@tensamin/audio 0.1.14 → 0.2.0

package/dist/pipeline/audio-pipeline.mjs

@@ -1,11 +1,10 @@
 import {
   createAudioPipeline
-} from "../chunk-R5M2DGAQ.mjs";
-import "../chunk-K4YLH73B.mjs";
+} from "../chunk-K4J3UUOR.mjs";
 import "../chunk-OZ7KMC4S.mjs";
-import "../chunk-UFKIAMG3.mjs";
-import "../chunk-XO6B3D4A.mjs";
-import "../chunk-2G2JFHJY.mjs";
+import "../chunk-IS37FHDN.mjs";
+import "../chunk-QNQK6QFB.mjs";
+import "../chunk-AQ5RVY33.mjs";
 export {
   createAudioPipeline
 };

package/dist/types.d.mts

@@ -1,270 +1,55 @@
 import { Emitter } from 'mitt';
 
-/**
- * Configuration for the audio processing pipeline.
- */
-interface AudioProcessingConfig {
-    /**
-     * Noise suppression configuration.
-     */
-    noiseSuppression?: {
-        enabled: boolean;
-        /**
-         * Path or URL to the RNNoise WASM binary.
-         * REQUIRED if enabled.
-         */
-        wasmUrl?: string;
-        /**
-         * Path or URL to the RNNoise SIMD WASM binary.
-         * REQUIRED if enabled.
-         */
-        simdUrl?: string;
-        /**
-         * Path or URL to the RNNoise worklet script.
-         * REQUIRED if enabled.
-         */
-        workletUrl?: string;
-        /**
-         * Plugin name to use. Defaults to 'rnnoise-ns'.
-         */
-        pluginName?: string;
-    };
-    /**
-     * Voice Activity Detection (VAD) configuration.
-     */
-    vad?: {
-        enabled: boolean;
-        /**
-         * Plugin name to use. Defaults to 'energy-vad'.
-         */
-        pluginName?: string;
-        /**
-         * Probability threshold for speech onset (0-1).
-         * When VAD probability rises above this, audio is unmuted.
-         * Lower = more sensitive (catches quiet speech, may include noise)
-         * Higher = less sensitive (only confident speech, may clip quiet parts)
-         * Default: 0.8 (aggressive noise rejection)
-         */
-        startThreshold?: number;
-        /**
-         * Probability threshold for speech offset (0-1).
-         * When VAD probability drops below this (after hangover), audio is muted.
-         * Lower = keeps audio on longer (less aggressive gating)
-         * Higher = mutes faster (more aggressive noise suppression)
-         * Default: 0.3 (wide hysteresis for stability)
-         */
-        stopThreshold?: number;
-        /**
-         * Time in ms to wait after speech stops before muting.
-         * Prevents rapid on/off toggling during pauses.
-         * Lower = more aggressive gating, may clip between words
-         * Higher = smoother but may let trailing noise through
-         * Default: 300ms
-         */
-        hangoverMs?: number;
-        /**
-         * Time in ms of audio to buffer before speech onset.
-         * Prevents cutting off the beginning of speech.
-         * Default: 250ms (generous pre-roll for voice)
-         */
-        preRollMs?: number;
-        /**
-         * Minimum speech duration in ms to consider it valid speech.
-         * Filters out brief transients like keyboard clicks.
-         * Default: 250ms (aggressive transient rejection)
-         */
-        minSpeechDurationMs?: number;
-        /**
-         * Minimum silence duration in ms before allowing another speech segment.
-         * Prevents false positives from quick noise bursts.
-         * Default: 150ms
-         */
-        minSilenceDurationMs?: number;
-        /**
-         * Advanced: Energy VAD specific parameters
-         */
-        energyVad?: {
-            /**
-             * Smoothing factor for energy calculation (0-1).
-             * Higher = more smoothing, slower to react
-             * Default: 0.95
-             */
-            smoothing?: number;
-            /**
-             * Initial noise floor estimate.
-             * Default: 0.001
-             */
-            initialNoiseFloor?: number;
-            /**
-             * Rate at which noise floor adapts to quiet signals (0-1).
-             * Default: 0.002 (very slow downward drift)
-             */
-            noiseFloorAdaptRateQuiet?: number;
-            /**
-             * Rate at which noise floor adapts to loud signals (0-1).
-             * Applied to low-energy, low-crest-factor signals (background noise).
-             * Default: 0.02
-             */
-            noiseFloorAdaptRateLoud?: number;
-            /**
-             * Minimum SNR (Signal-to-Noise Ratio) in dB for speech detection.
-             * Default: 12.0 (aggressive noise rejection)
-             */
-            minSNR?: number;
-            /**
-             * SNR range in dB for probability scaling.
-             * Default: 10.0 (probability scales from minSNR to minSNR+snrRange)
-             */
-            snrRange?: number;
-            /**
-             * Minimum absolute RMS energy to consider as speech.
-             * Prevents triggering on very quiet background noise.
-             * Default: 0.003 (approx -50dB, voice-appropriate level)
-             */
-            minEnergy?: number;
-        };
-    };
-    /**
-     * Output gain and muting configuration.
-     */
-    output?: {
-        /**
-         * Gain to apply when speaking (0-infinity).
