Onyx Kokoro 82M
A mixed-precision MLX quantization of Kokoro-82M (StyleTTS2-derived, Apache-2.0), packaged for Onyx — an on-device iOS AI assistant. Used by OnyxTTS, Onyx's own from-scratch Swift + MLX text-to-speech engine, to...
Modellquelle
- Art
- Other
- Dateigröße
- 510 KB–148 MB
- Basismodell
- hexgrad/Kokoro-82M
- Version
- vRevision 17a1eb7
- Aktualisiert
- 13. Okt. 2026
- Downloads
- 106
- Quelle
- Hugging Face
Quellenbeschreibung
Onyx-Kokoro-82M
A mixed-precision MLX quantization of Kokoro-82M (StyleTTS2-derived, Apache-2.0), packaged for Onyx — an on-device iOS AI assistant. Used by OnyxTTS, Onyx's own from-scratch Swift + MLX text-to-speech engine, to speak assistant replies in a natural neural voice fully on-device (Apple's AVSpeechSynthesizer is the fallback).
What this is
- Mixed 4-bit quant (group size 64): the PL-BERT (ALBERT) text encoder's Linear weights are 4-bit; the
entire vocoder path — the prosody predictor (LSTMs + AdaIN convs → pitch/energy) and the iSTFTNet generator —
stays fp16. The vocoder directly shapes the waveform and is quant-sensitive, so it is deliberately NOT
quantized. ~330 MB fp32 → 148 MB (
model.safetensors). - Voice packs (
voices/*.safetensors): 54 voices (af_/am_ American, bf_/bm_ British, plus jf/jm/zf/zm/ ef/em/ff/hf/hm/if/im/pf/pm for other languages). Each is a[N, 1, 256]style tensor indexed by phoneme-sequence length. config.jsoncarries the standard Kokoro config plus aquantizationblock ({group_size: 64, bits: 4}) so the loader dequantizes the right layers.
Provenance
Weights derived from prince-canuma/Kokoro-82M (the mlx-audio mirror of hexgrad/Kokoro-82M) via Onyx's
quantize_kokoro.py. Verified to load + synthesize valid 24 kHz audio end-to-end through OnyxTTS on Apple
Silicon. Apache-2.0, same as the base model.
Quellen
1 Quelle
Quellen
1 QuelleVerifiziert 15. Aug.
Modellartefakte
55 Artefakte
Modellartefakte
55 Artefaktevoices/af_alloy.safetensors
safetensors · 510 KB · SHA-256 7997b001760b…82bc · Hugging Face
Herunterladenvoices/af_aoede.safetensors
safetensors · 510 KB · SHA-256 fce9bf78661a…96d2 · Hugging Face
Herunterladenvoices/af_bella.safetensors
safetensors · 510 KB · SHA-256 a18024b9332f…f6ce · Hugging Face
Herunterladenvoices/af_heart.safetensors
safetensors · 510 KB · SHA-256 4e40b08984cd…37ca · Hugging Face
Herunterladenvoices/af_jessica.safetensors
safetensors · 510 KB · SHA-256 93930bbe363c…7003 · Hugging Face
Herunterladenvoices/af_kore.safetensors
safetensors · 510 KB · SHA-256 71fb664533d3…ed1b · Hugging Face
Herunterladenvoices/af_nicole.safetensors
safetensors · 510 KB · SHA-256 769ee38efeb1…05c6 · Hugging Face
Herunterladenvoices/af_nova.safetensors
safetensors · 510 KB · SHA-256 6bf350f6649b…03bd · Hugging Face
Herunterladenvoices/af_river.safetensors
safetensors · 510 KB · SHA-256 4068d50ad247…3019 · Hugging Face
Herunterladenvoices/af_sarah.safetensors
safetensors · 510 KB · SHA-256 2bc36ea1b089…d3f9 · Hugging Face
Herunterladenvoices/af_sky.safetensors
safetensors · 510 KB · SHA-256 7c8fafbc903f…f007 · Hugging Face
Herunterladenvoices/am_adam.safetensors
