Skip to content

Support Chatterbox multilingual frontends #375

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
ivan-digital merged 2 commits into from
Jul 5, 2026
Merged

Support Chatterbox multilingual frontends #375

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
9a41ca6
Support Chatterbox multilingual frontends
Jul 5, 2026
1c31dd2
Add safe Hebrew support for Chatterbox
Jul 5, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 4 additions & 4 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,8 +46,8 @@ On-device speech recognition, synthesis, and understanding for Mac and iOS. Runs
- **[VibeVoice TTS](https://soniqo.audio/guides/vibevoice)** — Long-form / multi-speaker TTS (Microsoft VibeVoice Realtime-0.5B + 1.5B, MLX, up to 90-min podcast/audiobook synthesis, EN/ZH)
- **[Magpie TTS](https://soniqo.audio/guides/magpie)** — Multilingual TTS (NVIDIA Magpie-TTS Multilingual 357M, MLX INT8 411 MB or CoreML INT8 342 MB, 9 languages, 5 baked speakers, streaming on MLX)
- **[Supertonic TTS](https://soniqo.audio/guides/supertonic)** — On-device flow-matching TTS (Supertone Supertonic-3 99M, CoreML/Neural Engine, 31 languages, 10 voices, G2P-free, 44.1 kHz)
- **[Chatterbox TTS](https://huggingface.co/aufklarer/Chatterbox-Multilingual-MLX-fp16)** — Multilingual TTS with zero-shot voice cloning (Resemble AI Chatterbox Multilingual, MLX fp16 ~1.3 GB, 23 languages, MIT)
- **[OmniVoice TTS](https://huggingface.co/aufklarer/OmniVoice-MLX-int8)** — Non-autoregressive diffusion TTS with zero-shot voice cloning (k2-fsa OmniVoice, Qwen3 backbone, MLX int8 ~1 GB / fp16, 600+ languages, Apache-2.0)
- **[Chatterbox TTS](https://huggingface.co/aufklarer/Chatterbox-Multilingual-MLX-fp16)** — Multilingual TTS with zero-shot voice cloning (Resemble AI Chatterbox Multilingual, MLX fp16 ~1.3 GB, 23 runtime languages; Hebrew requires niqqud, MIT)
- **[OmniVoice TTS](https://huggingface.co/aufklarer/OmniVoice-MLX-fp16)** — Non-autoregressive diffusion TTS with zero-shot voice cloning (k2-fsa OmniVoice, Qwen3 backbone, MLX fp16 default / int8 available, 600+ languages, Apache-2.0)
- **[Indic-Mio](docs/models/indic-mio-tts.md)** — Hindi/Indic TTS with inline emotion markers and optional reference-voice cloning (MLX, 24 kHz)

**LLMs & Translation**
Expand Down Expand Up @@ -168,8 +168,8 @@ Compact view below. **[Full model catalogue with sizes, quantisations, download
| [VibeVoice Realtime-0.5B](https://soniqo.audio/guides/vibevoice) | Text → Speech (long-form, multi-speaker) | MLX | 0.5B | EN/ZH |
| [VibeVoice 1.5B](https://soniqo.audio/guides/vibevoice) | Text → Speech (up to 90-min podcast) | MLX | 1.5B | EN/ZH |
| [Magpie-TTS Multilingual](https://soniqo.audio/guides/magpie) | Text → Speech (5 baked speakers, streaming) | MLX / CoreML | 357M (MLX INT8, CoreML INT8) | 9 (CoreML excludes JA) |