-         * Values > 1.0 will amplify the voice.
-         * Default: 1.0 (unity gain)
-         */
-        speechGain?: number;
-        /**
-         * Gain to apply when silent (0-1).
-         * 0.0 = complete mute (recommended for voice-only)
-         * 0.1-0.3 = allow some background ambience
-         * Default: 0.0 (full mute for voice-only)
-         */
-        silenceGain?: number;
-        /**
-         * Time in seconds to ramp gain changes.
-         * Lower = faster transitions (may cause clicks)
-         * Higher = smoother transitions (may sound sluggish)
-         * Default: 0.015 (fast but smooth for voice)
-         */
-        gainRampTime?: number;
-        /**
-         * Apply additional gain reduction during the transition to silence.
-         * Helps create cleaner cutoffs without abrupt clicks.
-         * Default: true
-         */
-        smoothTransitions?: boolean;
-        /**
-         * Maximum gain in dB to apply (prevents clipping).
-         * Default: 6.0 dB (roughly 2x amplitude)
-         */
-        maxGainDb?: number;
-        /**
-         * Apply dynamic range compression when speaking.
-         * Makes quiet parts louder and loud parts quieter.
-         * Default: false (transparent audio)
-         */
-        enableCompression?: boolean;
-        /**
-         * Compression settings (when enabled)
-         */
-        compression?: {
-            /**
-             * Threshold in dB above which compression starts.
-             * Default: -24.0 dB
-             */
-            threshold?: number;
-            /**
-             * Compression ratio (1:N).
-             * Default: 3.0 (3:1 ratio)
-             */
-            ratio?: number;
-            /**
-             * Attack time in seconds.
-             * Default: 0.003 (3ms)
-             */
-            attack?: number;
-            /**
-             * Release time in seconds.
-             * Default: 0.05 (50ms)
-             */
-            release?: number;
-        };
-    };
-    /**
-     * LiveKit integration configuration.
-     */
-    livekit?: {
-        /**
-         * Whether to call track.mute()/unmute() on the LocalAudioTrack based on VAD.
-         * This saves bandwidth but has more signaling overhead.
-         * Default: false (uses gain gating only)
-         */
-        manageTrackMute?: boolean;
+interface NoiseSuppressionConfig {
+    enabled?: boolean;
+    noiseReductionLevel?: number;
+    assetConfig?: {
+        cdnUrl?: string;
     };
 }
-/**
- * Represents the state of Voice Activity Detection.
- */
-interface VADState {
-    /**
-     * Whether speech is currently detected (after hysteresis).
-     */
-    isSpeaking: boolean;
-    /**
-     * Raw probability of speech from the VAD model (0-1).
-     */
-    probability: number;
-    /**
-     * Current state enum.
-     */
-    state: "silent" | "speech_starting" | "speaking" | "speech_ending";
+interface SpeakingDetectionConfig {
+    minDb: number;
+    maxDb: number;
+    speakOnRatio?: number;
+    speakOffRatio?: number;
+    hangoverMs?: number;
+    attackMs?: number;
+    releaseMs?: number;
+}
+interface OutputGainConfig {
+    speechGain?: number;
+    silenceGain?: number;
+    gainRampTime?: number;
+    maxGainDb?: number;
+    smoothTransitions?: boolean;
 }
-/**
- * Events emitted by the audio pipeline.
- */
-type AudioPipelineEvents = {
-    vadChange: VADState;
+interface LivekitSpeakingOptions {
+    noiseSuppression?: NoiseSuppressionConfig;
+    speaking?: SpeakingDetectionConfig;
+    output?: OutputGainConfig;
+    muteWhenSilent?: boolean;
+}
+interface SpeakingState {
+    speaking: boolean;
+    levelDb: number;
+}
+type SpeakingEvents = {
+    speakingChange: SpeakingState;
     error: Error;
 };
-/**
- * Handle to a running audio processing pipeline.
- */
 interface AudioPipelineHandle {
-    /**
-     * The processed MediaStreamTrack.
-     */
     readonly processedTrack: MediaStreamTrack;
-    /**
-     * Event emitter for VAD state and errors.
-     */
-    readonly events: Emitter<AudioPipelineEvents>;
-    /**
-     * Current VAD state.
-     */
-    readonly state: VADState;
-    /**
-     * Update configuration at runtime.
-     */
-    setConfig(config: Partial<AudioProcessingConfig>): void;
-    /**
-     * Stop processing and release resources.
-     */
+    readonly events: Emitter<SpeakingEvents>;
+    readonly state: SpeakingState;
+    setConfig(config: Partial<LivekitSpeakingOptions>): void;
     dispose(): void;
 }
-/**
- * Interface for a Noise Suppression Plugin.
- */
-interface NoiseSuppressionPlugin {
-    name: string;
-    createNode(context: AudioContext, config: AudioProcessingConfig["noiseSuppression"]): Promise<AudioNode>;
-}
-/**
- * Interface for a VAD Plugin.
- */
-interface VADPlugin {
-    name: string;
-    createNode(context: AudioContext, config: AudioProcessingConfig["vad"], onDecision: (probability: number) => void): Promise<AudioNode>;
+interface SpeakingController {
+    readonly speaking: boolean;
+    readonly levelDb: number;
+    onChange(listener: (state: SpeakingState) => void): () => void;
+    setConfig(config: Partial<LivekitSpeakingOptions>): void;
+    dispose(): void;
 }
 
