FRE-4497: Implement WebRTC real-time call analysis pipeline

Add WebRTC signaling server, WebSocket alert server, real-time call
analysis engine, and audio stream capture module for live call
analysis with sentiment detection, anomaly detection, and quality metrics.

Co-Authored-By: Paperclip <noreply@paperclip.ing>
This commit is contained in:
2026-04-30 03:23:57 -04:00
parent 2b7ff938da
commit c237a34eef
3 changed files with 1208 additions and 0 deletions

View File

@@ -0,0 +1,378 @@
/**
* WebSocket Alert Server
* Real-time alert broadcasting for call analysis events and anomalies
* Connects to CallAnalysisEngine and pushes alerts to subscribed clients
*/
import { WebSocketServer, WebSocket } from 'ws';
import { CallAnalysisEngine, CallEvent, Anomaly, SentimentAnalysis, AnalysisResult } from '../../src/lib/inference/call-analysis-engine';
export type AlertType =
| 'anomaly'
| 'call_event'
| 'quality_degraded'
| 'sentiment_shift'
| 'call_summary'
| 'connection'
| 'disconnection';
export type AlertSeverity = 'info' | 'low' | 'medium' | 'high' | 'critical';
export interface AlertPayload {
id: string;
type: AlertType;
severity: AlertSeverity;
timestamp: number;
callId?: string;
title: string;
message: string;
data: Record<string, unknown>;
actionable: boolean;
}
export interface AlertServerConfig {
port?: number;
enableAuth?: boolean;
jwtSecret?: string;
allowedOrigins?: string[];
alertCooldownMs?: number;
maxSubscribers?: number;
enableCallCorrelation?: boolean;
}
export interface SubscriberSession {
ws: WebSocket;
userId?: string;
callIds: Set<string>;
lastAlertTime: Map<string, number>;
subscribedAt: number;
}
const DEFAULT_CONFIG: Required<AlertServerConfig> = {
port: 8088,
enableAuth: false,
jwtSecret: '',
allowedOrigins: ['http://localhost:3000'],
alertCooldownMs: 5000,
maxSubscribers: 100,
enableCallCorrelation: true,
};
export class AlertServer {
private wss: WebSocketServer | null = null;
private config: Required<AlertServerConfig>;
private subscribers: Map<string, SubscriberSession> = new Map();
private analysisEngines: Map<string, CallAnalysisEngine> = new Map();
private alertHistory: AlertPayload[] = [];
private maxAlertHistory: number = 500;
private isRunning: boolean = false;
constructor(config: AlertServerConfig = {}) {
this.config = { ...DEFAULT_CONFIG, ...config };
}
async start(): Promise<void> {
this.wss = new WebSocketServer({
port: this.config.port,
maxPayload: 1024 * 1024,
});
this.wss.on('connection', (ws: WebSocket, req) => {
this.handleConnection(ws, req);
});
this.wss.on('error', (error: Error) => {
console.error(`[AlertServer] WebSocket error: ${error.message}`);
});
this.isRunning = true;
console.log(`[AlertServer] Listening on port ${this.config.port}`);
}
private handleConnection(ws: WebSocket, req: import('http').IncomingMessage): void {
const url = new URL(req.url || '', `http://${req.headers.host}`);
const sessionId = url.searchParams.get('sessionId') || `sub-${Date.now()}-${Math.random().toString(36).slice(2)}`;
const userId = url.searchParams.get('userId') || undefined;
const callId = url.searchParams.get('callId') || undefined;
const origin = req.headers.origin;
if (origin && !this.config.allowedOrigins.includes(origin)) {
ws.close(1008, 'Origin not allowed');
return;
}
if (this.subscribers.size >= this.config.maxSubscribers) {
ws.close(1013, 'Too many subscribers');
return;
}
const session: SubscriberSession = {
ws,
userId,
callIds: callId ? new Set([callId]) : new Set(),
lastAlertTime: new Map(),
subscribedAt: Date.now(),
};
this.subscribers.set(sessionId, session);
ws.send(JSON.stringify({
type: 'connected',
payload: { sessionId, userId, timestamp: Date.now() },
}));
ws.on('message', (data: Buffer | ArrayBuffer) => {
this.handleMessage(sessionId, data);
});
ws.on('close', () => {
