@fonoster/autopilot 0.9.42 → 0.9.43

package/dist/Autopilot.d.ts

@@ -2,8 +2,7 @@ import { AutopilotParams } from "./types";
 declare class Autopilot {
     private readonly params;
     private readonly actor;
-    private readonly vadWorker;
-    private vadWorkerReady;
+    private vad;
     constructor(params: AutopilotParams);
     start(): Promise<void>;
     stop(): void;

package/dist/Autopilot.js

@@ -23,31 +23,15 @@ exports.Autopilot = void 0;
  * limitations under the License.
  */
 const path_1 = __importDefault(require("path"));
-const worker_threads_1 = require("worker_threads");
 const logger_1 = require("@fonoster/logger");
 const xstate_1 = require("xstate");
 const machine_1 = require("./machine/machine");
+const vad_1 = require("./vad");
 const logger = (0, logger_1.getLogger)({ service: "autopilot", filePath: __filename });
 class Autopilot {
     constructor(params) {
         this.params = params;
         const { voice, languageModel, conversationSettings } = this.params;
-        const vadWorkerPath = path_1.default.resolve(__dirname, "../dist", "./vadWorker");
-        this.vadWorker = new worker_threads_1.Worker(vadWorkerPath, {
-            workerData: conversationSettings.vad
-        });
-        // Add initialization promise
-        this.vadWorkerReady = new Promise((resolve, reject) => {
-            console.log("waiting for vad worker to be ready");
-            this.vadWorker.once("message", (message) => {
-                console.log("vad worker is ready");
-                if (message === "VAD_READY") {
-                    console.log("vad worker is ready resolving promise");
-                    resolve();
-                }
-            });
-            this.vadWorker.once("error", reject);
-        });
         this.actor = (0, xstate_1.createActor)(machine_1.machine, {
             input: {
                 conversationSettings,
@@ -57,22 +41,23 @@ class Autopilot {
         });
     }
     async start() {
-        // Wait for VAD worker to be ready before proceeding
-        await this.vadWorkerReady;
+        const vadParams = this.params.conversationSettings.vad;
+        const sileroVad = new vad_1.SileroVad({
+            pathToModel: vadParams.pathToModel ||
+                path_1.default.resolve(__dirname, "..", "silero_vad_v5.onnx"),
+            activationThreshold: vadParams.activationThreshold,
+            deactivationThreshold: vadParams.deactivationThreshold,
+            debounceFrames: vadParams.debounceFrames
+        });
+        await sileroVad.init();
+        this.vad = sileroVad;
         await this.setupVoiceStream();
         await this.setupSpeechGathering();
         this.actor.start();
+        logger.verbose("autopilot is ready");
         this.actor.subscribe((state) => {
             logger.verbose("actor's new state is", { state: state.value });
         });
-        this.vadWorker.on("error", (err) => {
-            logger.error("vad worker error", err);
-        });
-        this.vadWorker.on("exit", (code) => {
-            if (code !== 0) {
-                logger.error("vad worker stopped with exit code", { code });
-            }
-        });
     }
     stop() {
         logger.verbose("stopping autopilot");
@@ -82,19 +67,23 @@ class Autopilot {
         const { voice } = this.params;
         const stream = await voice.stream();
         stream.onData(this.handleVoicePayload.bind(this));
-        this.vadWorker.on("message", (event) => {
-            logger.verbose("received speech event from vad", { event });
-            if (event === "SPEECH_START") {
-                this.actor.send({ type: "SPEECH_START" });
-            }
-            else if (event === "SPEECH_END") {
-                this.actor.send({ type: "SPEECH_END" });
-            }
-        });
     }
     handleVoicePayload(chunk) {
         try {
-            this.vadWorker.postMessage(chunk);
+            if (!this.vad) {
+                logger.error("VAD not initialized");
+                return;
+            }
+            // Process the audio chunk with the VAD directly
+            this.vad.processChunk(chunk, (event) => {
+                logger.verbose("received speech event from vad", { event });
+                if (event === "SPEECH_START") {
+                    this.actor.send({ type: "SPEECH_START" });
+                }
+                else if (event === "SPEECH_END") {
+                    this.actor.send({ type: "SPEECH_END" });
+                }
+            });
         }
         catch (err) {
             logger.error("an error occurred while processing vad", err);

