwen3 CustomVoice, Voxtral, Magpie	Qwen3 Base, XTTS, VoxCPM2, Fish S2 Pro, Chatterbox
Portuguese	Kokoro, Qwen3 CustomVoice, Voxtral	Qwen3 Base, XTTS, VoxCPM2, Fish S2 Pro, Chatterbox
Russian	Qwen3 CustomVoice, Silero	Qwen3 Base, XTTS, VoxCPM2, Fish S2 Pro, Chatterbox
Chinese, Japanese	Kokoro, Qwen3 CustomVoice, Magpie	Qwen3 Base, XTTS, VoxCPM2, Fish S2 Pro, Chatterbox
Korean	Qwen3 CustomVoice	Qwen3 Base, XTTS, VoxCPM2, Fish S2 Pro, Chatterbox
Hindi	Kokoro, Voxtral, Silero, Magpie	VoxCPM2, Fish S2 Pro, Chatterbox
Arabic, Dutch	Voxtral	XTTS, VoxCPM2, Fish S2 Pro, Chatterbox
Polish, Turkish	—	XTTS, VoxCPM2, Fish S2 Pro, Chatterbox
Czech, Hungarian	—	XTTS, Fish S2 Pro
Vietnamese	Magpie	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
Additional regional and low-resource languages	Selected Silero packs	Fish S2 Pro; consult the complete catalogue

Transcription

CrispASR exposes both packaged engines with word timestamps, configurable VAD, quantization, and compute backend.

Whisper large-v3

FP16 or Q5_0

100 languages; the broadest packaged option.

Parakeet TDT 0.6B v3

FP16, Q8_0, Q5_0, or Q4_K

25 primarily European languages.

Input and output formats

Documents: TXT, PDF, EPUB, DOCX, MOBI, pasted text
Subtitles: SRT
Audio sources: AAC, AIFF, FLAC, M4A/MKA, MP3, OGG, Opus, WAV, WMA
Video sources: MP4, MKV, WebM, AVI, MOV
Audio output: M4B, MP3, Opus, FLAC, WAV
Video output: MP4 with selectable audio and subtitle tracks

Catalogue snapshot reviewed 25 July 2026. Update this site-owned snapshot when the Pandrator installer catalogue changes. 2026-07-25