| [Chatterbox Multilingual](https://huggingface.co/aufklarer/Chatterbox-Multilingual-MLX-fp16) | Text → Speech (zero-shot cloning) | MLX | 0.8B (fp16) | 23 |
| [OmniVoice](https://huggingface.co/aufklarer/OmniVoice-MLX-int8) | Text → Speech (NAR diffusion, zero-shot cloning) | MLX | 0.8B (int8/fp16) | **600+** |
| [Chatterbox Multilingual](https://huggingface.co/aufklarer/Chatterbox-Multilingual-MLX-fp16) | Text → Speech (zero-shot cloning) | MLX | 0.8B (fp16) | 23 (HE requires niqqud) |
| [OmniVoice](https://huggingface.co/aufklarer/OmniVoice-MLX-fp16) | Text → Speech (NAR diffusion, zero-shot cloning) | MLX | 0.8B (fp16 default / int8) | **600+** |
| [Indic-Mio](docs/models/indic-mio-tts.md) | Text → Speech (Hindi/Indic, emotion tags, voice cloning) | MLX | fp16 | Hindi / Indic |
| [Fish Audio S2 Pro](docs/models/fish-audio-s2-pro.md) | Text → Speech (zero-shot cloning, explicit style markers) | MLX | 0.5B-class (fp16) | Multilingual |
| [Qwen3.5 Chat](docs/models/qwen35-chat.md) | Text → Text (LLM) | MLX, CoreML | 0.8B | Multi |
Expand Down
8 changes: 4 additions & 4 deletions README_ar.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -56,8 +56,8 @@
- **[VibeVoice TTS](https://soniqo.audio/ar/guides/vibevoice)** — تحويل النص إلى كلام للنصوص الطويلة / متعدد المتحدثين (Microsoft VibeVoice Realtime-0.5B + 1.5B، MLX، توليد بودكاست/كتب صوتية حتى 90 دقيقة، EN/ZH)
- **[Magpie TTS](https://soniqo.audio/ar/guides/magpie)** — تحويل النص إلى كلام متعدد اللغات (NVIDIA Magpie-TTS Multilingual 357M، MLX INT8 411 ميغابايت أو CoreML INT8 342 ميغابايت، 9 لغات، 5 متحدثين جاهزين، تدفق على MLX)
- **[Supertonic TTS](https://soniqo.audio/guides/supertonic)** — تحويل النص إلى كلام على الجهاز بمطابقة التدفق (Supertone Supertonic-3 99M، CoreML/Neural Engine، 31 لغة، 10 أصوات، بدون G2P، 44.1 كيلوهرتز)
- **[Chatterbox TTS](https://huggingface.co/aufklarer/Chatterbox-Multilingual-MLX-fp16)** — تحويل النص إلى كلام متعدد اللغات مع استنساخ الصوت بدون تدريب مسبق (Resemble AI Chatterbox Multilingual، MLX fp16 ~1.3 غيغابايت، 23 لغة، MIT)
- **[OmniVoice TTS](https://huggingface.co/aufklarer/OmniVoice-MLX-int8)** — تحويل النص إلى كلام بالانتشار غير الانحداري التلقائي مع استنساخ الصوت بدون تدريب مسبق (k2-fsa OmniVoice، عمود Qwen3 الفقري، MLX int8 ~1 GB / fp16، أكثر من 600 لغة، Apache-2.0)
- **[Chatterbox TTS](https://huggingface.co/aufklarer/Chatterbox-Multilingual-MLX-fp16)** — تحويل النص إلى كلام متعدد اللغات مع استنساخ الصوت بدون تدريب مسبق (Resemble AI Chatterbox Multilingual، MLX fp16 ~1.3 غيغابايت، 23 لغة وقت تشغيل؛ العبرية تتطلب النِّقُود، MIT)
- **[OmniVoice TTS](https://huggingface.co/aufklarer/OmniVoice-MLX-fp16)** — تحويل النص إلى كلام بالانتشار غير الانحداري التلقائي مع استنساخ الصوت بدون تدريب مسبق (k2-fsa OmniVoice، عمود Qwen3 الفقري، MLX fp16 افتراضي / int8 متاح، أكثر من 600 لغة، Apache-2.0)
- **[Indic-Mio](docs/models/indic-mio-tts.md)** — Hindi/Indic TTS with inline emotion markers and optional reference-voice cloning (MLX, 24 kHz)