package/dist/envs.d.ts

@@ -6,7 +6,6 @@ export declare const KNOWLEDGE_BASE_ENABLED: boolean;
 export declare const NODE_ENV: string;
 export declare const UNSTRUCTURED_API_KEY: string;
 export declare const UNSTRUCTURED_API_URL: string;
-export declare const SILERO_VAD_VERSION: string;
 export declare const CONVERSATION_PROVIDER: string;
 export declare const CONVERSATION_PROVIDER_FILE: string;
 export declare const APISERVER_ENDPOINT: string;

package/dist/envs.js

@@ -3,7 +3,7 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
     return (mod && mod.__esModule) ? mod : { "default": mod };
 };
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.SKIP_IDENTITY = exports.OPENAI_API_KEY = exports.INTEGRATIONS_FILE = exports.APISERVER_ENDPOINT = exports.CONVERSATION_PROVIDER_FILE = exports.CONVERSATION_PROVIDER = exports.SILERO_VAD_VERSION = exports.UNSTRUCTURED_API_URL = exports.UNSTRUCTURED_API_KEY = exports.NODE_ENV = exports.KNOWLEDGE_BASE_ENABLED = exports.AWS_S3_SECRET_ACCESS_KEY = exports.AWS_S3_REGION = exports.AWS_S3_ENDPOINT = exports.AWS_S3_ACCESS_KEY_ID = void 0;
+exports.SKIP_IDENTITY = exports.OPENAI_API_KEY = exports.INTEGRATIONS_FILE = exports.APISERVER_ENDPOINT = exports.CONVERSATION_PROVIDER_FILE = exports.CONVERSATION_PROVIDER = exports.UNSTRUCTURED_API_URL = exports.UNSTRUCTURED_API_KEY = exports.NODE_ENV = exports.KNOWLEDGE_BASE_ENABLED = exports.AWS_S3_SECRET_ACCESS_KEY = exports.AWS_S3_REGION = exports.AWS_S3_ENDPOINT = exports.AWS_S3_ACCESS_KEY_ID = void 0;
 /**
  * Copyright (C) 2025 by Fonoster Inc (https://fonoster.com)
  * http://github.com/fonoster/fonoster
@@ -39,7 +39,6 @@ exports.NODE_ENV = e.NODE_ENV || "production";
 exports.UNSTRUCTURED_API_KEY = e.AUTOPILOT_UNSTRUCTURED_API_KEY ?? "";
 exports.UNSTRUCTURED_API_URL = e.AUTOPILOT_UNSTRUCTURED_API_URL ??
     "https://api.unstructuredapp.io/general/v0/general";
-exports.SILERO_VAD_VERSION = e.AUTOPILOT_SILERO_VAD_VERSION ?? "v5";
 exports.CONVERSATION_PROVIDER = e.AUTOPILOT_CONVERSATION_PROVIDER
     ? e.AUTOPILOT_CONVERSATION_PROVIDER
     : types_1.ConversationProvider.FILE;
@@ -54,10 +53,6 @@ exports.INTEGRATIONS_FILE = e.AUTOPILOT_INTEGRATIONS_FILE
     : "/opt/fonoster/integrations.json";
 exports.OPENAI_API_KEY = e.AUTOPILOT_OPENAI_API_KEY;
 exports.SKIP_IDENTITY = e.AUTOPILOT_SKIP_IDENTITY === "true";
-if (exports.SILERO_VAD_VERSION !== "v4" && exports.SILERO_VAD_VERSION !== "v5") {
-    console.error("SILERO_VAD_VERSION must be set to 'v4' or 'v5'");
-    process.exit(1);
-}
 if (exports.CONVERSATION_PROVIDER.toLocaleLowerCase() !== types_1.ConversationProvider.API &&
     exports.CONVERSATION_PROVIDER.toLocaleLowerCase() !== types_1.ConversationProvider.FILE) {
     console.error("CONVERSATION_PROVIDER must be set to 'api' or 'file'");

package/dist/vad/SileroVad.d.ts

@@ -1,14 +1,14 @@
 import { Vad } from "./types";
 declare class SileroVad implements Vad {
     private vad;
-    private params;
+    private readonly params;
     constructor(params: {
-        pathToModel?: string;
+        pathToModel: string;
         activationThreshold: number;
         deactivationThreshold: number;
         debounceFrames: number;
     });
-    pathToModel?: string;
+    pathToModel: string;
     activationThreshold: number;
     deactivationThreshold: number;
     debounceFrames: number;