safetensors · 510 KB · SHA-256 1dff8c742132…28ce · Hugging Face
Herunterladenvoices/am_echo.safetensors
safetensors · 510 KB · SHA-256 73f4473b2b74…9cd4 · Hugging Face
Herunterladenvoices/am_eric.safetensors
safetensors · 510 KB · SHA-256 91bcd1a2c8b1…d034 · Hugging Face
Herunterladenvoices/am_fenrir.safetensors
safetensors · 510 KB · SHA-256 843c5a6abcd2…beee · Hugging Face
Herunterladenvoices/am_liam.safetensors
safetensors · 510 KB · SHA-256 dd0b4852ec6a…136d · Hugging Face
Herunterladenvoices/am_michael.safetensors
safetensors · 510 KB · SHA-256 19a866143045…a43c · Hugging Face
Herunterladenvoices/am_onyx.safetensors
safetensors · 510 KB · SHA-256 0d6ee75af61c…ba40 · Hugging Face
Herunterladenvoices/am_puck.safetensors
safetensors · 510 KB · SHA-256 a6be98717a13…9d47 · Hugging Face
Herunterladenvoices/am_santa.safetensors
safetensors · 510 KB · SHA-256 207b91319c81…8edc · Hugging Face
Herunterladenvoices/bf_alice.safetensors
safetensors · 510 KB · SHA-256 23cc78a40920…faa7 · Hugging Face
Herunterladenvoices/bf_emma.safetensors
safetensors · 510 KB · SHA-256 ed92055e1ed9…ec01 · Hugging Face
Herunterladenvoices/bf_isabella.safetensors
safetensors · 510 KB · SHA-256 3a3f580cd789…9f08 · Hugging Face
Herunterladenvoices/bf_lily.safetensors
safetensors · 510 KB · SHA-256 4cfbc152f58d…56a9 · Hugging Face
Herunterladenvoices/bm_daniel.safetensors
safetensors · 510 KB · SHA-256 35b044b866eb…582a · Hugging Face
Herunterladenvoices/bm_fable.safetensors
safetensors · 510 KB · SHA-256 146cd42da875…aa54 · Hugging Face
Herunterladenvoices/bm_george.safetensors
safetensors · 510 KB · SHA-256 a3a6682cde62…5a21 · Hugging Face
Herunterladenvoices/bm_lewis.safetensors
safetensors · 510 KB · SHA-256 a3a1a67f5f1d…8f22 · Hugging Face
Herunterladenvoices/ef_dora.safetensors
safetensors · 510 KB · SHA-256 716c9b0106ed…3266 · Hugging Face
Herunterladenvoices/em_alex.safetensors
safetensors · 510 KB · SHA-256 2af20f22c382…8e90 · Hugging Face
Herunterladenvoices/em_santa.safetensors
safetensors · 510 KB · SHA-256 c50a40686659…b68c · Hugging Face
Herunterladenvoices/ff_siwis.safetensors
safetensors · 510 KB · SHA-256 1e06666932ea…7296 · Hugging Face
Herunterladenvoices/hf_alpha.safetensors
safetensors · 510 KB · SHA-256 ca6dce871c8e…da9c · Hugging Face
Herunterladenvoices/hf_beta.safetensors
safetensors · 510 KB · SHA-256 8c8df5d3db72…68cd · Hugging Face
Herunterladenvoices/hm_omega.safetensors
safetensors · 510 KB · SHA-256 2f320b2eb557…0fb6 · Hugging Face
Herunterladenvoices/hm_psi.safetensors
safetensors · 510 KB · SHA-256 002b660d4c97…d5f2 · Hugging Face
Herunterladenvoices/if_sara.safetensors
safetensors · 510 KB · SHA-256 abbfb90313d9…88bc · Hugging Face
Herunterladenvoices/im_nicola.safetensors
safetensors · 510 KB · SHA-256 e67e89aa864a…868c · Hugging Face
Herunterladenvoices/jf_alpha.safetensors
safetensors · 510 KB · SHA-256 99b5c42d92fb…2012 · Hugging Face
Herunterladenvoices/jf_gongitsune.safetensors
safetensors · 510 KB · SHA-256 fa63996f4374…4ed4 · Hugging Face
Herunterladenvoices/jf_nezumi.safetensors
safetensors · 510 KB · SHA-256 251263822976…e443 · Hugging Face
Herunterladenvoices/jf_tebukuro.safetensors