**نماذج LLM والترجمة**
Expand Down Expand Up @@ -207,8 +207,8 @@ struct DictateView: View {
| [Qwen3-TTS](https://soniqo.audio/ar/guides/speak) | نص → كلام | MLX, CoreML | 0.6B, 1.7B | 10 |
| [CosyVoice3](https://soniqo.audio/ar/guides/cosyvoice) | نص → كلام | MLX | 0.5B | 9 |
| [VoxCPM2](https://soniqo.audio/ar/speech-generation) | نص → كلام (48 كيلوهرتز، تصميم + استنساخ صوت) | MLX | 2B (bf16/int8) | 30 |
| [Chatterbox Multilingual](https://huggingface.co/aufklarer/Chatterbox-Multilingual-MLX-fp16) | نص → كلام (استنساخ صوت بدون تدريب) | MLX | 0.8B (fp16) | 23 |
| [OmniVoice](https://huggingface.co/aufklarer/OmniVoice-MLX-int8) | نص → كلام (انتشار NAR، استنساخ بدون تدريب) | MLX | 0.8B (int8/fp16) | 600+ |
| [Chatterbox Multilingual](https://huggingface.co/aufklarer/Chatterbox-Multilingual-MLX-fp16) | نص → كلام (استنساخ صوت بدون تدريب) | MLX | 0.8B (fp16) | 23 (HE تتطلب النِّقُود) |
| [OmniVoice](https://huggingface.co/aufklarer/OmniVoice-MLX-fp16) | نص → كلام (انتشار NAR، استنساخ بدون تدريب) | MLX | 0.8B (fp16 افتراضي / int8) | 600+ |
| [Indic-Mio](docs/models/indic-mio-tts.md) | Text → Speech (Hindi/Indic, emotion tags, voice cloning) | MLX | fp16 | Hindi / Indic |
| [Fish Audio S2 Pro](docs/models/fish-audio-s2-pro.md) | نص → كلام (استنساخ صوت بدون تدريب، علامات نمط صريحة) | MLX | 0.5B-class (fp16) | متعدد اللغات |
| [Kokoro-82M](https://soniqo.audio/ar/guides/kokoro) | نص → كلام | CoreML (ANE) | 82M | 10 |
Expand Down
8 changes: 4 additions & 4 deletions README_de.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,8 +46,8 @@ Spracherkennung, -synthese und -verständnis auf dem Gerät für Mac und iOS. L
- **[VibeVoice TTS](https://soniqo.audio/de/guides/vibevoice)** — Langform-/Multi-Speaker-TTS (Microsoft VibeVoice Realtime-0.5B + 1.5B, MLX, bis zu 90 Min. Podcast-/Hörbuch-Synthese, EN/ZH)
- **[Magpie TTS](https://soniqo.audio/de/guides/magpie)** — Mehrsprachiges TTS (NVIDIA Magpie-TTS Multilingual 357M, MLX INT8 411 MB oder CoreML INT8 342 MB, 9 Sprachen, 5 vordefinierte Sprecher, Streaming auf MLX)
- **[Supertonic TTS](https://soniqo.audio/guides/supertonic)** — Flow-Matching-TTS auf dem Gerät (Supertone Supertonic-3 99M, CoreML/Neural Engine, 31 Sprachen, 10 Stimmen, G2P-frei, 44,1 kHz)
- **[Chatterbox TTS](https://huggingface.co/aufklarer/Chatterbox-Multilingual-MLX-fp16)** — Mehrsprachiges TTS mit Zero-Shot-Stimmklonen (Resemble AI Chatterbox Multilingual, MLX fp16 ~1,3 GB, 23 Sprachen, MIT)
- **[OmniVoice TTS](https://huggingface.co/aufklarer/OmniVoice-MLX-int8)** — Nicht-autoregressives Diffusions-TTS mit Zero-Shot-Stimmklonen (k2-fsa OmniVoice, Qwen3-Backbone, MLX int8 ~1 GB / fp16, 600+ Sprachen, Apache-2.0)
- **[Chatterbox TTS](https://huggingface.co/aufklarer/Chatterbox-Multilingual-MLX-fp16)** — Mehrsprachiges TTS mit Zero-Shot-Stimmklonen (Resemble AI Chatterbox Multilingual, MLX fp16 ~1,3 GB, 23 Laufzeitsprachen; Hebräisch erfordert Nikkud, MIT)
