Upload 2 files

punctuator.js

@@ -36,13 +36,22 @@ const MULTILINGUAL_LABELS = {
 const MULTILINGUAL_LANGS = ['de', 'fr', 'it', 'nl', 'es', 'pt'];
 
 // Load the English punctuator model and vocab
+async function cachedFetch(url) {
+    const cache = await caches.open('granite-speech-local-models');
+    const cached = await cache.match(url);
+    if (cached) return cached;
+    const response = await fetch(url);
+    if (response.ok) await cache.put(url, response.clone());
+    return response;
+}
+
 async function loadEnglishPunctuator() {
     if (pcsSession) return;
 
     console.log('Loading English punctuator model...');
 
     // Load vocab
-    const vocabResponse = await fetch('./pcs_vocab.json');
+    const vocabResponse = await cachedFetch('./pcs_vocab.json');
     const vocabData = await vocabResponse.json();
     pcsVocab = vocabData.vocab;
 
@@ -53,7 +62,9 @@ async function loadEnglishPunctuator() {
     }
 
     // Load ONNX model
-    pcsSession = await ort.InferenceSession.create('./punct_cap_seg_en.onnx', {
+    const modelResponse = await cachedFetch('./punct_cap_seg_en.onnx');
+    const buffer = await modelResponse.arrayBuffer();
+    pcsSession = await ort.InferenceSession.create(buffer, {
         executionProviders: ['wasm'],
     });
 
@@ -71,7 +82,9 @@ async function loadMultilingualPunctuator() {
     multilingualTokenizer = await AutoTokenizer.from_pretrained('oliverguhr/fullstop-punctuation-multilingual-base');
 
     // Load ONNX model
-    multilingualSession = await ort.InferenceSession.create('./punct_multilingual_q8.onnx', {
+    const modelResponse = await cachedFetch('./punct_multilingual_q8.onnx');
+    const buffer = await modelResponse.arrayBuffer();
+    multilingualSession = await ort.InferenceSession.create(buffer, {
         executionProviders: ['wasm'],
     });
 

vad.js

@@ -7,12 +7,26 @@ let vadSession = null;
 const VAD_SAMPLE_RATE = 16000;
 const VAD_CHUNK_SIZE = 512;  // 32ms chunks at 16kHz
 
+const MODEL_CACHE_NAME = 'granite-speech-local-models';
+
+// Fetch with Cache API persistence
+async function cachedFetch(url) {
+    const cache = await caches.open(MODEL_CACHE_NAME);
+    const cached = await cache.match(url);
+    if (cached) return cached;
+    const response = await fetch(url);
+    if (response.ok) await cache.put(url, response.clone());
+    return response;
+}
+
 // Load VAD model
 async function loadVAD() {
     if (vadSession) return;
 
     console.log('Loading VAD model...');
-    vadSession = await ort.InferenceSession.create('./silero_vad.onnx', {
+    const response = await cachedFetch('./silero_vad.onnx');
+    const buffer = await response.arrayBuffer();
+    vadSession = await ort.InferenceSession.create(buffer, {
         executionProviders: ['wasm'],
     });
     console.log('VAD model loaded');