-export type { AudioPipelineEvents, AudioPipelineHandle, AudioProcessingConfig, NoiseSuppressionPlugin, VADPlugin, VADState };
+export type { AudioPipelineHandle, LivekitSpeakingOptions, NoiseSuppressionConfig, OutputGainConfig, SpeakingController, SpeakingDetectionConfig, SpeakingEvents, SpeakingState };

package/dist/types.d.ts

@@ -1,270 +1,55 @@
 import { Emitter } from 'mitt';
 
-/**
- * Configuration for the audio processing pipeline.
- */
-interface AudioProcessingConfig {
-    /**
-     * Noise suppression configuration.
-     */
-    noiseSuppression?: {
-        enabled: boolean;
-        /**
-         * Path or URL to the RNNoise WASM binary.
-         * REQUIRED if enabled.
-         */
-        wasmUrl?: string;
-        /**
-         * Path or URL to the RNNoise SIMD WASM binary.
-         * REQUIRED if enabled.
-         */
-        simdUrl?: string;
-        /**
-         * Path or URL to the RNNoise worklet script.
-         * REQUIRED if enabled.
-         */
-        workletUrl?: string;
-        /**
-         * Plugin name to use. Defaults to 'rnnoise-ns'.
-         */
-        pluginName?: string;
-    };
-    /**
-     * Voice Activity Detection (VAD) configuration.
-     */
-    vad?: {
-        enabled: boolean;
-        /**
-         * Plugin name to use. Defaults to 'energy-vad'.
-         */
-        pluginName?: string;
-        /**
-         * Probability threshold for speech onset (0-1).
-         * When VAD probability rises above this, audio is unmuted.
-         * Lower = more sensitive (catches quiet speech, may include noise)
-         * Higher = less sensitive (only confident speech, may clip quiet parts)
-         * Default: 0.8 (aggressive noise rejection)
-         */
-        startThreshold?: number;
-        /**
-         * Probability threshold for speech offset (0-1).
-         * When VAD probability drops below this (after hangover), audio is muted.
-         * Lower = keeps audio on longer (less aggressive gating)
-         * Higher = mutes faster (more aggressive noise suppression)
-         * Default: 0.3 (wide hysteresis for stability)
-         */
-        stopThreshold?: number;
-        /**
-         * Time in ms to wait after speech stops before muting.
-         * Prevents rapid on/off toggling during pauses.
-         * Lower = more aggressive gating, may clip between words
-         * Higher = smoother but may let trailing noise through
-         * Default: 300ms
-         */
-        hangoverMs?: number;
-        /**
-         * Time in ms of audio to buffer before speech onset.
-         * Prevents cutting off the beginning of speech.
-         * Default: 250ms (generous pre-roll for voice)
-         */
-        preRollMs?: number;
-        /**
-         * Minimum speech duration in ms to consider it valid speech.
-         * Filters out brief transients like keyboard clicks.
-         * Default: 250ms (aggressive transient rejection)
-         */
-        minSpeechDurationMs?: number;
-        /**
-         * Minimum silence duration in ms before allowing another speech segment.
-         * Prevents false positives from quick noise bursts.
-         * Default: 150ms
-         */
-        minSilenceDurationMs?: number;
-        /**
-         * Advanced: Energy VAD specific parameters
-         */
-        energyVad?: {
-            /**
-             * Smoothing factor for energy calculation (0-1).
-             * Higher = more smoothing, slower to react
-             * Default: 0.95
-             */
-            smoothing?: number;
-            /**
-             * Initial noise floor estimate.
-             * Default: 0.001
-             */
-            initialNoiseFloor?: number;
-            /**
-             * Rate at which noise floor adapts to quiet signals (0-1).
-             * Default: 0.002 (very slow downward drift)
-             */
-            noiseFloorAdaptRateQuiet?: number;
-            /**
-             * Rate at which noise floor adapts to loud signals (0-1).
-             * Applied to low-energy, low-crest-factor signals (background noise).
-             * Default: 0.02
-             */
-            noiseFloorAdaptRateLoud?: number;
-            /**
-             * Minimum SNR (Signal-to-Noise Ratio) in dB for speech detection.