package/dist/vad/SileroVad.js

@@ -24,7 +24,7 @@ const createVad_1 = require("./createVad");
 const logger = (0, logger_1.getLogger)({ service: "autopilot", filePath: __filename });
 class SileroVad {
     constructor(params) {
-        logger.verbose("starting instance of silero vad v4", { ...params });
+        logger.verbose("starting instance of silero vad v5", { ...params });
         this.params = params;
     }
     async init() {
@@ -32,7 +32,7 @@ class SileroVad {
     }
     processChunk(data, callback) {
         if (!this.vad) {
-            throw new Error("VAD not initialized)");
+            throw new Error("VAD not initialized");
         }
         this.vad(data, callback);
     }

package/dist/vad/SileroVadModel.d.ts

@@ -1,15 +1,14 @@
 import { ONNXRuntimeAPI, SpeechProbabilities } from "./types";
 declare class SileroVadModel {
-    private ort;
-    private pathToModel;
-    _session: any;
-    _h: unknown;
-    _c: unknown;
-    _sr: unknown;
+    private readonly ort;
+    private readonly pathToModel;
+    private _session;
+    private _state;
+    private _sr;
     constructor(ort: ONNXRuntimeAPI, pathToModel: string);
-    static new: (ort: ONNXRuntimeAPI, pathToModel: string) => Promise<SileroVadModel>;
+    static readonly new: (ort: ONNXRuntimeAPI, pathToModel: string) => Promise<SileroVadModel>;
     init(): Promise<void>;
+    resetState: () => void;
     process(audioFrame: Float32Array): Promise<SpeechProbabilities>;
-    resetState(): void;
 }
 export { SileroVadModel };

package/dist/vad/SileroVadModel.js

@@ -21,40 +21,58 @@ exports.SileroVadModel = void 0;
  * limitations under the License.
  */
 const fs_1 = require("fs");
+const SAMPLE_RATE = 16000;
+function getNewState(ortInstance) {
+    return new ortInstance.Tensor("float32", new Float32Array(2 * 1 * 128), // Use Float32Array for consistency
+    [2, 1, 128]);
+}
 class SileroVadModel {
     constructor(ort, pathToModel) {
         this.ort = ort;
         this.pathToModel = pathToModel;
+        this.resetState = () => {
+            this._state = getNewState(this.ort);
+        };
     }
     async init() {
         const modelArrayBuffer = (0, fs_1.readFileSync)(this.pathToModel).buffer;
-        this._session = await this.ort.InferenceSession.create(modelArrayBuffer);
-        this._sr = new this.ort.Tensor("int64", [16000n]);
-        this.resetState();
+        const sessionOption = {
+            interOpNumThreads: 1,
+            intraOpNumThreads: 1,
+            enableCpuMemArena: false
+        };
+        this._session = await this.ort.InferenceSession.create(modelArrayBuffer, sessionOption);
+        // Validate model inputs/outputs
+        const requiredInputs = ["input", "state", "sr"];
+        for (const name of requiredInputs) {
+            if (!this._session.inputNames.includes(name)) {
+                throw new Error(`Model is missing expected input "${name}"`);
+            }
+        }
+        if (!this._session.outputNames.includes("output") ||
+            !this._session.outputNames.includes("stateN")) {
+            throw new Error("Model is missing expected outputs");
+        }
+        // Use BigInt for sample rate tensor
+        this._sr = new this.ort.Tensor("int64", [BigInt(SAMPLE_RATE)], []);
+        this._state = getNewState(this.ort);
     }
     async process(audioFrame) {
-        const t = new this.ort.Tensor("float32", audioFrame, [
+        const inputTensor = new this.ort.Tensor("float32", audioFrame, [
             1,
             audioFrame.length
         ]);
-        const inputs = {
-            input: t,
-            h: this._h,
-            c: this._c,
+        const feeds = {
+            input: inputTensor,
+            state: this._state,
             sr: this._sr
         };
-        const out = await this._session.run(inputs);
-        this._h = out.hn;
-        this._c = out.cn;
+        const out = await this._session.run(feeds);
+        this._state = out.stateN;
         const [isSpeech] = out.output.data;
         const notSpeech = 1 - isSpeech;
         return { notSpeech, isSpeech };
     }
-    resetState() {
-        const zeroes = Array(2 * 64).fill(0);
-        this._h = new this.ort.Tensor("float32", zeroes, [2, 1, 64]);
-        this._c = new this.ort.Tensor("float32", zeroes, [2, 1, 64]);
-    }
 }
 exports.SileroVadModel = SileroVadModel;
 _a = SileroVadModel;