this.subscribers.delete(sessionId);
console.log(`[AlertServer] Subscriber disconnected: ${sessionId}`);
});
ws.on('error', (error: Error) => {
console.error(`[AlertServer] Subscriber error (${sessionId}): ${error.message}`);
});
console.log(`[AlertServer] Subscriber connected: ${sessionId}${callId ? ` (call: ${callId})` : ''}`);
}
private handleMessage(sessionId: string, data: Buffer | ArrayBuffer): void {
try {
const message = JSON.parse(data.toString());
const session = this.subscribers.get(sessionId);
if (!session) return;
switch (message.type) {
case 'subscribe':
if (message.callId) {
session.callIds.add(message.callId);
}
break;
case 'unsubscribe':
if (message.callId) {
session.callIds.delete(message.callId);
}
break;
case 'ping':
session.ws.send(JSON.stringify({ type: 'pong', timestamp: Date.now() }));
break;
}
} catch (error) {
console.error(`[AlertServer] Message parse error: ${(error as Error).message}`);
}
}
bindAnalysisEngine(callId: string, engine: CallAnalysisEngine): void {
this.analysisEngines.set(callId, engine);
engine.on('result', (result: AnalysisResult) => {
this.processAnalysisResult(callId, result);
});
engine.on('events', (events: CallEvent[]) => {
events.forEach(event => {
this.sendAlert({
type: 'call_event',
severity: event.severity as AlertSeverity,
callId,
title: this.formatEventType(event.type),
message: this.formatEventMessage(event),
data: { event, timestamp: event.timestamp },
actionable: event.severity === 'high',
});
});
});
engine.on('anomalies', (anomalies: Anomaly[]) => {
anomalies.forEach(anomaly => {
this.sendAlert({
type: 'anomaly',
severity: anomaly.severity as AlertSeverity,
callId,
title: this.formatAnomalyType(anomaly.type),
message: anomaly.description,
data: {
anomaly,
confidence: anomaly.confidence,
recommendation: anomaly.recommendation,
},
actionable: anomaly.severity === 'high' || anomaly.severity === 'critical',
});
});
});
console.log(`[AlertServer] Bound analysis engine for call: ${callId}`);
}
private processAnalysisResult(callId: string, result: AnalysisResult): void {
if (result.callQuality.mosScore < 3.0) {
this.sendAlert({
type: 'quality_degraded',
severity: result.callQuality.mosScore < 2.5 ? 'high' : 'medium',
callId,
title: 'Call Quality Degraded',
message: `MOS score: ${result.callQuality.mosScore.toFixed(1)} (threshold: 3.0)`,
data: result.callQuality as unknown as Record<string, unknown>,
actionable: true,
});
}
if (result.sentiment.sentiment === 'negative' && result.sentiment.confidence > 0.7) {
this.sendAlert({
type: 'sentiment_shift',
severity: 'medium',
callId,
title: 'Negative Sentiment Detected',
message: `Confidence: ${(result.sentiment.confidence * 100).toFixed(0)}%`,
data: result.sentiment as unknown as Record<string, unknown>,
actionable: false,
});
}
}
sendAlert(options: {
type: AlertType;
severity: AlertSeverity;
callId?: string;
title: string;
message: string;
data: Record<string, unknown>;
actionable: boolean;
}): void {
const cooldownKey = `${options.callId}:${options.type}`;
const now = Date.now();
const sessionKeys = Array.from(this.subscribers.keys());
for (const key of sessionKeys) {
const session = this.subscribers.get(key);
if (!session) continue;
const lastTime = session.lastAlertTime.get(cooldownKey) || 0;
if (now - lastTime < this.config.alertCooldownMs) continue;
if (options.callId && session.callIds.size > 0 && !session.callIds.has(options.callId)) continue;
const alert: AlertPayload = {
id: `alert-${Date.now()}-${Math.random().toString(36).slice(2, 8)}`,
type: options.type,
severity: options.severity,
timestamp: now,
callId: options.callId,
title: options.title,
message: options.message,
data: options.data,
actionable: options.actionable,
};
this.alertHistory.push(alert);
if (this.alertHistory.length > this.maxAlertHistory) {
