Update: Gemma 3 1B INT4, Gemma 4 E2B + abliterated .litertlm, Kokoro v1.0
Browse files- NOTICE.md +7 -10
- README.md +4 -2
- kokoro-tts/model_q8f16.onnx +3 -0
NOTICE.md
CHANGED
|
@@ -33,19 +33,16 @@ used by the ChatBack Android app. All models run 100% on-device.
|
|
| 33 |
- **Original:** https://huggingface.co/rhasspy/piper-voices
|
| 34 |
- American English medium-quality VITS ONNX models
|
| 35 |
|
| 36 |
-
## Kokoro TTS
|
| 37 |
-
- **Author:** Kokoro TTS project
|
| 38 |
- **License:** Apache 2.0 (see LICENSE-kokoro)
|
| 39 |
-
- **Original:** https://huggingface.co/
|
| 40 |
-
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
- **Author:** espeak-ng contributors
|
| 44 |
-
- **License:** GPL 3.0 (see LICENSE-espeak-ng)
|
| 45 |
-
- **Original:** https://github.com/espeak-ng/espeak-ng
|
| 46 |
-
- Used at runtime by Sherpa-ONNX for phonemisation (not compiled into the app)
|
| 47 |
|
| 48 |
## Sherpa-ONNX
|
| 49 |
- **Author:** k2-fsa / Next-gen Kaldi
|
| 50 |
- **License:** Apache 2.0 (see LICENSE-sherpa-onnx)
|
| 51 |
- **Original:** https://github.com/k2-fsa/sherpa-onnx
|
|
|
|
|
|
| 33 |
- **Original:** https://huggingface.co/rhasspy/piper-voices
|
| 34 |
- American English medium-quality VITS ONNX models
|
| 35 |
|
| 36 |
+
## Kokoro TTS English v1.0 (q8f16 ONNX)
|
| 37 |
+
- **Author:** Kokoro TTS project (hexgrad); ONNX export by onnx-community
|
| 38 |
- **License:** Apache 2.0 (see LICENSE-kokoro)
|
| 39 |
+
- **Original:** https://huggingface.co/onnx-community/Kokoro-82M-v1.0-ONNX
|
| 40 |
+
- 28 English voices (American & British, female & male)
|
| 41 |
+
- Mixed int8-weight / float16-activation quantization for better quality than pure int8
|
| 42 |
+
- espeak-ng data NOT included — app uses its own Java IPA phonemizer
|
|
|
|
|
|
|
|
|
|
|
|
|
| 43 |
|
| 44 |
## Sherpa-ONNX
|
| 45 |
- **Author:** k2-fsa / Next-gen Kaldi
|
| 46 |
- **License:** Apache 2.0 (see LICENSE-sherpa-onnx)
|
| 47 |
- **Original:** https://github.com/k2-fsa/sherpa-onnx
|
| 48 |
+
- Previously used for Kokoro TTS packaging; no longer a dependency
|
README.md
CHANGED
|
@@ -36,8 +36,10 @@ NSFW mode is enabled in the app.
|
|
| 36 |
| Model | Size | License |
|
| 37 |
|-------|------|---------|
|
| 38 |
| **Piper TTS** — Amy, Lessac, LibriTTS-R, Ryan, Joe (ONNX medium) | ~178 MB total | MIT |
|
| 39 |
-
| **Kokoro TTS**
|
| 40 |
-
|
|
|
|
|
|
|
| 41 |
|
| 42 |
---
|
| 43 |
|
|
|
|
| 36 |
| Model | Size | License |
|
| 37 |
|-------|------|---------|
|
| 38 |
| **Piper TTS** — Amy, Lessac, LibriTTS-R, Ryan, Joe (ONNX medium) | ~178 MB total | MIT |
|
| 39 |
+
| **Kokoro TTS** English v1.0 (q8f16 ONNX, 28 voices) | ~100 MB | Apache 2.0 |
|
| 40 |
+
|
| 41 |
+
Kokoro uses mixed int8-weight/float16-activation quantization for better audio quality
|
| 42 |
+
vs pure int8. The app uses its own Java IPA phonemizer — no espeak-ng dependency.
|
| 43 |
|
| 44 |
---
|
| 45 |
|
kokoro-tts/model_q8f16.onnx
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:04c658aec1b6008857c2ad10f8c589d4180d0ec427e7e6118ceb487e215c3cd0
|
| 3 |
+
size 86033585
|