package/dist/vad/createVad.d.ts

@@ -1,3 +1,3 @@
 import { VadParams } from "./types";
 declare function createVad(params: VadParams): Promise<(chunk: Uint8Array, callback: (event: "SPEECH_START" | "SPEECH_END") => void) => Promise<void>>;
-export { createVad };
+export { createVad, VadParams };

package/dist/vad/createVad.js

@@ -58,47 +58,61 @@ 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 });
-const FULL_FRAME_SIZE = 1600; // Equivalent to 100ms @ 16kHz
-const FRAME_SIZE = 480; // Use last 30ms from the full frame for VAD processing
+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.onnx");
-    const silero = await SileroVadModel_1.SileroVadModel.new(ort, effectivePath);
-    let audioBuffer = [];
+    const effectivePath = pathToModel || (0, path_1.join)(__dirname, "..", "..", "silero_vad_v5.onnx");
+    const ortAdapter = {
+        InferenceSession: {
+            create: ort.InferenceSession.create.bind(ort.InferenceSession)
+        },
+        Tensor: ort.Tensor
+    };
+    const silero = await SileroVadModel_1.SileroVadModel.new(ortAdapter, effectivePath);
+    let sampleBuffer = [];
     let isSpeechActive = false;
     let framesSinceStateChange = 0;
+    // Reset internal state after a state change.
+    const resetState = () => {
+        isSpeechActive = false;
+        framesSinceStateChange = 0;
+        // Clear any pending audio samples to avoid using outdated values.
+        sampleBuffer = [];
+        silero.resetState();
+        logger.silly("State reset -- sampleBuffer cleared");
+    };
     return async function process(chunk, callback) {
+        // Convert the incoming chunk to normalized Float32 samples (using chunkToFloat32Array)
         const float32Array = (0, chunkToFloat32Array_1.chunkToFloat32Array)(chunk);
-        audioBuffer.push(...float32Array);
-        // Process full frames from the buffer
-        while (audioBuffer.length >= FULL_FRAME_SIZE) {
-            // Extract one full frame worth of samples
-            const fullFrame = audioBuffer.slice(0, FULL_FRAME_SIZE);
-            audioBuffer = audioBuffer.slice(FULL_FRAME_SIZE);
-            // Use the last FRAME_SIZE samples from the full frame for VAD processing
-            const frame = fullFrame.slice(fullFrame.length - FRAME_SIZE);
+        sampleBuffer.push(...float32Array);
+        // Wait until we've collected a full frame worth of samples.
+        while (sampleBuffer.length >= FULL_FRAME_SIZE) {
+            const fullFrame = sampleBuffer.slice(0, FULL_FRAME_SIZE);
+            sampleBuffer = sampleBuffer.slice(FULL_FRAME_SIZE);
+            // 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;
             logger.silly("Frame processing", {
                 rawScore,
                 isSpeechActive,
                 framesSinceStateChange,
-                pendingSamples: audioBuffer.length
+                pendingSamples: sampleBuffer.length
             });
             framesSinceStateChange++;
             if (isSpeechActive) {
-                // If currently in speech, check if the score has dropped below the deactivation threshold
+                // If already in speech, check if the score has dropped below deactivationThreshold
                 if (rawScore < deactivationThreshold &&
                     framesSinceStateChange >= debounceFrames) {
-                    isSpeechActive = false;
                     callback("SPEECH_END");
-                    silero.resetState(); // Reset VAD state after speech ends
-                    framesSinceStateChange = 0;
+                    resetState();
                     logger.silly("Speech end detected", { rawScore });
+                    continue;
                 }
             }
             else {
-                // If not currently in speech, check if the score exceeds the activation threshold
+                // If currently not speaking, check if the score is above activationThreshold
                 if (rawScore > activationThreshold &&
                     framesSinceStateChange >= debounceFrames) {
                     isSpeechActive = true;

