LAM A2E β Lip Animation Model
Wav2Vec2-based model fine-tuned for real-time audio-to-expression (A2E) inference. Outputs 52 ARKit blendshape weights at 30fps for avatar lip sync.
Part of the @omote/core SDK - the first open source client-side AI character SDK for the web.
Variants
Desktop (WebGPU)
| File | Size | Description |
|---|---|---|
model_fp16.onnx |
385 KB | ONNX graph (external data format) |
model_fp16.onnx.data |
192 MB | fp16 weights |
- Backend: WebGPU (preferred) / WASM
- Input: 16kHz audio, fixed 16000 samples
- Output: 52 ARKit blendshapes + CTC ASR (dual-head)
- Precision: Surgical mixed fp16 -- decomposed LayerNorm subgraphs kept in fp32
iOS / Safari (WASM)
| File | Size | Description |
|---|---|---|
lam_ios.onnx |
230 KB | ONNX graph (external data format, opset 18) |
lam_ios.onnx.data |
192 MB | fp16 weights |
- Backend: WASM only
- Input: 16kHz audio, fixed 16000 samples
- Output: 52 ARKit blendshapes (no CTC head)
- Optimizations: 32 fused native LayerNormalization ops, 54.6% node reduction (1339β608)
Quick Start
import { createA2E } from '@omote/core';
// Zero-config β auto-detects platform, uses this HuggingFace CDN
const a2e = createA2E();
await a2e.load();
const { blendshapes } = await a2e.infer(audioSamples);
// blendshapes: Float32Array[52] β ARKit weights at 30fps
ARKit Blendshapes (52)
eyeBlinkLeft, eyeLookDownLeft, eyeLookInLeft, eyeLookOutLeft, eyeLookUpLeft,
eyeSquintLeft, eyeWideLeft, eyeBlinkRight, eyeLookDownRight, eyeLookInRight,
eyeLookOutRight, eyeLookUpRight, eyeSquintRight, eyeWideRight,
jawForward, jawLeft, jawRight, jawOpen,
mouthClose, mouthFunnel, mouthPucker, mouthLeft, mouthRight,
mouthSmileLeft, mouthSmileRight, mouthFrownLeft, mouthFrownRight,
mouthDimpleLeft, mouthDimpleRight, mouthStretchLeft, mouthStretchRight,
mouthRollLower, mouthRollUpper, mouthShrugLower, mouthShrugUpper,
mouthPressLeft, mouthPressRight, mouthLowerDownLeft, mouthLowerDownRight,
mouthUpperUpLeft, mouthUpperUpRight,
browDownLeft, browDownRight, browInnerUp, browOuterUpLeft, browOuterUpRight,
cheekPuff, cheekSquintLeft, cheekSquintRight,
noseSneerLeft, noseSneerRight, tongueOut
License
Apache 2.0
- Downloads last month
- 127
Inference Providers NEW
This model isn't deployed by any Inference Provider. π Ask for provider support