Hugging Face's logo

Pixel-Labs
/
threadcast-neural-models

Text-to-Speech
ONNX
KittenTTS
English
tts
kokoro
piper
melotts
vits
vits2
styletts2
sherpa-onnx
on-device
threadcast
Model card Files
xet
 Community
Pixel-Labs commited on
Commit
b786514
·
verified ·
1 Parent(s): 378619e

Upload README.md

Browse files
Files changed (1) hide show
  1. README.md +9 -4
README.md CHANGED
@@ -7,6 +7,7 @@ tags:
7
  - tts
8
  - kokoro
9
  - piper
 
10
  - vits
11
  - styletts2
12
  - onnx
@@ -31,7 +32,7 @@ pipeline_tag: text-to-speech
31
 
32
  ---
33
 
34
- Self-hosted mirror of the on-device neural TTS models used by **[ThreadCast](https://threadcast.app)** across both shipping platforms — the Chrome extension and the Android app. Two engines, two packagings, one source of truth.
35
 
36
  This repository exists so each platform can ship a stable, version-pinned set of model weights without depending on the availability or rate-limits of upstream Hugging Face repos at runtime.
37
 
@@ -48,7 +49,9 @@ threadcast-neural-models/
48
  │ └── neural-82m/ Kokoro StyleTTS2 — 1 model + 11 voice embeddings
49
  │
50
  └── mobile-android/ ← Android app — production zips fetched at runtime
51
- └── v1/ 7 zips: 1 shared + 5 per-voice Piper + 1 Kokoro
 
 
52
  ```
53
 
54
  | Subtree | Format | Consumed by | Sub-README |
@@ -64,8 +67,9 @@ The two subtrees parallel each other on purpose — same engine families (Piper
64
 
65
  | Engine | Architecture | Params | Per-voice cost | Quality tier |
66
  |---|---|---|---|---|
67
- | **`neural-28m`** | Piper VITS | ~28 M | One ONNX file per voice (~63 MB) | Standard — fast, CPU-friendly, single-thread WASM real-time on a laptop |
68
- | **`neural-82m`** | Kokoro StyleTTS2 | ~82 M | Single model + 256-dim style vectors per voice (one ~325 MB file serves all) | Premium — more natural prosody, GPU-accelerated on Chrome (WebGPU); CPU-only on Android (perf-gated) |
 
69
 
70
  ---
71
 
@@ -74,6 +78,7 @@ The two subtrees parallel each other on purpose — same engine families (Piper
74
  This repository **mirrors** upstream models for distribution stability. Each upstream project retains its own license:
75
 
76
  - **Kokoro-82M:** Apache-2.0 ([upstream model card](https://huggingface.co/hexgrad/Kokoro-82M))
 
77
  - **Piper voices:** MIT, with individual voice attributions in each `.onnx.json`
78
  - **transformers.js, onnxruntime-web, onnxruntime-android:** Apache-2.0
79
  - **sherpa-onnx:** Apache-2.0
 
7
  - tts
8
  - kokoro
9
  - piper
10
+ - kittentts
11
  - vits
12
  - styletts2
13
  - onnx
 
32
 
33
  ---
34
 
35
+ Self-hosted mirror of the on-device neural TTS models used by **[ThreadCast](https://threadcast.app)** across both shipping platforms — the Chrome extension and the Android app. Three engine families on Android (Piper VITS, KittenTTS-nano VITS, Kokoro StyleTTS2), two on the extension, one source of truth.
36
 
37
  This repository exists so each platform can ship a stable, version-pinned set of model weights without depending on the availability or rate-limits of upstream Hugging Face repos at runtime.
38
 
 
49
  │ └── neural-82m/ Kokoro StyleTTS2 — 1 model + 11 voice embeddings
50
  │
51
  └── mobile-android/ ← Android app — production zips fetched at runtime
52
+ └── v1/ 8 zips: 1 shared espeak + 5 per-voice Piper
53
+ + 1 KittenTTS-nano ("Local AI Plus")
54
+ + 1 Kokoro ("Local AI Studio")
55
  ```
56
 
57
  | Subtree | Format | Consumed by | Sub-README |
 
67
 
68
  | Engine | Architecture | Params | Per-voice cost | Quality tier |
69
  |---|---|---|---|---|
70
+ | **`neural-28m`** | Piper VITS | ~28 M | One ONNX file per voice (~63 MB) | Standard — fast, CPU-friendly, single-thread WASM real-time on a laptop. Surfaced on Android as **Local AI Lite**, on the extension as **AI Neural CPU**. |
71
+ | **`neural-15m`** | KittenTTS-nano VITS | ~15 M | Single fp16 model + 8 speaker embeddings (one ~26 MB file serves all) | Sweet spot — 8 voices with style-vector switching at a fraction of the storage cost. Android-only, surfaced as **Local AI Plus**. |
72
+ | **`neural-82m`** | Kokoro StyleTTS2 | ~82 M | Single model + 256-dim style vectors per voice (one ~325 MB file serves all) | Premium — more natural prosody, GPU-accelerated on Chrome (WebGPU); CPU-only on Android (perf-gated). Surfaced on Android as **Local AI Studio**, on the extension as **AI Neural GPU**. |
73
 
74
  ---
75
 
 
78
  This repository **mirrors** upstream models for distribution stability. Each upstream project retains its own license:
79
 
80
  - **Kokoro-82M:** Apache-2.0 ([upstream model card](https://huggingface.co/hexgrad/Kokoro-82M))
81
+ - **KittenTTS-nano (v0.1):** Apache-2.0 ([upstream model card](https://huggingface.co/KittenML/kitten-tts-nano-0.1))
82
  - **Piper voices:** MIT, with individual voice attributions in each `.onnx.json`
83
  - **transformers.js, onnxruntime-web, onnxruntime-android:** Apache-2.0
84
  - **sherpa-onnx:** Apache-2.0