@fonoster/autopilot 0.8.49 → 0.8.50

package/dist/envs.js

@@ -39,7 +39,7 @@ exports.KNOWLEDGE_BASE_ENABLED = e.KNOWLEDGE_BASE_ENABLED === "true";
 exports.NODE_ENV = e.NODE_ENV || "production";
 exports.UNSTRUCTURED_API_KEY = e.UNSTRUCTURED_API_KEY;
 exports.UNSTRUCTURED_API_URL = e.UNSTRUCTURED_API_URL ?? "https://api.unstructuredapp.io/general/v0/general";
-exports.SILERO_VAD_VERSION = e.SILERO_VAD_VERSION ?? "v4";
+exports.SILERO_VAD_VERSION = e.SILERO_VAD_VERSION ?? "v5";
 exports.CONVERSATION_PROVIDER = e.CONVERSATION_PROVIDER
     ? e.CONVERSATION_PROVIDER
     : types_1.ConversationProvider.FILE;

package/dist/vad/createVad.js

@@ -59,7 +59,6 @@ const ort = __importStar(require("onnxruntime-node"));
 const chunkToFloat32Array_1 = require("./chunkToFloat32Array");
 const SileroVadModel_1 = require("./SileroVadModel");
 const logger = (0, logger_1.getLogger)({ service: "autopilot", filePath: __filename });
-// Updated constants from v5
 const FULL_FRAME_SIZE = 1600; // Equivalent to 100ms @ 16kHz
 const FRAME_SIZE = 480; // Use last 30ms from the full frame for VAD processing
 async function createVad(params) {
@@ -78,7 +77,6 @@ async function createVad(params) {
         logger.silly("State reset -- audioBuffer cleared");
     };
     return async function process(chunk, callback) {
-        // Append new samples to the audio buffer
         const float32Array = (0, chunkToFloat32Array_1.chunkToFloat32Array)(chunk);
         audioBuffer.push(...float32Array);
         // Process full frames from the buffer
@@ -96,11 +94,11 @@ async function createVad(params) {
                 framesSinceStateChange,
                 pendingSamples: audioBuffer.length
             });
-            // Increment the frames counter
             framesSinceStateChange++;
             if (isSpeechActive) {
                 // If currently in speech, check if the score has dropped below the deactivation threshold
-                if (rawScore < deactivationThreshold && framesSinceStateChange >= debounceFrames) {
+                if (rawScore < deactivationThreshold &&
+                    framesSinceStateChange >= debounceFrames) {
                     isSpeechActive = false;
                     callback("SPEECH_END");
                     silero.resetState(); // Reset VAD state after speech ends
@@ -110,7 +108,8 @@ async function createVad(params) {
             }
             else {
                 // If not currently in speech, check if the score exceeds the activation threshold
-                if (rawScore > activationThreshold && framesSinceStateChange >= debounceFrames) {
+                if (rawScore > activationThreshold &&
+                    framesSinceStateChange >= debounceFrames) {
                     isSpeechActive = true;
                     framesSinceStateChange = 0;
                     callback("SPEECH_START");

package/dist/vadv5/SileroVadModel.js

@@ -54,7 +54,10 @@ class SileroVadModel {
         this._state = getNewState(this.ort);
     }
     async process(audioFrame) {
-        const inputTensor = new this.ort.Tensor("float32", audioFrame, [1, audioFrame.length]);
+        const inputTensor = new this.ort.Tensor("float32", audioFrame, [
+            1,
+            audioFrame.length
+        ]);
         const feeds = {
             input: inputTensor,
             state: this._state,

package/dist/vadv5/createVad.js

@@ -59,8 +59,8 @@ const SileroVadModel_1 = require("./SileroVadModel");
 const logger_1 = require("@fonoster/logger");
 const path_1 = require("path");
 const logger = (0, logger_1.getLogger)({ service: "autopilot", filePath: __filename });
-const FULL_FRAME_SIZE = 1024; // 100ms @ 16kHz
-const BUFFER_SIZE = 512; // 30ms @ 16kHz
+const FULL_FRAME_SIZE = 1024; // 64ms @ 16kHz
+const BUFFER_SIZE = 512; // 32ms @ 16kHz
 async function createVad(params) {
     const { pathToModel, activationThreshold, deactivationThreshold, debounceFrames } = params;
     const effectivePath = pathToModel || (0, path_1.join)(__dirname, "..", "..", "silero_vad_v5.onnx");
@@ -88,7 +88,7 @@ async function createVad(params) {
             // Use the last BUFFER_SIZE samples from the full frame.
             const frame = fullFrame.slice(fullFrame.length - BUFFER_SIZE);
             const result = await silero.process(new Float32Array(frame));
-            const rawScore = result.isSpeech; // rawScore is assumed to be between 0 and 1
+            const rawScore = result.isSpeech;
             logger.silly("Frame processing", {
                 rawScore,
                 isSpeechActive,
@@ -98,7 +98,8 @@ async function createVad(params) {
             framesSinceStateChange++;
             if (isSpeechActive) {
                 // If already in speech, check if the score has dropped below deactivationThreshold
-                if (rawScore < deactivationThreshold && framesSinceStateChange >= debounceFrames) {
+                if (rawScore < deactivationThreshold &&
+                    framesSinceStateChange >= debounceFrames) {
                     callback("SPEECH_END");
                     resetState();
                     logger.silly("Speech end detected", { rawScore });
@@ -107,7 +108,8 @@ async function createVad(params) {
             }
             else {
                 // If currently not speaking, check if the score is above activationThreshold
-                if (rawScore > activationThreshold && framesSinceStateChange >= debounceFrames) {
+                if (rawScore > activationThreshold &&
+                    framesSinceStateChange >= debounceFrames) {
                     isSpeechActive = true;
                     framesSinceStateChange = 0;
                     callback("SPEECH_START");

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@fonoster/autopilot",
-  "version": "0.8.49",
+  "version": "0.8.50",
   "description": "Voice AI for the Fonoster platform",
   "author": "Pedro Sanders <psanders@fonoster.com>",
   "homepage": "https://github.com/fonoster/fonoster#readme",
@@ -33,11 +33,11 @@
   },
   "dependencies": {
     "@aws-sdk/client-s3": "^3.712.0",
-    "@fonoster/common": "^0.8.49",
+    "@fonoster/common": "^0.8.50",
     "@fonoster/logger": "^0.8.47",
-    "@fonoster/sdk": "^0.8.49",
+    "@fonoster/sdk": "^0.8.50",
     "@fonoster/types": "^0.8.47",
-    "@fonoster/voice": "^0.8.49",
+    "@fonoster/voice": "^0.8.50",
     "@langchain/community": "^0.3.19",
     "@langchain/core": "^0.3.23",
     "@langchain/groq": "^0.1.2",
@@ -55,5 +55,5 @@
   "devDependencies": {
     "typescript": "^5.5.4"
   },
-  "gitHead": "d4aa82f7926c0f451d2f580c59ab812c74a7f579"
+  "gitHead": "d0e373668e8e77295e3847c99a346c4aa1c8d3d7"
 }