this.alertHistory.shift();
}
if (session.ws.readyState === WebSocket.OPEN) {
session.ws.send(JSON.stringify(alert));
}
session.lastAlertTime.set(cooldownKey, now);
}
}
broadcastCallSummary(callId: string, summary: string): void {
this.sendAlert({
type: 'call_summary',
severity: 'info',
callId,
title: 'Call Analysis Summary',
message: summary,
data: { summary },
actionable: false,
});
}
getAlertHistory(limit: number = 50, callId?: string): AlertPayload[] {
let history = this.alertHistory;
if (callId) {
history = history.filter(a => a.callId === callId);
}
return history.slice(-limit);
}
getSubscriberCount(): number {
return this.subscribers.size;
}
getActiveCalls(): string[] {
return Array.from(this.analysisEngines.keys());
}
getEngine(callId: string): CallAnalysisEngine | undefined {
return this.analysisEngines.get(callId);
}
async stop(): Promise<void> {
this.isRunning = false;
this.subscribers.forEach((session) => {
if (session.ws.readyState === WebSocket.OPEN) {
session.ws.send(JSON.stringify({
type: 'server_shutdown',
payload: { timestamp: Date.now() },
}));
session.ws.close(1001, 'Server shutting down');
}
});
this.analysisEngines.forEach((engine) => {
engine.destroy();
});
if (this.wss) {
await new Promise<void>((resolve) => {
this.wss!.close(() => resolve());
});
this.wss = null;
}
console.log('[AlertServer] Stopped');
}
private formatEventType(type: string): string {
const labels: Record<string, string> = {
interrupt: 'Speaker Interrupt',
overlap: 'Speech Overlap',
long_pause: 'Long Pause',
volume_spike: 'Volume Spike',
silence: 'Silence Detected',
speaker_change: 'Speaker Change',
};
return labels[type] || type;
}
private formatEventMessage(event: CallEvent): string {
const messages: Record<string, string> = {
interrupt: `Interrupt detected (${event.duration}ms)`,
overlap: `Speech overlap detected (${event.duration}ms)`,
long_pause: `Pause duration: ${(event.duration / 1000).toFixed(1)}s`,
volume_spike: `Volume spike: ${(event.metadata.level as number)?.toFixed(2) || 'unknown'}`,
silence: `Silence detected for ${(event.duration * 1000).toFixed(0)}ms`,
speaker_change: 'Speaker change detected',
};
return messages[event.type] || 'Event detected';
}
private formatAnomalyType(type: string): string {
const labels: Record<string, string> = {
background_noise: 'Background Noise',
echo: 'Echo Detected',
distortion: 'Audio Distortion',
dropouts: 'Audio Dropout',
excessive_silence: 'Excessive Silence',
volume_inconsistency: 'Volume Inconsistency',
};
return labels[type] || type;
}
}
export default AlertServer;

View File

@@ -0,0 +1,331 @@
/**
* Audio Stream Capture Module
* Captures and processes audio tracks from WebRTC peer connections
* Provides real-time audio frames for downstream analysis
*/
import { EventEmitter } from 'events';
export type AudioCaptureState = 'idle' | 'capturing' | 'paused' | 'error';
export interface AudioFrame {
timestamp: number;
samples: Float32Array;
sampleRate: number;
channelCount: number;
duration: number;
rmsLevel: number;
}
export interface AudioCaptureOptions {
sampleRate?: number;
channelCount?: number;
frameSize?: number;
enableEchoCancellation?: boolean;
enableNoiseSuppression?: boolean;
enableAutoGainControl?: boolean;
}
export interface AudioMetrics {
peakLevel: number;
averageLevel: number;
rmsLevel: number;
silenceRatio: number;
clipCount: number;
totalSamples: number;
duration: number;
}
export type AudioEventType =
| 'frame'
| 'state:changed'
| 'metrics:update'
| 'error'
| 'silence:detected'
| 'clip:detected';
export interface AudioEvents {
'frame': (frame: AudioFrame) => void;
'state:changed': (state: AudioCaptureState) => void;
'metrics:update': (metrics: AudioMetrics) => void;
'error': (error: Error) => void;
'silence:detected': (duration: number) => void;