safetensors · 510 KB · SHA-256 55708217a673…df64 · Hugging Face
Herunterladenvoices/jm_kumo.safetensors
safetensors · 510 KB · SHA-256 afe4eba3e9cc…9a8d · Hugging Face
Herunterladenvoices/pf_dora.safetensors
safetensors · 510 KB · SHA-256 908e42f986bc…91f8 · Hugging Face
Herunterladenvoices/pm_alex.safetensors
safetensors · 510 KB · SHA-256 a9bbd244dd96…a035 · Hugging Face
Herunterladenvoices/pm_santa.safetensors
safetensors · 510 KB · SHA-256 3cd98fbea1d2…6199 · Hugging Face
Herunterladenvoices/zf_xiaobei.safetensors
safetensors · 510 KB · SHA-256 07514e1ffb72…b502 · Hugging Face
Herunterladenvoices/zf_xiaoni.safetensors
safetensors · 510 KB · SHA-256 2d1331a40a7c…8a16 · Hugging Face
Herunterladenvoices/zf_xiaoxiao.safetensors
safetensors · 510 KB · SHA-256 a91ecaa5467b…df21 · Hugging Face
Herunterladenvoices/zf_xiaoyi.safetensors
safetensors · 510 KB · SHA-256 3da318864cdd…6996 · Hugging Face
Herunterladenvoices/zm_yunjian.safetensors
safetensors · 510 KB · SHA-256 dc7446f673d6…df56 · Hugging Face
Herunterladenvoices/zm_yunxi.safetensors
safetensors · 510 KB · SHA-256 f8a9865df082…8cee · Hugging Face
Herunterladenvoices/zm_yunxia.safetensors
safetensors · 510 KB · SHA-256 84dd67028466…7eae · Hugging Face
Herunterladenvoices/zm_yunyang.safetensors
safetensors · 510 KB · SHA-256 b645dadf7bbe…f126 · Hugging Face
HerunterladenQuellenauszüge
2 Auszüge
Quellenauszüge
2 Auszüge--- license: apache-2.0 language: - en pipeline_tag: text-to-speech tags: - mlx - kokoro - tts - onyx base_model: hexgrad/Kokoro-82M --- # Onyx-Kokoro-82M A mixed-precision MLX quantization of [Kokoro-82M](https://huggingface.co/hexgrad/Kokoro-82M) (StyleTTS2-derived, Apache-2.0), packaged for **[Onyx](https://github.com/)** — an on-device iOS AI assistant. Used by OnyxTTS, Onyx's own from-scratch Swift + MLX text-to-speech engine, to speak assistant replies in a natural neural voice fully on-device (Apple's AVSpeechSynthesizer is the fallback). ## What this is - **Mixed 4-bit quant** (group size 64): the PL-BERT (ALBERT) text encoder's Linear weights are 4-bit; the entire vocoder path — the prosody predictor (LSTMs + AdaIN convs → pitch/energy) and the iSTFTNet generator — stays fp16. The vocoder directly shapes the waveform and is quant-sensitive, so it is deliberately NOT quantized. ~330 MB fp32 → **148 MB** (`model.safetensors`). - **Voice packs** (`voices/*.safetensors`): 54 voices (af_/am_ American, bf_/bm_ British, plus jf/jm/zf/zm/ ef/em/ff/hf/hm/if/im/pf/pm for other languages). Each is a `[N, 1, 256]` style tensor indexed by phoneme-sequence length. - `config.json` carries the standard Kokoro config plus a `quantization` block (`{group_size: 64, bits: 4}`) so the loader dequantizes the right layers. ## Provenance Weights derived from `prince-canuma/Kokoro-82M` (the mlx-audio mirror of `hexgrad/Kokoro-82M`) via Onyx's `quantize_kokoro.py`. Verified to load + synthesize valid 24 kHz audio end-to-end through OnyxTTS on Apple Silicon. Apache-2.0, same as the base model.
Source context: 106 downloads · 1 likes · Pipeline text-to-speech · Library mlx · Repo wabibito/Onyx-Kokoro-82M