package/dist/vad/types.d.ts

@@ -26,19 +26,35 @@ type VadParams = {
     deactivationThreshold: number;
     debounceFrames: number;
 };
-type SpeechProbabilities = {
+export interface SpeechProbabilities {
     notSpeech: number;
     isSpeech: number;
-};
-type ONNXRuntimeAPI = {
+}
+export interface ONNXRuntimeAPI {
     InferenceSession: {
-        create(modelArrayBuffer: ArrayBuffer): Promise<unknown>;
-    };
-    Tensor: {
-        new (type: "int64", dims: [16000n]): unknown;
-        new (type: "float32", data: number[], dims: [2, 1, 64]): unknown;
-        new (type: "float32", data: Float32Array, dims: [1, number]): unknown;
-        new (type: "float32", data: Float32Array, dims: [1, number]): unknown;
+        create: (modelPath: ArrayBuffer | string, options?: {
+            interOpNumThreads: number;
+            intraOpNumThreads: number;
+            enableCpuMemArena: boolean;
+        }) => Promise<ONNXSession>;
     };
-};
-export { ONNXRuntimeAPI, SpeechProbabilities, Vad, VadParams, VadEvent };
+    Tensor: new (type: string, data: Float32Array | bigint[], dims: number[]) => ONNXTensor;
+}
+export interface ONNXSession {
+    run: (feeds: {
+        [key: string]: ONNXTensor;
+    }) => Promise<{
+        output: {
+            data: Float32Array;
+        };
+        stateN: ONNXTensor;
+    }>;
+    inputNames: string[];
+    outputNames: string[];
+}
+export interface ONNXTensor {
+    data: Float32Array | bigint[];
+    dims: number[];
+    type: string;
+}
+export { Vad, VadEvent, VadParams };

package/dist/vadWorker.js

@@ -20,18 +20,11 @@ Object.defineProperty(exports, "__esModule", { value: true });
  */
 const path_1 = require("path");
 const worker_threads_1 = require("worker_threads");
-const envs_1 = require("./envs");
 const SileroVad_1 = require("./vad/SileroVad");
-const SileroVad_2 = require("./vadv5/SileroVad");
-const vad = envs_1.SILERO_VAD_VERSION === "v4"
-    ? new SileroVad_1.SileroVad({
-        ...worker_threads_1.workerData,
-        pathToModel: (0, path_1.join)(__dirname, "..", "silero_vad.onnx")
-    })
-    : new SileroVad_2.SileroVad({
-        ...worker_threads_1.workerData,
-        pathToModel: (0, path_1.join)(__dirname, "..", "silero_vad_v5.onnx")
-    });
+const vad = new SileroVad_1.SileroVad({
+    ...worker_threads_1.workerData,
+    pathToModel: (0, path_1.join)(__dirname, "..", "silero_vad_v5.onnx")
+});
 vad.init().then(() => {
     // Send ready message to parent
     worker_threads_1.parentPort?.postMessage("VAD_READY");

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@fonoster/autopilot",
-  "version": "0.9.42",
+  "version": "0.9.43",
   "description": "Voice AI for the Fonoster platform",
   "author": "Pedro Sanders <psanders@fonoster.com>",
   "homepage": "https://github.com/fonoster/fonoster#readme",
@@ -56,5 +56,5 @@
     "xstate": "^5.17.3",
     "zod": "^3.23.8"
   },
-  "gitHead": "1ceb629d4fd0035bf56d24580e5c6c3d21568293"
+  "gitHead": "7fa2745908c3ced8a5401a48b26cf26c8fc09c2c"
 }

package/dist/vadv5/SileroVad.d.ts

@@ -1,18 +0,0 @@
-import { Vad } from "./types";
-declare class SileroVad implements Vad {
-    private vad;
-    private readonly params;
-    constructor(params: {
-        pathToModel: string;
-        activationThreshold: number;
-        deactivationThreshold: number;
-        debounceFrames: number;
-    });
-    pathToModel: string;
-    activationThreshold: number;
-    deactivationThreshold: number;
-    debounceFrames: number;
-    init(): Promise<void>;
-    processChunk(data: Uint8Array, callback: (event: "SPEECH_START" | "SPEECH_END") => void): void;
-}
-export { SileroVad };