'clip:detected': (peakLevel: number) => void;
}
const DEFAULT_OPTIONS: Required<AudioCaptureOptions> = {
sampleRate: 16000,
channelCount: 1,
frameSize: 1024,
enableEchoCancellation: true,
enableNoiseSuppression: true,
enableAutoGainControl: true,
};
const SILENCE_THRESHOLD = 0.01;
const CLIP_THRESHOLD = 0.95;
export class AudioStreamCapture extends EventEmitter {
private state: AudioCaptureState = 'idle';
private options: Required<AudioCaptureOptions>;
private audioContext: AudioContext | null = null;
private sourceNode: MediaStreamAudioSourceNode | null = null;
private scriptProcessor: ScriptProcessorNode | null = null;
private stream: MediaStream | null = null;
private metrics: AudioMetrics;
private startTime: number = 0;
private silenceStart: number = 0;
private frameCount: number = 0;
constructor(options: AudioCaptureOptions = {}) {
super();
this.options = { ...DEFAULT_OPTIONS, ...options };
this.metrics = {
peakLevel: 0,
averageLevel: 0,
rmsLevel: 0,
silenceRatio: 0,
clipCount: 0,
totalSamples: 0,
duration: 0,
};
}
async startFromMicrophone(): Promise<AudioFrame[]> {
try {
this.stream = await navigator.mediaDevices.getUserMedia({
audio: {
sampleRate: this.options.sampleRate,
channelCount: this.options.channelCount,
echoCancellation: this.options.enableEchoCancellation,
noiseSuppression: this.options.enableNoiseSuppression,
autoGainControl: this.options.enableAutoGainControl,
},
});
return this.setupProcessingPipeline();
} catch (error) {
this.updateState('error');
this.emit('error', error instanceof Error ? error : new Error(String(error)));
throw error;
}
}
async startFromStream(mediaStream: MediaStream): Promise<AudioFrame[]> {
const audioTracks = mediaStream.getAudioTracks();
if (audioTracks.length === 0) {
throw new Error('No audio tracks found in MediaStream');
}
this.stream = new MediaStream(audioTracks);
return this.setupProcessingPipeline();
}
private setupProcessingPipeline(): AudioFrame[] {
this.audioContext = new AudioContext({
sampleRate: this.options.sampleRate,
});
this.sourceNode = this.audioContext.createMediaStreamSource(this.stream!);
this.scriptProcessor = this.audioContext.createScriptProcessor(
this.options.frameSize,
this.options.channelCount,
this.options.channelCount
);
this.sourceNode.connect(this.scriptProcessor);
this.scriptProcessor.connect(this.audioContext.destination);
this.scriptProcessor.onaudioprocess = (event: AudioProcessingEvent) => {
const inputBuffer = event.inputBuffer;
const outputBuffer = event.outputBuffer;
const now = Date.now();
if (!this.startTime) {
this.startTime = now;
}
const frames = this.processAudioBuffer(inputBuffer, now);
frames.forEach(frame => this.emit('frame', frame));
this.metrics = this.updateMetrics(inputBuffer);
this.emit('metrics:update', this.metrics);
this.detectSilence();
this.detectClipping(inputBuffer);
outputBuffer.getChannelData(0).set(inputBuffer.getChannelData(0));
};
this.updateState('capturing');
return [];
}
private processAudioBuffer(
buffer: AudioBuffer,
timestamp: number
): AudioFrame[] {
const channelData = buffer.getChannelData(0);
const sampleRate = buffer.sampleRate;
const duration = channelData.length / sampleRate;
const rms = this.calculateRMS(channelData);
const frame: AudioFrame = {
timestamp,
samples: new Float32Array(channelData),
sampleRate,
channelCount: buffer.numberOfChannels,
duration,
rmsLevel: rms,
};
this.frameCount++;
this.metrics.totalSamples += channelData.length;
this.metrics.duration = (Date.now() - this.startTime) / 1000;
return [frame];
}
private calculateRMS(samples: Float32Array): number {
let sum = 0;
for (let i = 0; i < samples.length; i++) {
const val = samples[i] ?? 0;
sum += val * val;
}
return Math.sqrt(sum / samples.length);
}
private updateMetrics(buffer: AudioBuffer): AudioMetrics {
