@omote/core 0.5.4 → 0.5.5

package/dist/index.d.mts

@@ -584,6 +584,14 @@ interface FullFacePipelineOptions {
     chunkSize?: number;
     /** A2E inference engine */
     lam: A2EBackend;
+    /**
+     * Identity/style index for the A2E model (default: 0).
+     *
+     * The LAM model uses a 12-class one-hot identity vector as style conditioning.
+     * Different indices produce different expression intensity across face regions.
+     * Only affects Wav2Vec2Inference (GPU). Wav2ArkitCpuInference has identity 11 baked in.
+     */
+    identityIndex?: number;
     /** Per-character expression weight scaling */
     profile?: ExpressionProfile;
     /**
@@ -2259,6 +2267,17 @@ interface A2EProcessorConfig {
     sampleRate?: number;
     /** Samples per inference chunk (default: 16000 = 1s) */
     chunkSize?: number;
+    /**
+     * Identity/style index for the A2E model (default: 0).
+     *
+     * The LAM model uses a one-hot identity vector (12 classes, indices 0-11) as
+     * style conditioning alongside audio features. Different indices produce
+     * different expression intensity across face regions (brows, eyes, cheeks).
+     *
+     * Only affects Wav2Vec2Inference (GPU model). Wav2ArkitCpuInference has
+     * identity 11 baked into the model weights.
+     */
+    identityIndex?: number;
     /** Callback fired with each blendshape frame (push mode) */
     onFrame?: (frame: Float32Array) => void;
     /** Error callback */
@@ -2268,6 +2287,7 @@ declare class A2EProcessor {
     private readonly backend;
     private readonly sampleRate;
     private readonly chunkSize;
+    private readonly identityIndex;
     private readonly onFrame?;
     private readonly onError?;
     private bufferCapacity;

package/dist/index.d.ts

@@ -584,6 +584,14 @@ interface FullFacePipelineOptions {
     chunkSize?: number;
     /** A2E inference engine */
     lam: A2EBackend;
+    /**
+     * Identity/style index for the A2E model (default: 0).
+     *
+     * The LAM model uses a 12-class one-hot identity vector as style conditioning.
+     * Different indices produce different expression intensity across face regions.
+     * Only affects Wav2Vec2Inference (GPU). Wav2ArkitCpuInference has identity 11 baked in.
+     */
+    identityIndex?: number;
     /** Per-character expression weight scaling */
     profile?: ExpressionProfile;
     /**
@@ -2259,6 +2267,17 @@ interface A2EProcessorConfig {
     sampleRate?: number;
     /** Samples per inference chunk (default: 16000 = 1s) */
     chunkSize?: number;
+    /**
+     * Identity/style index for the A2E model (default: 0).
+     *
+     * The LAM model uses a one-hot identity vector (12 classes, indices 0-11) as
+     * style conditioning alongside audio features. Different indices produce
+     * different expression intensity across face regions (brows, eyes, cheeks).
+     *
+     * Only affects Wav2Vec2Inference (GPU model). Wav2ArkitCpuInference has
+     * identity 11 baked into the model weights.
+     */
+    identityIndex?: number;
     /** Callback fired with each blendshape frame (push mode) */
     onFrame?: (frame: Float32Array) => void;
     /** Error callback */
@@ -2268,6 +2287,7 @@ declare class A2EProcessor {
     private readonly backend;
     private readonly sampleRate;
     private readonly chunkSize;
+    private readonly identityIndex;
     private readonly onFrame?;
     private readonly onError?;
     private bufferCapacity;

package/dist/index.js

@@ -916,6 +916,7 @@ var A2EProcessor = class {
     this.backend = config.backend;
     this.sampleRate = config.sampleRate ?? 16e3;
     this.chunkSize = config.chunkSize ?? config.backend.chunkSize ?? 16e3;
+    this.identityIndex = config.identityIndex ?? 0;
     this.onFrame = config.onFrame;
     this.onError = config.onError;
     this.bufferCapacity = this.chunkSize * 2;
@@ -1123,7 +1124,7 @@ var A2EProcessor = class {
         const { chunk, timestamp } = this.pendingChunks.shift();
         try {
           const t0 = performance.now();
-          const result = await this.backend.infer(chunk);
+          const result = await this.backend.infer(chunk, this.identityIndex);
           const inferMs = Math.round(performance.now() - t0);
           const actualDuration = chunk.length / this.sampleRate;
           const actualFrameCount = Math.ceil(actualDuration * FRAME_RATE);
@@ -3247,6 +3248,7 @@ var FullFacePipeline = class extends EventEmitter {
       backend: options.lam,
       sampleRate,
       chunkSize,
+      identityIndex: options.identityIndex,
       onError: (error) => {
         logger4.error("A2E inference error", { message: error.message, stack: error.stack });
         this.emit("error", error);