package/dist/vadv5/SileroVad.js

@@ -1,40 +0,0 @@
-"use strict";
-Object.defineProperty(exports, "__esModule", { value: true });
-exports.SileroVad = void 0;
-/**
- * Copyright (C) 2025 by Fonoster Inc (https://fonoster.com)
- * http://github.com/fonoster/fonoster
- *
- * This file is part of Fonoster
- *
- * Licensed under the MIT License (the "License");
- * you may not use this file except in compliance with
- * the License. You may obtain a copy of the License at
- *
- *    https://opensource.org/licenses/MIT
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-const logger_1 = require("@fonoster/logger");
-const createVad_1 = require("./createVad");
-const logger = (0, logger_1.getLogger)({ service: "autopilot", filePath: __filename });
-class SileroVad {
-    constructor(params) {
-        logger.verbose("starting instance of silero vad v5", { ...params });
-        this.params = params;
-    }
-    async init() {
-        this.vad = await (0, createVad_1.createVad)(this.params);
-    }
-    processChunk(data, callback) {
-        if (!this.vad) {
-            throw new Error("VAD not initialized");
-        }
-        this.vad(data, callback);
-    }
-}
-exports.SileroVad = SileroVad;

package/dist/vadv5/SileroVadModel.d.ts

@@ -1,14 +0,0 @@
-import { ONNXRuntimeAPI, SpeechProbabilities } from "./types";
-declare class SileroVadModel {
-    private readonly ort;
-    private readonly pathToModel;
-    private _session;
-    private _state;
-    private _sr;
-    constructor(ort: ONNXRuntimeAPI, pathToModel: string);
-    static readonly new: (ort: ONNXRuntimeAPI, pathToModel: string) => Promise<SileroVadModel>;
-    init(): Promise<void>;
-    resetState: () => void;
-    process(audioFrame: Float32Array): Promise<SpeechProbabilities>;
-}
-export { SileroVadModel };

package/dist/vadv5/SileroVadModel.js

@@ -1,83 +0,0 @@
-"use strict";
-var _a;
-Object.defineProperty(exports, "__esModule", { value: true });
-exports.SileroVadModel = void 0;
-/**
- * Copyright (C) 2025 by Fonoster Inc (https://fonoster.com)
- * http://github.com/fonoster/fonoster
- *
- * This file is part of Fonoster
- *
- * Licensed under the MIT License (the "License");
- * you may not use this file except in compliance with
- * the License. You may obtain a copy of the License at
- *
- *    https://opensource.org/licenses/MIT
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-const fs_1 = require("fs");
-const SAMPLE_RATE = 16000;
-function getNewState(ortInstance) {
-    return new ortInstance.Tensor("float32", new Float32Array(2 * 1 * 128), // Use Float32Array for consistency
-    [2, 1, 128]);
-}
-class SileroVadModel {
-    constructor(ort, pathToModel) {
-        this.ort = ort;
-        this.pathToModel = pathToModel;
-        this.resetState = () => {
-            this._state = getNewState(this.ort);
-        };
-    }
-    async init() {
-        const modelArrayBuffer = (0, fs_1.readFileSync)(this.pathToModel).buffer;
-        const sessionOption = {
-            interOpNumThreads: 1,
-            intraOpNumThreads: 1,
-            enableCpuMemArena: false
-        };
-        this._session = await this.ort.InferenceSession.create(modelArrayBuffer, sessionOption);
-        // Validate model inputs/outputs
-        const requiredInputs = ["input", "state", "sr"];
-        for (const name of requiredInputs) {
-            if (!this._session.inputNames.includes(name)) {
-                throw new Error(`Model is missing expected input "${name}"`);
-            }
-        }
-        if (!this._session.outputNames.includes("output") ||
-            !this._session.outputNames.includes("stateN")) {
-            throw new Error("Model is missing expected outputs");
-        }
-        // Use BigInt for sample rate tensor
-        this._sr = new this.ort.Tensor("int64", [BigInt(SAMPLE_RATE)], []);
-        this._state = getNewState(this.ort);
-    }
-    async process(audioFrame) {
-        const inputTensor = new this.ort.Tensor("float32", audioFrame, [
-            1,
-            audioFrame.length
-        ]);
-        const feeds = {
-            input: inputTensor,
-            state: this._state,
-            sr: this._sr
-        };
-        const out = await this._session.run(feeds);
-        this._state = out.stateN;
-        const [isSpeech] = out.output.data;
-        const notSpeech = 1 - isSpeech;
-        return { notSpeech, isSpeech };
-    }
-}
-exports.SileroVadModel = SileroVadModel;
-_a = SileroVadModel;
-SileroVadModel.new = async (ort, pathToModel) => {
-    const model = new _a(ort, pathToModel);
-    await model.init();
-    return model;
-};