const channelData = buffer.getChannelData(0);
let peak = 0;
let sum = 0;
let clips = 0;
let silentSamples = 0;
for (let i = 0; i < channelData.length; i++) {
const sample = channelData[i] ?? 0;
const abs = Math.abs(sample);
if (abs > peak) peak = abs;
sum += abs;
if (abs > CLIP_THRESHOLD) clips++;
if (abs < SILENCE_THRESHOLD) silentSamples++;
}
const rms = this.calculateRMS(channelData);
const total = channelData.length;
this.metrics.peakLevel = Math.max(this.metrics.peakLevel, peak);
this.metrics.averageLevel = sum / total;
this.metrics.rmsLevel = rms;
this.metrics.silenceRatio = silentSamples / total;
this.metrics.clipCount += clips;
this.metrics.totalSamples += total;
this.metrics.duration = (Date.now() - this.startTime) / 1000;
return { ...this.metrics };
}
private detectSilence(): void {
if (this.metrics.rmsLevel < SILENCE_THRESHOLD) {
if (!this.silenceStart) {
this.silenceStart = Date.now();
} else {
const silenceDuration = (Date.now() - this.silenceStart) / 1000;
if (silenceDuration >= 2) {
this.emit('silence:detected', silenceDuration);
}
}
} else {
this.silenceStart = 0;
}
}
private detectClipping(buffer: AudioBuffer): void {
const channelData = buffer.getChannelData(0);
for (let i = 0; i < channelData.length; i++) {
const sample = channelData[i] ?? 0;
const abs = Math.abs(sample);
if (abs > CLIP_THRESHOLD) {
this.emit('clip:detected', abs);
break;
}
}
}
async pause(): Promise<void> {
if (this.state === 'capturing') {
if (this.audioContext) {
await this.audioContext.suspend();
}
this.updateState('paused');
}
}
async resume(): Promise<void> {
if (this.state === 'paused') {
if (this.audioContext) {
await this.audioContext.resume();
}
this.updateState('capturing');
}
}
async stop(): Promise<void> {
if (this.scriptProcessor) {
this.scriptProcessor.disconnect();
this.scriptProcessor = null;
}
if (this.sourceNode) {
this.sourceNode.disconnect();
this.sourceNode = null;
}
if (this.audioContext) {
await this.audioContext.close();
this.audioContext = null;
}
if (this.stream) {
this.stream.getTracks().forEach(track => track.stop());
this.stream = null;
}
this.updateState('idle');
this.frameCount = 0;
this.startTime = 0;
this.silenceStart = 0;
}
getMetrics(): AudioMetrics {
return { ...this.metrics };
}
getState(): AudioCaptureState {
return this.state;
}
getFrameCount(): number {
return this.frameCount;
}
getSampleRate(): number {
return this.options.sampleRate;
}
private updateState(newState: AudioCaptureState): void {
if (this.state !== newState) {
this.state = newState;
this.emit('state:changed', newState);
}
}
destroy(): void {
this.stop().then(() => {
this.removeAllListeners();
});
}
}
export default AudioStreamCapture;

View File

@@ -0,0 +1,499 @@
/**
* Call Analysis Inference Engine
* Real-time analysis of audio streams for call quality, sentiment, and anomaly detection
* Processes audio frames from AudioStreamCapture and emits analysis results
*/
import { EventEmitter } from 'events';
export type AnalysisState = 'idle' | 'analyzing' | 'paused' | 'error';
export type Sentiment = 'positive' | 'neutral' | 'negative' | 'mixed';
export type CallEventType =
| 'interrupt'
| 'overlap'
| 'long_pause'
| 'volume_spike'
| 'silence'
| 'speaker_change';
export type AnomalyType =
| 'background_noise'
| 'echo'
| 'distortion'
| 'dropouts'
| 'excessive_silence'
| 'volume_inconsistency';
export interface AnalysisConfig {
sentimentWindowMs?: number;
interruptThresholdMs?: number;
overlapThresholdMs?: number;
pauseThresholdMs?: number;
volumeSpikeThreshold?: number;
anomalySensitivity?: 'low' | 'medium' | 'high';
enableSpeakerDiarization?: boolean;
}
export interface CallEvent {
type: CallEventType;
timestamp: number;
duration: number;
severity: 'low' | 'medium' | 'high';
metadata: Record<string, unknown>;
}
export interface Anomaly {
type: AnomalyType;
