Document iOS iPhone Core AI conversion target
Browse files
README.md
CHANGED
|
@@ -1,4 +1,70 @@
|
|
| 1 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
|
| 3 |
Local iOS workbench for incrementally porting and profiling the split Kokoro
|
| 4 |
pipeline in Swift.
|
|
|
|
| 1 |
+
---
|
| 2 |
+
tags:
|
| 3 |
+
- text-to-speech
|
| 4 |
+
- coreai
|
| 5 |
+
- ios
|
| 6 |
+
- iphone
|
| 7 |
+
- arm64
|
| 8 |
+
- kokoro
|
| 9 |
+
- supertonic
|
| 10 |
+
---
|
| 11 |
+
|
| 12 |
+
# TTS Core AI Lab — converted for iOS / iPhone
|
| 13 |
+
|
| 14 |
+
This repository contains **Core AI `.aimodel` assets converted for an arm64
|
| 15 |
+
iOS application running on physical iPhone hardware**. It is the model-asset
|
| 16 |
+
companion to the [TTSCoreAILab iOS app](https://github.com/augustzheng-RED/TTSCoreAILab).
|
| 17 |
+
|
| 18 |
+
**Deployment target:** iOS 27.0+ on arm64 iPhone hardware.
|
| 19 |
+
|
| 20 |
+
**Benchmark device:** iPhone 17 Pro. This identifies the device used for the
|
| 21 |
+
current RTF benchmark; it is not a guarantee that every included profile has
|
| 22 |
+
passed end-to-end validation on every iPhone.
|
| 23 |
+
|
| 24 |
+
## What was converted
|
| 25 |
+
|
| 26 |
+
| Runtime | Upstream model | Conversion output | Intended use |
|
| 27 |
+
| --- | --- | --- | --- |
|
| 28 |
+
| Kokoro | [`hexgrad/Kokoro-82M`](https://huggingface.co/hexgrad/Kokoro-82M) PyTorch model | 11-stage Core AI `.aimodel` pipeline plus fixed-shape profiles | On-device iOS TTS research and profiling |
|
| 29 |
+
| Supertonic 3 | [`Supertone/supertonic-3`](https://huggingface.co/Supertone/supertonic-3) ONNX pipeline | Four Core AI `.aimodel` stages per fixed-shape profile | On-device iOS TTS research and profiling |
|
| 30 |
+
|
| 31 |
+
Kokoro subgraphs are exported through `torch.export` and converted with
|
| 32 |
+
`coreai_torch`. Supertonic ONNX graphs are recovered at fixed shapes, converted
|
| 33 |
+
through `onnx2torch` and `torch.export`, then converted with `coreai_torch`.
|
| 34 |
+
The resulting assets use the versioned AICode layout required by the iOS 27
|
| 35 |
+
device compiler.
|
| 36 |
+
|
| 37 |
+
These are deployment assets, not general-purpose PyTorch, ONNX, Core ML, web,
|
| 38 |
+
Android, macOS, or iOS Simulator models. Use them only with the matching
|
| 39 |
+
`TTSCoreAILab` app and its Core AI runtime integration.
|
| 40 |
+
|
| 41 |
+
## Device and profile notes
|
| 42 |
+
|
| 43 |
+
- The app targets `arm64` iPhone hardware with deployment target iOS 27.0.
|
| 44 |
+
- The iPhone 17 Pro is the current benchmark device.
|
| 45 |
+
- Kokoro packages include `64x64`, `128x256`, and `256x512` routed profiles,
|
| 46 |
+
plus legacy/reference profiles.
|
| 47 |
+
- Supertonic packages include `64x32`, `128x64`, `128x128`, and `128x256`
|
| 48 |
+
fixed-shape profiles.
|
| 49 |
+
- Asset presence does not imply physical-device validation. Validate the exact
|
| 50 |
+
app revision, profile, prompt, and device before shipping or profiling.
|
| 51 |
+
|
| 52 |
+
## App integration
|
| 53 |
+
|
| 54 |
+
The iOS project downloads the two directories below into its repository root:
|
| 55 |
+
|
| 56 |
+
- `TTSCoreAILab/Models`
|
| 57 |
+
- `TTSCoreAILab/SupertonicResources`
|
| 58 |
+
|
| 59 |
+
From the app repository, run:
|
| 60 |
+
|
| 61 |
+
```bash
|
| 62 |
+
./scripts/download-assets.sh
|
| 63 |
+
```
|
| 64 |
+
|
| 65 |
+
---
|
| 66 |
+
|
| 67 |
+
## Project overview
|
| 68 |
|
| 69 |
Local iOS workbench for incrementally porting and profiling the split Kokoro
|
| 70 |
pipeline in Swift.
|