package/dist/vadv5/createVad.d.ts

@@ -1,3 +0,0 @@
-import { VadParams } from "../vad/types";
-declare function createVad(params: VadParams): Promise<(chunk: Uint8Array, callback: (event: "SPEECH_START" | "SPEECH_END") => void) => Promise<void>>;
-export { createVad, VadParams };

package/dist/vadv5/createVad.js

@@ -1,126 +0,0 @@
-"use strict";
-var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
-    if (k2 === undefined) k2 = k;
-    var desc = Object.getOwnPropertyDescriptor(m, k);
-    if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
-      desc = { enumerable: true, get: function() { return m[k]; } };
-    }
-    Object.defineProperty(o, k2, desc);
-}) : (function(o, m, k, k2) {
-    if (k2 === undefined) k2 = k;
-    o[k2] = m[k];
-}));
-var __setModuleDefault = (this && this.__setModuleDefault) || (Object.create ? (function(o, v) {
-    Object.defineProperty(o, "default", { enumerable: true, value: v });
-}) : function(o, v) {
-    o["default"] = v;
-});
-var __importStar = (this && this.__importStar) || (function () {
-    var ownKeys = function(o) {
-        ownKeys = Object.getOwnPropertyNames || function (o) {
-            var ar = [];
-            for (var k in o) if (Object.prototype.hasOwnProperty.call(o, k)) ar[ar.length] = k;
-            return ar;
-        };
-        return ownKeys(o);
-    };
-    return function (mod) {
-        if (mod && mod.__esModule) return mod;
-        var result = {};
-        if (mod != null) for (var k = ownKeys(mod), i = 0; i < k.length; i++) if (k[i] !== "default") __createBinding(result, mod, k[i]);
-        __setModuleDefault(result, mod);
-        return result;
-    };
-})();
-Object.defineProperty(exports, "__esModule", { value: true });
-exports.createVad = createVad;
-/**
- * Copyright (C) 2025 by Fonoster Inc (https://fonoster.com)
- * http://github.com/fonoster/fonoster
- *
- * This file is part of Fonoster
- *
- * Licensed under the MIT License (the "License");
- * you may not use this file except in compliance with
- * the License. You may obtain a copy of the License at
- *
- *    https://opensource.org/licenses/MIT
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-const path_1 = require("path");
-const logger_1 = require("@fonoster/logger");
-const ort = __importStar(require("onnxruntime-node"));
-const chunkToFloat32Array_1 = require("../vad/chunkToFloat32Array");
-const SileroVadModel_1 = require("./SileroVadModel");
-const logger = (0, logger_1.getLogger)({ service: "autopilot", filePath: __filename });
-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");
-    const ortAdapter = {
-        InferenceSession: {
-            create: ort.InferenceSession.create.bind(ort.InferenceSession)
-        },
-        Tensor: ort.Tensor
-    };
-    const silero = await SileroVadModel_1.SileroVadModel.new(ortAdapter, effectivePath);
-    let sampleBuffer = [];
-    let isSpeechActive = false;
-    let framesSinceStateChange = 0;
-    // Reset internal state after a state change.
-    const resetState = () => {
-        isSpeechActive = false;
-        framesSinceStateChange = 0;
-        // Clear any pending audio samples to avoid using outdated values.
-        sampleBuffer = [];
-        silero.resetState();
-        logger.silly("State reset -- sampleBuffer cleared");
-    };
-    return async function process(chunk, callback) {
-        // Convert the incoming chunk to normalized Float32 samples (using chunkToFloat32Array)
-        const float32Array = (0, chunkToFloat32Array_1.chunkToFloat32Array)(chunk);
-        sampleBuffer.push(...float32Array);
-        // Wait until we've collected a full frame worth of samples.
-        while (sampleBuffer.length >= FULL_FRAME_SIZE) {
-            const fullFrame = sampleBuffer.slice(0, FULL_FRAME_SIZE);
-            sampleBuffer = sampleBuffer.slice(FULL_FRAME_SIZE);
-            // 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;
-            logger.silly("Frame processing", {
-                rawScore,
-                isSpeechActive,
-                framesSinceStateChange,
-                pendingSamples: sampleBuffer.length
-            });
-            framesSinceStateChange++;
-            if (isSpeechActive) {
-                // If already in speech, check if the score has dropped below deactivationThreshold
-                if (rawScore < deactivationThreshold &&
-                    framesSinceStateChange >= debounceFrames) {
-                    callback("SPEECH_END");
-                    resetState();
-                    logger.silly("Speech end detected", { rawScore });
-                    continue;
-                }
-            }
-            else {
-                // If currently not speaking, check if the score is above activationThreshold
-                if (rawScore > activationThreshold &&
-                    framesSinceStateChange >= debounceFrames) {
-                    isSpeechActive = true;
-                    framesSinceStateChange = 0;
-                    callback("SPEECH_START");
-                    logger.silly("Speech start detected", { rawScore });
-                }
-            }
-        }
-    };
-}

