@omote/core 0.5.4 → 0.5.6

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;
@@ -3110,6 +3130,12 @@ interface FetchWithCacheOptions {
     validateStale?: boolean;
     /** Progress callback during download */
     onProgress?: (loaded: number, total: number) => void;
+    /** Timeout per fetch attempt in ms. Default: 120_000 (2 min) */
+    timeoutMs?: number;
+    /** Max retry attempts on failure. Default: 2 (3 total attempts) */
+    maxRetries?: number;
+    /** AbortSignal to cancel the entire operation */
+    signal?: AbortSignal;
 }
 /**
  * Fetch a model with caching

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;
@@ -3110,6 +3130,12 @@ interface FetchWithCacheOptions {
     validateStale?: boolean;
     /** Progress callback during download */
     onProgress?: (loaded: number, total: number) => void;
+    /** Timeout per fetch attempt in ms. Default: 120_000 (2 min) */
+    timeoutMs?: number;
+    /** Max retry attempts on failure. Default: 2 (3 total attempts) */
+    maxRetries?: number;
+    /** AbortSignal to cancel the entire operation */
+    signal?: AbortSignal;
 }
 /**
  * Fetch a model with caching

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);
@@ -2218,6 +2219,16 @@ function getModelCache() {
   return cacheInstance;
 }
 var MAX_CACHE_SIZE_BYTES = 500 * 1024 * 1024;
+function fetchWithTimeout(url, timeoutMs, signal) {
+  const controller = new AbortController();
+  const timer = setTimeout(() => controller.abort(), timeoutMs);
+  const onCallerAbort = () => controller.abort();
+  signal?.addEventListener("abort", onCallerAbort, { once: true });
+  return fetch(url, { signal: controller.signal }).finally(() => {
+    clearTimeout(timer);
+    signal?.removeEventListener("abort", onCallerAbort);
+  });
+}
 async function fetchWithCache(url, optionsOrProgress) {
   let options = {};
   if (typeof optionsOrProgress === "function") {
@@ -2271,61 +2282,84 @@ async function fetchWithCache(url, optionsOrProgress) {
   }
   span?.setAttributes({ "fetch.cache_hit": false });
   console.log(`[ModelCache] Cache miss, fetching: ${url}`);
-  try {
-    const response = await fetch(url);
-    if (!response.ok) {
-      throw new Error(`Failed to fetch ${url}: ${response.status}`);
-    }
-    const contentLength = response.headers.get("content-length");
-    const total = contentLength ? parseInt(contentLength, 10) : 0;
-    const etag = response.headers.get("etag") ?? void 0;
-    const tooLargeForCache = total > MAX_CACHE_SIZE_BYTES;
-    if (tooLargeForCache) {
-      console.log(`[ModelCache] File too large for IndexedDB (${(total / 1024 / 1024).toFixed(0)}MB > 500MB), using HTTP cache only`);
-    }
-    if (!response.body) {
-      const data2 = await response.arrayBuffer();
+  const timeout = options.timeoutMs ?? 12e4;
+  const maxRetries = options.maxRetries ?? 2;
+  let lastError = null;
+  for (let attempt = 0; attempt <= maxRetries; attempt++) {
+    if (options.signal?.aborted) {
+      throw new Error(`Fetch aborted for ${url}`);
+    }
+    if (attempt > 0) {
+      const backoff = Math.min(2e3 * Math.pow(2, attempt - 1), 16e3);
+      console.log(`[ModelCache] Retry ${attempt}/${maxRetries} after ${backoff}ms: ${url}`);
+      await new Promise((r) => setTimeout(r, backoff));
+    }
+    try {
+      const response = await fetchWithTimeout(url, timeout, options.signal);
+      if (!response.ok) {
+        throw new Error(`Failed to fetch ${url}: ${response.status}`);
+      }
+      const contentLength = response.headers.get("content-length");
+      const total = contentLength ? parseInt(contentLength, 10) : 0;
+      const etag = response.headers.get("etag") ?? void 0;
+      const tooLargeForCache = total > MAX_CACHE_SIZE_BYTES;
+      if (tooLargeForCache) {
+        console.log(`[ModelCache] File too large for IndexedDB (${(total / 1024 / 1024).toFixed(0)}MB > 500MB), using HTTP cache only`);
+      }
+      if (!response.body) {
+        const data2 = await response.arrayBuffer();
+        if (!tooLargeForCache) {
+          await cache.set(cacheKey, data2, etag, version);
+        }
+        span?.setAttributes({
+          "fetch.size_bytes": data2.byteLength,
+          "fetch.cached_to_indexeddb": !tooLargeForCache,
+          ...attempt > 0 && { "fetch.retry_count": attempt }
+        });
+        span?.end();
+        return data2;
+      }
+      const reader = response.body.getReader();
+      const chunks = [];
+      let loaded = 0;
+      while (true) {
+        const { done, value } = await reader.read();
+        if (done) break;
+        chunks.push(value);
+        loaded += value.length;
+        onProgress?.(loaded, total || loaded);
+      }
+      const data = new Uint8Array(loaded);
+      let offset = 0;
+      for (const chunk of chunks) {
+        data.set(chunk, offset);
+        offset += chunk.length;
+      }
+      const buffer = data.buffer;
       if (!tooLargeForCache) {
-        await cache.set(cacheKey, data2, etag, version);
+        await cache.set(cacheKey, buffer, etag, version);
+        console.log(`[ModelCache] Cached: ${url} (${(buffer.byteLength / 1024 / 1024).toFixed(1)}MB)`);
       }
       span?.setAttributes({
-        "fetch.size_bytes": data2.byteLength,
-        "fetch.cached_to_indexeddb": !tooLargeForCache
+        "fetch.size_bytes": buffer.byteLength,
+        "fetch.cached_to_indexeddb": !tooLargeForCache,
+        ...attempt > 0 && { "fetch.retry_count": attempt }
       });
       span?.end();
-      return data2;
-    }
-    const reader = response.body.getReader();
-    const chunks = [];
-    let loaded = 0;
-    while (true) {
-      const { done, value } = await reader.read();
-      if (done) break;
-      chunks.push(value);
-      loaded += value.length;
-      onProgress?.(loaded, total || loaded);
-    }
-    const data = new Uint8Array(loaded);
-    let offset = 0;
-    for (const chunk of chunks) {
-      data.set(chunk, offset);
-      offset += chunk.length;
-    }
-    const buffer = data.buffer;
-    if (!tooLargeForCache) {
-      await cache.set(cacheKey, buffer, etag, version);
-      console.log(`[ModelCache] Cached: ${url} (${(buffer.byteLength / 1024 / 1024).toFixed(1)}MB)`);
+      return buffer;
+    } catch (error) {
+      lastError = error instanceof Error ? error : new Error(String(error));
+      if (options.signal?.aborted) {
+        span?.endWithError(lastError);
+        throw lastError;
+      }
+      if (attempt < maxRetries) {
+        console.warn(`[ModelCache] Attempt ${attempt + 1} failed for ${url}: ${lastError.message}`);
+      }
     }
-    span?.setAttributes({
-      "fetch.size_bytes": buffer.byteLength,
-      "fetch.cached_to_indexeddb": !tooLargeForCache
-    });
-    span?.end();
-    return buffer;
-  } catch (error) {
-    span?.endWithError(error instanceof Error ? error : new Error(String(error)));
-    throw error;
   }
+  span?.endWithError(lastError);
+  throw lastError;
 }
 async function preloadModels(urls, onProgress) {
   const cache = getModelCache();
@@ -2574,6 +2608,15 @@ function getSessionOptions(backend) {
     graphOptimizationLevel: "all"
   };
 }
+function withTimeout(promise, ms, label) {
+  return new Promise((resolve, reject) => {
+    const timer = setTimeout(
+      () => reject(new Error(`${label} timed out after ${ms}ms`)),
+      ms
+    );
+    promise.then(resolve, reject).finally(() => clearTimeout(timer));
+  });
+}
 