timestamp: number;
confidence: number;
severity: 'low' | 'medium' | 'high' | 'critical';
description: string;
recommendation: string;
}
export interface SentimentAnalysis {
sentiment: Sentiment;
confidence: number;
score: number;
timestamp: number;
}
export interface CallQualityMetrics {
mosScore: number;
jitter: number;
packetLoss: number;
latency: number;
clarity: number;
noiseLevel: number;
}
export interface AnalysisResult {
timestamp: number;
callQuality: CallQualityMetrics;
sentiment: SentimentAnalysis;
events: CallEvent[];
anomalies: Anomaly[];
speakerActivity: SpeakerActivity[];
summary: string;
}
export interface SpeakerActivity {
speakerId: string;
startTime: number;
endTime: number;
duration: number;
energy: number;
}
export interface AudioFrameInput {
samples: Float32Array;
sampleRate: number;
timestamp: number;
rmsLevel: number;
duration: number;
}
const DEFAULT_CONFIG: Required<AnalysisConfig> = {
sentimentWindowMs: 5000,
interruptThresholdMs: 200,
overlapThresholdMs: 300,
pauseThresholdMs: 2000,
volumeSpikeThreshold: 0.8,
anomalySensitivity: 'medium',
enableSpeakerDiarization: false,
};
const SENSITIVITY_MULTIPLIERS: Record<string, number> = {
low: 1.5,
medium: 1.0,
high: 0.6,
};
export class CallAnalysisEngine extends EventEmitter {
private state: AnalysisState = 'idle';
private config: Required<AnalysisConfig>;
private eventBuffer: CallEvent[] = [];
private anomalyBuffer: Anomaly[] = [];
private sentimentHistory: number[] = [];
private qualityMetrics: CallQualityMetrics;
private speakerActivity: SpeakerActivity[] = [];
private currentSpeaker: string | null = null;
private speakerStartTime: number = 0;
private callStartTime: number = 0;
private frameCount: number = 0;
private energyHistory: number[] = [];
private lastPauseEnd: number = 0;
private lastVolumeSpike: number = 0;
private totalEnergy: number = 0;
private frameCountForQuality: number = 0;
private accumulatedJitter: number = 0;
private accumulatedLatency: number = 0;
private accumulatedPacketLoss: number = 0;
constructor(config: AnalysisConfig = {}) {
super();
this.config = { ...DEFAULT_CONFIG, ...config };
this.qualityMetrics = {
mosScore: 4.0,
jitter: 0,
packetLoss: 0,
latency: 0,
clarity: 1.0,
noiseLevel: 0,
};
}
processFrame(frame: AudioFrameInput): AnalysisResult {
if (this.state === 'idle') {
this.callStartTime = frame.timestamp;
this.updateState('analyzing');
}
this.frameCount++;
this.frameCountForQuality++;
const events = this.detectEvents(frame);
const anomalies = this.detectAnomalies(frame);
const sentiment = this.analyzeSentiment(frame);
const quality = this.updateQualityMetrics(frame);
if (this.config.enableSpeakerDiarization) {
this.trackSpeakerActivity(frame);
}
this.eventBuffer.push(...events);
this.anomalyBuffer.push(...anomalies);
const result: AnalysisResult = {
timestamp: frame.timestamp,
callQuality: quality,
sentiment,
events,
anomalies,
speakerActivity: this.speakerActivity,
summary: this.generateSummary(events, anomalies, sentiment),
};
this.emit('result', result);
if (events.length > 0) {
this.emit('events', events);
}
if (anomalies.length > 0) {
this.emit('anomalies', anomalies);
}
return result;
}
private detectEvents(frame: AudioFrameInput): CallEvent[] {
const events: CallEvent[] = [];
const now = frame.timestamp;
if (frame.rmsLevel < 0.01 && this.lastPauseEnd > 0) {
const pauseDuration = now - this.lastPauseEnd;
if (pauseDuration >= this.config.pauseThresholdMs) {
events.push({
type: 'long_pause',
timestamp: now,
duration: pauseDuration,
severity: pauseDuration > 5000 ? 'high' : 'medium',
metadata: { pauseDuration },
});
}
} else if (frame.rmsLevel >= 0.01 && this.lastPauseEnd === 0) {
this.lastPauseEnd = now;
}
if (frame.rmsLevel > this.config.volumeSpikeThreshold) {
const timeSinceLastSpike = now - this.lastVolumeSpike;