- **[OmniVoice TTS](https://huggingface.co/aufklarer/OmniVoice-MLX-fp16)** — Nicht-autoregressives Diffusions-TTS mit Zero-Shot-Stimmklonen (k2-fsa OmniVoice, Qwen3-Backbone, MLX fp16 Standard / int8 verfügbar, 600+ Sprachen, Apache-2.0)
- **[Indic-Mio](docs/models/indic-mio-tts.md)** — Hindi/Indic TTS with inline emotion markers and optional reference-voice cloning (MLX, 24 kHz)

**LLMs und Übersetzung**
Expand Down Expand Up @@ -163,8 +163,8 @@ Kompakte Übersicht unten. **[Vollständiger Modellkatalog mit Größen, Quantis
| [Qwen3-TTS](https://soniqo.audio/de/guides/speak) | Text → Sprache | MLX, CoreML | 0.6B, 1.7B | 10 |
| [CosyVoice3](https://soniqo.audio/de/guides/cosyvoice) | Text → Sprache | MLX | 0.5B | 9 |
| [VoxCPM2](https://soniqo.audio/de/speech-generation) | Text → Sprache (48 kHz, Voice Design + Klonen) | MLX | 2B (bf16/int8) | 30 |
| [Chatterbox Multilingual](https://huggingface.co/aufklarer/Chatterbox-Multilingual-MLX-fp16) | Text → Sprache (Zero-Shot-Klonen) | MLX | 0.8B (fp16) | 23 |
| [OmniVoice](https://huggingface.co/aufklarer/OmniVoice-MLX-int8) | Text → Sprache (NAR-Diffusion, Zero-Shot-Klonen) | MLX | 0.8B (int8/fp16) | 600+ |
| [Chatterbox Multilingual](https://huggingface.co/aufklarer/Chatterbox-Multilingual-MLX-fp16) | Text → Sprache (Zero-Shot-Klonen) | MLX | 0.8B (fp16) | 23 (HE erfordert Nikkud) |
| [OmniVoice](https://huggingface.co/aufklarer/OmniVoice-MLX-fp16) | Text → Sprache (NAR-Diffusion, Zero-Shot-Klonen) | MLX | 0.8B (fp16 Standard / int8) | 600+ |
| [Indic-Mio](docs/models/indic-mio-tts.md) | Text → Speech (Hindi/Indic, emotion tags, voice cloning) | MLX | fp16 | Hindi / Indic |
| [Fish Audio S2 Pro](docs/models/fish-audio-s2-pro.md) | Text → Sprache (Zero-Shot-Klonen, explizite Stilmarker) | MLX | 0.5B-class (fp16) | Mehrsprachig |
| [Kokoro-82M](https://soniqo.audio/de/guides/kokoro) | Text → Sprache | CoreML (ANE) | 82M | 10 |
Expand Down
8 changes: 4 additions & 4 deletions README_es.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,8 +46,8 @@ Reconocimiento, síntesis y comprensión de voz en el dispositivo para Mac e iOS
- **[VibeVoice TTS](https://soniqo.audio/es/guides/vibevoice)** — TTS de formato largo / múltiples hablantes (Microsoft VibeVoice Realtime-0.5B + 1.5B, MLX, síntesis de podcast/audiolibro de hasta 90 min, EN/ZH)
- **[Magpie TTS](https://soniqo.audio/es/guides/magpie)** — TTS multilingüe (NVIDIA Magpie-TTS Multilingual 357M, MLX INT8 411 MB o CoreML INT8 342 MB, 9 idiomas, 5 hablantes preconfigurados, streaming en MLX)
- **[Supertonic TTS](https://soniqo.audio/guides/supertonic)** — TTS de flow-matching en el dispositivo (Supertone Supertonic-3 99M, CoreML/Neural Engine, 31 idiomas, 10 voces, sin G2P, 44.1 kHz)
- **[Chatterbox TTS](https://huggingface.co/aufklarer/Chatterbox-Multilingual-MLX-fp16)** — TTS multilingüe con clonación de voz zero-shot (Resemble AI Chatterbox Multilingual, MLX fp16 ~1,3 GB, 23 idiomas, MIT)