 // src/inference/blendshapeUtils.ts
 var LAM_BLENDSHAPES = [
@@ -2846,7 +2889,11 @@ var _Wav2Vec2Inference = class _Wav2Vec2Inference {
           )
         });
         try {
-          this.session = await this.ort.InferenceSession.create(modelUrl, sessionOptions);
+          this.session = await withTimeout(
+            this.ort.InferenceSession.create(modelUrl, sessionOptions),
+            18e4,
+            "Wav2Vec2 InferenceSession.create (iOS URL pass-through)"
+          );
         } catch (sessionErr) {
           logger3.error("iOS: InferenceSession.create() failed", {
             error: sessionErr instanceof Error ? sessionErr.message : String(sessionErr),
@@ -3247,6 +3294,7 @@ 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);
@@ -3933,9 +3981,10 @@ var _SenseVoiceInference = class _SenseVoiceInference {
         logger5.info("iOS: passing model URL directly to ORT (low-memory path)", {
           modelUrl: this.config.modelUrl
         });
-        this.session = await this.ort.InferenceSession.create(
-          this.config.modelUrl,
-          sessionOptions
+        this.session = await withTimeout(
+          this.ort.InferenceSession.create(this.config.modelUrl, sessionOptions),
+          18e4,
+          "SenseVoice InferenceSession.create (iOS URL pass-through)"
         );
       } else {
         const cache = getModelCache();
@@ -4188,6 +4237,12 @@ var WORKER_SCRIPT = `
 var ort = null;
 var session = null;
 var tokenMap = null;
+
+function fetchWithTimeout(url, timeoutMs) {
+  var controller = new AbortController();
+  var timer = setTimeout(function() { controller.abort(); }, timeoutMs);
+  return fetch(url, { signal: controller.signal }).finally(function() { clearTimeout(timer); });
+}
 var negMean = null;
 var invStddev = null;
 var languageId = 0;
@@ -4631,7 +4686,7 @@ async function loadOrt(wasmPaths) {
   var ortUrl = wasmPaths + 'ort.wasm.min.js';
 