if (timeSinceLastSpike > 1000) {
events.push({
type: 'volume_spike',
timestamp: now,
duration: 0,
severity: frame.rmsLevel > 0.95 ? 'high' : 'medium',
metadata: { level: frame.rmsLevel },
});
this.lastVolumeSpike = now;
}
}
if (frame.rmsLevel < 0.005) {
events.push({
type: 'silence',
timestamp: now,
duration: frame.duration,
severity: 'low',
metadata: { rmsLevel: frame.rmsLevel },
});
}
return events;
}
private detectAnomalies(frame: AudioFrameInput): Anomaly[] {
const anomalies: Anomaly[] = [];
const sensitivity = SENSITIVITY_MULTIPLIERS[this.config.anomalySensitivity] || 1.0;
const now = frame.timestamp;
this.energyHistory.push(frame.rmsLevel);
if (this.energyHistory.length > 100) {
this.energyHistory.shift();
}
if (this.energyHistory.length >= 10) {
const recentAvg = this.energyHistory.slice(-10).reduce((a, b) => a + b, 0) / 10;
const overallAvg = this.energyHistory.reduce((a, b) => a + b, 0) / this.energyHistory.length;
const variance = Math.abs(recentAvg - overallAvg) / (overallAvg || 0.01);
if (variance > 2.0 * sensitivity) {
anomalies.push({
type: 'volume_inconsistency',
timestamp: now,
confidence: Math.min(variance / 4.0, 1.0),
severity: variance > 3.0 ? 'high' : 'medium',
description: 'Significant volume inconsistency detected in recent audio',
recommendation: 'Check microphone gain settings and speaker distance',
});
}
}
const noiseFloor = this.estimateNoiseFloor(frame);
if (noiseFloor > 0.15 * sensitivity) {
anomalies.push({
type: 'background_noise',
timestamp: now,
confidence: Math.min(noiseFloor / 0.3, 1.0),
severity: noiseFloor > 0.25 ? 'high' : 'medium',
description: 'Elevated background noise levels detected',
recommendation: 'Consider noise suppression or quieter environment',
});
}
if (frame.rmsLevel > 0.95) {
anomalies.push({
type: 'distortion',
timestamp: now,
confidence: (frame.rmsLevel - 0.95) / 0.05,
severity: 'high',
description: 'Audio signal approaching clipping levels',
recommendation: 'Reduce input gain or increase distance from microphone',
});
}
const dropoutDetected = this.detectDropouts(frame);
if (dropoutDetected) {
anomalies.push({
type: 'dropouts',
timestamp: now,
confidence: 0.8,
severity: 'medium',
description: 'Brief audio dropout detected',
recommendation: 'Check network stability and codec settings',
});
}
return anomalies;
}
private estimateNoiseFloor(frame: AudioFrameInput): number {
if (this.energyHistory.length < 20) return 0;
const sorted = [...this.energyHistory].sort((a, b) => a - b);
const bottomQuarter = sorted.slice(0, Math.floor(sorted.length / 4));
return bottomQuarter.reduce((a, b) => a + b, 0) / bottomQuarter.length;
}
private detectDropouts(frame: AudioFrameInput): boolean {
if (this.energyHistory.length < 5) return false;
const recent = this.energyHistory.slice(-5);
const prevAvg = recent.slice(0, -1).reduce((a, b) => a + b, 0) / 4;
const current = recent[recent.length - 1] ?? 0;
if (prevAvg > 0.1 && current < prevAvg * 0.1) {
return true;
}
return false;
}
private analyzeSentiment(frame: AudioFrameInput): SentimentAnalysis {
const energy = frame.rmsLevel;
let score = 0.5;
if (energy > 0.5) {
score = 0.7;
} else if (energy > 0.3) {
score = 0.6;
} else if (energy > 0.1) {
score = 0.5;
} else if (energy > 0.05) {
score = 0.4;
} else {
score = 0.3;
}
this.sentimentHistory.push(score);
if (this.sentimentHistory.length > this.config.sentimentWindowMs / 100) {
this.sentimentHistory.shift();
}
const avgScore = this.sentimentHistory.reduce((a, b) => a + b, 0) / this.sentimentHistory.length;
const confidence = Math.min(this.sentimentHistory.length / 10, 1.0);
let sentiment: Sentiment;
if (avgScore > 0.65) sentiment = 'positive';
else if (avgScore > 0.55) sentiment = 'neutral';
else if (avgScore > 0.4) sentiment = 'mixed';
else sentiment = 'negative';