- **[OmniVoice TTS](https://huggingface.co/aufklarer/OmniVoice-MLX-int8)** — TTS de difusión no autorregresivo con clonación de voz zero-shot (k2-fsa OmniVoice, backbone Qwen3, MLX int8 ~1 GB / fp16, 600+ idiomas, Apache-2.0)
- **[Chatterbox TTS](https://huggingface.co/aufklarer/Chatterbox-Multilingual-MLX-fp16)** — TTS multilingüe con clonación de voz zero-shot (Resemble AI Chatterbox Multilingual, MLX fp16 ~1,3 GB, 23 idiomas en runtime; hebreo requiere niqqud, MIT)
- **[OmniVoice TTS](https://huggingface.co/aufklarer/OmniVoice-MLX-fp16)** — TTS de difusión no autorregresivo con clonación de voz zero-shot (k2-fsa OmniVoice, backbone Qwen3, MLX fp16 por defecto / int8 disponible, 600+ idiomas, Apache-2.0)
- **[Indic-Mio](docs/models/indic-mio-tts.md)** — Hindi/Indic TTS with inline emotion markers and optional reference-voice cloning (MLX, 24 kHz)

**LLM y traducción**
Expand Down Expand Up @@ -163,8 +163,8 @@ Vista compacta a continuación. **[Catálogo completo de modelos con tamaños, c
| [Qwen3-TTS](https://soniqo.audio/es/guides/speak) | Texto → Voz | MLX, CoreML | 0.6B, 1.7B | 10 |
| [CosyVoice3](https://soniqo.audio/es/guides/cosyvoice) | Texto → Voz | MLX | 0.5B | 9 |
| [VoxCPM2](https://soniqo.audio/es/speech-generation) | Texto → Voz (48 kHz, diseño de voz + clonación) | MLX | 2B (bf16/int8) | 30 |
| [Chatterbox Multilingual](https://huggingface.co/aufklarer/Chatterbox-Multilingual-MLX-fp16) | Texto → Voz (clonación zero-shot) | MLX | 0.8B (fp16) | 23 |
| [OmniVoice](https://huggingface.co/aufklarer/OmniVoice-MLX-int8) | Texto → Voz (difusión NAR, clonación zero-shot) | MLX | 0.8B (int8/fp16) | 600+ |
| [Chatterbox Multilingual](https://huggingface.co/aufklarer/Chatterbox-Multilingual-MLX-fp16) | Texto → Voz (clonación zero-shot) | MLX | 0.8B (fp16) | 23 (HE requiere niqqud) |
| [OmniVoice](https://huggingface.co/aufklarer/OmniVoice-MLX-fp16) | Texto → Voz (difusión NAR, clonación zero-shot) | MLX | 0.8B (fp16 por defecto / int8) | 600+ |
| [Indic-Mio](docs/models/indic-mio-tts.md) | Text → Speech (Hindi/Indic, emotion tags, voice cloning) | MLX | fp16 | Hindi / Indic |
| [Fish Audio S2 Pro](docs/models/fish-audio-s2-pro.md) | Texto → Voz (clonación zero-shot, marcadores de estilo explícitos) | MLX | 0.5B-class (fp16) | Multilingüe |
| [Kokoro-82M](https://soniqo.audio/es/guides/kokoro) | Texto → Voz | CoreML (ANE) | 82M | 10 |
Expand Down
8 changes: 4 additions & 4 deletions README_fr.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,8 +46,8 @@ Reconnaissance, synthese et comprehension vocale embarquees pour Mac et iOS. S'e
- **[VibeVoice TTS](https://soniqo.audio/fr/guides/vibevoice)** -- TTS long format / multi-locuteurs (Microsoft VibeVoice Realtime-0.5B + 1.5B, MLX, synthese de podcast/livre audio jusqu'a 90 min, EN/ZH)