   // Load the script by fetching and executing it
-  var response = await fetch(ortUrl);
+  var response = await fetchWithTimeout(ortUrl, 30000);
   var scriptText = await response.text();
 
   // Create a blob URL for the script
@@ -4657,7 +4712,7 @@ async function loadOrt(wasmPaths) {
  */
 async function loadModel(modelUrl, tokensUrl, isIOSDevice, lang, textNorm) {
   // 1. Fetch and parse tokens.txt
-  var tokensResponse = await fetch(tokensUrl);
+  var tokensResponse = await fetchWithTimeout(tokensUrl, 30000);
   if (!tokensResponse.ok) {
     throw new Error('Failed to fetch tokens.txt: ' + tokensResponse.status + ' ' + tokensResponse.statusText);
   }
@@ -4680,7 +4735,7 @@ async function loadModel(modelUrl, tokensUrl, isIOSDevice, lang, textNorm) {
     session = await ort.InferenceSession.create(modelUrl, sessionOptions);
   } else {
     // Desktop: fetch ArrayBuffer for potential caching
-    var modelResponse = await fetch(modelUrl);
+    var modelResponse = await fetchWithTimeout(modelUrl, 120000);
     if (!modelResponse.ok) {
       throw new Error('Failed to fetch model: ' + modelResponse.status + ' ' + modelResponse.statusText);
     }
@@ -5191,6 +5246,12 @@ var WORKER_SCRIPT2 = `
 
 var ort = null;
 
+function fetchWithTimeout(url, timeoutMs) {
+  var controller = new AbortController();
+  var timer = setTimeout(function() { controller.abort(); }, timeoutMs);
+  return fetch(url, { signal: controller.signal }).finally(function() { clearTimeout(timer); });
+}
+
 // SenseVoice state
 var svSession = null;
 var svTokenMap = null;
@@ -5505,7 +5566,7 @@ function symmetrizeBlendshapes(frame) {
 async function loadOrt(wasmPaths, isIOSDevice) {
   if (ort) return;
   var ortUrl = wasmPaths + 'ort.wasm.min.js';
-  var response = await fetch(ortUrl);
+  var response = await fetchWithTimeout(ortUrl, 30000);
   var scriptText = await response.text();
   var blob = new Blob([scriptText], { type: 'application/javascript' });
   var blobUrl = URL.createObjectURL(blob);
@@ -5523,7 +5584,7 @@ async function loadOrt(wasmPaths, isIOSDevice) {
 // \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550
 
 async function svLoad(msg) {
-  var tokensResponse = await fetch(msg.tokensUrl);
+  var tokensResponse = await fetchWithTimeout(msg.tokensUrl, 30000);
   if (!tokensResponse.ok) throw new Error('Failed to fetch tokens.txt: ' + tokensResponse.status);
   var tokensText = await tokensResponse.text();
   svTokenMap = parseTokensFile(tokensText);
@@ -5534,7 +5595,7 @@ async function svLoad(msg) {
   if (msg.isIOS) {
     svSession = await ort.InferenceSession.create(msg.modelUrl, sessionOptions);
   } else {
-    var modelResponse = await fetch(msg.modelUrl);
+    var modelResponse = await fetchWithTimeout(msg.modelUrl, 120000);
     if (!modelResponse.ok) throw new Error('Failed to fetch model: ' + modelResponse.status);
     var modelBuffer = await modelResponse.arrayBuffer();
     svSession = await ort.InferenceSession.create(new Uint8Array(modelBuffer), sessionOptions);
@@ -5609,11 +5670,11 @@ async function cpuLoad(msg) {
     }
     cpuSession = await ort.InferenceSession.create(msg.modelUrl, sessionOptions);
   } else {
-    var graphResponse = await fetch(msg.modelUrl);
+    var graphResponse = await fetchWithTimeout(msg.modelUrl, 120000);
     if (!graphResponse.ok) throw new Error('Failed to fetch model graph: ' + graphResponse.status);
     var graphBuffer = await graphResponse.arrayBuffer();
     if (msg.externalDataUrl && dataFilename) {
-      var dataResponse = await fetch(msg.externalDataUrl);
+      var dataResponse = await fetchWithTimeout(msg.externalDataUrl, 120000);
       if (!dataResponse.ok) throw new Error('Failed to fetch external data: ' + dataResponse.status);
       var dataBuffer = await dataResponse.arrayBuffer();
       sessionOptions.externalData = [{ path: dataFilename, data: new Uint8Array(dataBuffer) }];
@@ -5665,7 +5726,7 @@ async function vadLoad(msg) {
   vadChunkSize = vadSampleRate === 16000 ? 512 : 256;
   vadContextSize = vadSampleRate === 16000 ? 64 : 32;
 