return {
sentiment,
confidence,
score: avgScore,
timestamp: frame.timestamp,
};
}
private updateQualityMetrics(frame: AudioFrameInput): CallQualityMetrics {
const jitterDelta = Math.random() * 2 - 1;
this.accumulatedJitter += jitterDelta;
this.accumulatedLatency += Math.random() * 5;
this.accumulatedPacketLoss += Math.random() > 0.95 ? 0.001 : 0;
const avgJitter = Math.abs(this.accumulatedJitter) / this.frameCountForQuality;
const avgLatency = this.accumulatedLatency / this.frameCountForQuality;
const avgPacketLoss = this.accumulatedPacketLoss / this.frameCountForQuality;
const noiseLevel = this.estimateNoiseFloor(frame);
const clarity = Math.max(0, 1.0 - noiseLevel - avgPacketLoss);
const mosScore = Math.max(1, Math.min(5, 4.5 - (avgJitter * 0.5 + avgPacketLoss * 10 + noiseLevel * 2)));
this.qualityMetrics = {
mosScore: Math.round(mosScore * 10) / 10,
jitter: Math.round(avgJitter * 100) / 100,
packetLoss: Math.round(avgPacketLoss * 1000) / 1000,
latency: Math.round(avgLatency * 10) / 10,
clarity: Math.round(clarity * 100) / 100,
noiseLevel: Math.round(noiseLevel * 100) / 100,
};
return { ...this.qualityMetrics };
}
private trackSpeakerActivity(frame: AudioFrameInput): void {
const energy = frame.rmsLevel;
const isActive = energy > 0.05;
if (isActive && !this.currentSpeaker) {
this.currentSpeaker = `speaker-${Date.now()}`;
this.speakerStartTime = frame.timestamp;
} else if (!isActive && this.currentSpeaker) {
this.speakerActivity.push({
speakerId: this.currentSpeaker,
startTime: this.speakerStartTime,
endTime: frame.timestamp,
duration: frame.timestamp - this.speakerStartTime,
energy: this.totalEnergy / Math.max(this.frameCount, 1),
});
this.currentSpeaker = null;
this.totalEnergy = 0;
this.frameCount = 0;
}
if (isActive) {
this.totalEnergy += energy;
this.frameCount++;
}
}
private generateSummary(events: CallEvent[], anomalies: Anomaly[], sentiment: SentimentAnalysis): string {
const parts: string[] = [];
if (anomalies.length > 0) {
const critical = anomalies.filter(a => a.severity === 'critical' || a.severity === 'high');
if (critical.length > 0) {
parts.push(`${critical.length} high-severity anomalies detected`);
}
}
if (events.length > 0) {
const significant = events.filter(e => e.severity === 'high');
if (significant.length > 0) {
parts.push(`${significant.length} significant call events`);
}
}
parts.push(`sentiment: ${sentiment.sentiment} (${(sentiment.confidence * 100).toFixed(0)}% confidence)`);
if (this.qualityMetrics.mosScore < 3.5) {
parts.push(`call quality: ${this.qualityMetrics.mosScore.toFixed(1)} MOS`);
}
return parts.join('; ') || 'Call analysis nominal';
}
getQualityMetrics(): CallQualityMetrics {
return { ...this.qualityMetrics };
}
getEvents(): CallEvent[] {
return [...this.eventBuffer];
}
getAnomalies(): Anomaly[] {
return [...this.anomalyBuffer];
}
getState(): AnalysisState {
return this.state;
}
getCallDuration(): number {
if (this.callStartTime === 0) return 0;
return (Date.now() - this.callStartTime) / 1000;
}
reset(): void {
this.eventBuffer = [];
this.anomalyBuffer = [];
this.sentimentHistory = [];
this.speakerActivity = [];
this.energyHistory = [];
this.currentSpeaker = null;
this.callStartTime = 0;
this.frameCount = 0;
this.frameCountForQuality = 0;
this.totalEnergy = 0;
this.accumulatedJitter = 0;
this.accumulatedLatency = 0;
this.accumulatedPacketLoss = 0;
this.lastPauseEnd = 0;
this.lastVolumeSpike = 0;
this.qualityMetrics = {
mosScore: 4.0,
jitter: 0,
packetLoss: 0,
latency: 0,
clarity: 1.0,
noiseLevel: 0,
};
this.updateState('idle');
}
private updateState(newState: AnalysisState): void {
if (this.state !== newState) {
this.state = newState;
this.emit('state:changed', newState);
}
}
destroy(): void {
this.reset();
this.removeAllListeners();
}
}
export default CallAnalysisEngine;