- **[Magpie TTS](https://soniqo.audio/fr/guides/magpie)** — TTS multilingue (NVIDIA Magpie-TTS Multilingual 357M, MLX INT8 411 Mo ou CoreML INT8 342 Mo, 9 langues, 5 voix prédéfinies, streaming sur MLX)
- **[Supertonic TTS](https://soniqo.audio/guides/supertonic)** — TTS embarque par appariement de flux (Supertone Supertonic-3 99M, CoreML/Neural Engine, 31 langues, 10 voix, sans G2P, 44.1 kHz)
- **[Chatterbox TTS](https://huggingface.co/aufklarer/Chatterbox-Multilingual-MLX-fp16)** — TTS multilingue avec clonage vocal zero-shot (Resemble AI Chatterbox Multilingual, MLX fp16 ~1,3 Go, 23 langues, MIT)
- **[OmniVoice TTS](https://huggingface.co/aufklarer/OmniVoice-MLX-int8)** — TTS à diffusion non autorégressive avec clonage vocal zero-shot (k2-fsa OmniVoice, backbone Qwen3, MLX int8 ~1 Go / fp16, 600+ langues, Apache-2.0)
- **[Chatterbox TTS](https://huggingface.co/aufklarer/Chatterbox-Multilingual-MLX-fp16)** — TTS multilingue avec clonage vocal zero-shot (Resemble AI Chatterbox Multilingual, MLX fp16 ~1,3 Go, 23 langues runtime; l'hébreu exige le niqqud, MIT)
- **[OmniVoice TTS](https://huggingface.co/aufklarer/OmniVoice-MLX-fp16)** — TTS à diffusion non autorégressive avec clonage vocal zero-shot (k2-fsa OmniVoice, backbone Qwen3, MLX fp16 par défaut / int8 disponible, 600+ langues, Apache-2.0)
- **[Indic-Mio](docs/models/indic-mio-tts.md)** — Hindi/Indic TTS with inline emotion markers and optional reference-voice cloning (MLX, 24 kHz)

**LLM et traduction**
Expand Down Expand Up @@ -163,8 +163,8 @@ Vue compacte ci-dessous. **[Catalogue complet des modeles avec tailles, quantifi
| [Qwen3-TTS](https://soniqo.audio/fr/guides/speak) | Texte → Parole | MLX, CoreML | 0.6B, 1.7B | 10 |
| [CosyVoice3](https://soniqo.audio/fr/guides/cosyvoice) | Texte → Parole | MLX | 0.5B | 9 |
| [VoxCPM2](https://soniqo.audio/fr/speech-generation) | Texte → Parole (48 kHz, conception vocale + clonage) | MLX | 2B (bf16/int8) | 30 |
| [Chatterbox Multilingual](https://huggingface.co/aufklarer/Chatterbox-Multilingual-MLX-fp16) | Texte → Voix (clonage zero-shot) | MLX | 0.8B (fp16) | 23 |
| [OmniVoice](https://huggingface.co/aufklarer/OmniVoice-MLX-int8) | Texte → Voix (diffusion NAR, clonage zero-shot) | MLX | 0.8B (int8/fp16) | 600+ |
| [Chatterbox Multilingual](https://huggingface.co/aufklarer/Chatterbox-Multilingual-MLX-fp16) | Texte → Voix (clonage zero-shot) | MLX | 0.8B (fp16) | 23 (HE exige le niqqud) |
| [OmniVoice](https://huggingface.co/aufklarer/OmniVoice-MLX-fp16) | Texte → Voix (diffusion NAR, clonage zero-shot) | MLX | 0.8B (fp16 par défaut / int8) | 600+ |
| [Indic-Mio](docs/models/indic-mio-tts.md) | Text → Speech (Hindi/Indic, emotion tags, voice cloning) | MLX | fp16 | Hindi / Indic |
| [Fish Audio S2 Pro](docs/models/fish-audio-s2-pro.md) | Texte → Voix (clonage zero-shot, marqueurs de style explicites) | MLX | 0.5B-class (fp16) | Multilingue |
| [Kokoro-82M](https://soniqo.audio/fr/guides/kokoro) | Texte → Parole | CoreML (ANE) | 82M | 10 |
Expand Down
Loading