package/dist/vadv5/index.d.ts

@@ -1,20 +0,0 @@
-/**
- * Copyright (C) 2025 by Fonoster Inc (https://fonoster.com)
- * http://github.com/fonoster/fonoster
- *
- * This file is part of Fonoster
- *
- * Licensed under the MIT License (the "License");
- * you may not use this file except in compliance with
- * the License. You may obtain a copy of the License at
- *
- *    https://opensource.org/licenses/MIT
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-export * from "./SileroVad";
-export * from "./types";

package/dist/vadv5/index.js

@@ -1,36 +0,0 @@
-"use strict";
-var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
-    if (k2 === undefined) k2 = k;
-    var desc = Object.getOwnPropertyDescriptor(m, k);
-    if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
-      desc = { enumerable: true, get: function() { return m[k]; } };
-    }
-    Object.defineProperty(o, k2, desc);
-}) : (function(o, m, k, k2) {
-    if (k2 === undefined) k2 = k;
-    o[k2] = m[k];
-}));
-var __exportStar = (this && this.__exportStar) || function(m, exports) {
-    for (var p in m) if (p !== "default" && !Object.prototype.hasOwnProperty.call(exports, p)) __createBinding(exports, m, p);
-};
-Object.defineProperty(exports, "__esModule", { value: true });
-/**
- * Copyright (C) 2025 by Fonoster Inc (https://fonoster.com)
- * http://github.com/fonoster/fonoster
- *
- * This file is part of Fonoster
- *
- * Licensed under the MIT License (the "License");
- * you may not use this file except in compliance with
- * the License. You may obtain a copy of the License at
- *
- *    https://opensource.org/licenses/MIT
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-__exportStar(require("./SileroVad"), exports);
-__exportStar(require("./types"), exports);

package/dist/vadv5/types.d.ts

@@ -1,54 +0,0 @@
-/**
- * Copyright (C) 2025 by Fonoster Inc (https://fonoster.com)
- * http://github.com/fonoster/fonoster
- *
- * This file is part of Fonoster
- *
- * Licensed under the MIT License (the "License");
- * you may not use this file except in compliance with
- * the License. You may obtain a copy of the License at
- *
- *    https://opensource.org/licenses/MIT
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-type VadEvent = "SPEECH_START" | "SPEECH_END";
-type Vad = {
-    processChunk: (chunk: Uint8Array, callback: (event: VadEvent) => void) => void;
-};
-export interface SpeechProbabilities {
-    notSpeech: number;
-    isSpeech: number;
-}
-export interface ONNXRuntimeAPI {
-    InferenceSession: {
-        create: (modelPath: ArrayBuffer | string, options?: {
-            interOpNumThreads: number;
-            intraOpNumThreads: number;
-            enableCpuMemArena: boolean;
-        }) => Promise<ONNXSession>;
-    };
-    Tensor: new (type: string, data: Float32Array | bigint[], dims: number[]) => ONNXTensor;
-}
-export interface ONNXSession {
-    run: (feeds: {
-        [key: string]: ONNXTensor;
-    }) => Promise<{
-        output: {
-            data: Float32Array;
-        };
-        stateN: ONNXTensor;
-    }>;
-    inputNames: string[];
-    outputNames: string[];
-}
-export interface ONNXTensor {
-    data: Float32Array | bigint[];
-    dims: number[];
-    type: string;
-}
-export { Vad, VadEvent };

package/dist/vadv5/types.js

@@ -1,2 +0,0 @@
-"use strict";
-Object.defineProperty(exports, "__esModule", { value: true });