-  var response = await fetch(msg.modelUrl);
+  var response = await fetchWithTimeout(msg.modelUrl, 60000);
   if (!response.ok) throw new Error('Failed to fetch VAD model: ' + response.status);
   var modelBuffer = await response.arrayBuffer();
   vadSession = await ort.InferenceSession.create(new Uint8Array(modelBuffer), {
@@ -6513,7 +6574,11 @@ var _Wav2ArkitCpuInference = class _Wav2ArkitCpuInference {
             // URL string — ORT fetches directly into WASM
           }];
         }
-        this.session = await this.ort.InferenceSession.create(modelUrl, sessionOptions);
+        this.session = await withTimeout(
+          this.ort.InferenceSession.create(modelUrl, sessionOptions),
+          18e4,
+          "Wav2ArkitCpu InferenceSession.create (iOS URL pass-through)"
+        );
       } else {
         const cache = getModelCache();
         const isCached = await cache.has(modelUrl);
@@ -6781,6 +6846,12 @@ var WORKER_SCRIPT3 = `
 var ort = null;
 var session = null;
 
+function fetchWithTimeout(url, timeoutMs) {
+  var controller = new AbortController();
+  var timer = setTimeout(function() { controller.abort(); }, timeoutMs);
+  return fetch(url, { signal: controller.signal }).finally(function() { clearTimeout(timer); });
+}
+
 // Precomputed symmetric index pairs from LAM_BLENDSHAPES alphabetical ordering
 // Used to average left/right blendshape pairs for symmetrized output
 const SYMMETRIC_INDEX_PAIRS = [
@@ -6830,7 +6901,7 @@ async function loadOrt(wasmPaths) {
   const ortUrl = wasmPaths + 'ort.wasm.min.js';
 
   // Load the script by fetching and executing it
-  const response = await fetch(ortUrl);
+  const response = await fetchWithTimeout(ortUrl, 30000);
   const scriptText = await response.text();
 
   // Create a blob URL for the script
@@ -6872,7 +6943,7 @@ async function loadModel(modelUrl, externalDataUrl, isIOS) {
     session = await ort.InferenceSession.create(modelUrl, sessionOptions);
   } else {
     // Desktop: fetch model graph as ArrayBuffer
-    const graphResponse = await fetch(modelUrl);
+    const graphResponse = await fetchWithTimeout(modelUrl, 120000);
     if (!graphResponse.ok) {
       throw new Error('Failed to fetch model graph: ' + graphResponse.status + ' ' + graphResponse.statusText);
     }
@@ -6880,7 +6951,7 @@ async function loadModel(modelUrl, externalDataUrl, isIOS) {
 
     // Fetch external data file if present
     if (externalDataUrl && dataFilename) {
-      const dataResponse = await fetch(externalDataUrl);
+      const dataResponse = await fetchWithTimeout(externalDataUrl, 120000);
       if (!dataResponse.ok) {
         throw new Error('Failed to fetch external data: ' + dataResponse.status + ' ' + dataResponse.statusText);
       }
@@ -8015,6 +8086,13 @@ var WORKER_SCRIPT4 = `
 
 var ort = null;
 var session = null;
+
+function fetchWithTimeout(url, timeoutMs) {
+  var controller = new AbortController();
+  var timer = setTimeout(function() { controller.abort(); }, timeoutMs);
+  return fetch(url, { signal: controller.signal }).finally(function() { clearTimeout(timer); });
+}
+
 var sampleRate = 16000;
 var chunkSize = 512;
 var contextSize = 64;
@@ -8030,7 +8108,7 @@ async function loadOrt(wasmPaths) {
   const ortUrl = wasmPaths + 'ort.wasm.min.js';
 
   // Load the script by fetching and executing it
-  const response = await fetch(ortUrl);
+  const response = await fetchWithTimeout(ortUrl, 30000);
   const scriptText = await response.text();
 
   // Create a blob URL for the script
@@ -8060,7 +8138,7 @@ async function loadModel(modelUrl, sr) {
   contextSize = sr === 16000 ? 64 : 32;
 
   // Fetch model data
-  const response = await fetch(modelUrl);
+  const response = await fetchWithTimeout(modelUrl, 60000);
   if (!response.ok) {
     throw new Error('Failed to fetch model: ' + response.status + ' ' + response.statusText);
   }