-             * Default: 12.0 (aggressive noise rejection)
-             */
-            minSNR?: number;
-            /**
-             * SNR range in dB for probability scaling.
-             * Default: 10.0 (probability scales from minSNR to minSNR+snrRange)
-             */
-            snrRange?: number;
-            /**
-             * Minimum absolute RMS energy to consider as speech.
-             * Prevents triggering on very quiet background noise.
-             * Default: 0.003 (approx -50dB, voice-appropriate level)
-             */
-            minEnergy?: number;
-        };
-    };
-    /**
-     * Output gain and muting configuration.
-     */
-    output?: {
-        /**
-         * Gain to apply when speaking (0-infinity).
-         * Values > 1.0 will amplify the voice.
-         * Default: 1.0 (unity gain)
-         */
-        speechGain?: number;
-        /**
-         * Gain to apply when silent (0-1).
-         * 0.0 = complete mute (recommended for voice-only)
-         * 0.1-0.3 = allow some background ambience
-         * Default: 0.0 (full mute for voice-only)
-         */
-        silenceGain?: number;
-        /**
-         * Time in seconds to ramp gain changes.
-         * Lower = faster transitions (may cause clicks)
-         * Higher = smoother transitions (may sound sluggish)
-         * Default: 0.015 (fast but smooth for voice)
-         */
-        gainRampTime?: number;
-        /**
-         * Apply additional gain reduction during the transition to silence.
-         * Helps create cleaner cutoffs without abrupt clicks.
-         * Default: true
-         */
-        smoothTransitions?: boolean;
-        /**
-         * Maximum gain in dB to apply (prevents clipping).
-         * Default: 6.0 dB (roughly 2x amplitude)
-         */
-        maxGainDb?: number;
-        /**
-         * Apply dynamic range compression when speaking.
-         * Makes quiet parts louder and loud parts quieter.
-         * Default: false (transparent audio)
-         */
-        enableCompression?: boolean;
-        /**
-         * Compression settings (when enabled)
-         */
-        compression?: {
-            /**
-             * Threshold in dB above which compression starts.
-             * Default: -24.0 dB
-             */
-            threshold?: number;
-            /**
-             * Compression ratio (1:N).
-             * Default: 3.0 (3:1 ratio)
-             */
-            ratio?: number;
-            /**
-             * Attack time in seconds.
-             * Default: 0.003 (3ms)
-             */
-            attack?: number;
-            /**
-             * Release time in seconds.
-             * Default: 0.05 (50ms)
-             */
-            release?: number;
-        };
-    };
-    /**
-     * LiveKit integration configuration.
-     */
-    livekit?: {
-        /**
-         * Whether to call track.mute()/unmute() on the LocalAudioTrack based on VAD.
-         * This saves bandwidth but has more signaling overhead.
-         * Default: false (uses gain gating only)
-         */
-        manageTrackMute?: boolean;
+interface NoiseSuppressionConfig {
+    enabled?: boolean;
+    noiseReductionLevel?: number;
+    assetConfig?: {
+        cdnUrl?: string;
     };
 }
-/**
- * Represents the state of Voice Activity Detection.
- */
-interface VADState {
-    /**
-     * Whether speech is currently detected (after hysteresis).
-     */
-    isSpeaking: boolean;
-    /**
-     * Raw probability of speech from the VAD model (0-1).
-     */
-    probability: number;
-    /**
-     * Current state enum.
-     */
-    state: "silent" | "speech_starting" | "speaking" | "speech_ending";
+interface SpeakingDetectionConfig {
+    minDb: number;
+    maxDb: number;
+    speakOnRatio?: number;
+    speakOffRatio?: number;
+    hangoverMs?: number;
+    attackMs?: number;
+    releaseMs?: number;
+}
+interface OutputGainConfig {
+    speechGain?: number;
+    silenceGain?: number;
+    gainRampTime?: number;
+    maxGainDb?: number;
+    smoothTransitions?: boolean;
 }
-/**
- * Events emitted by the audio pipeline.
- */
-type AudioPipelineEvents = {
-    vadChange: VADState;
+interface LivekitSpeakingOptions {
+    noiseSuppression?: NoiseSuppressionConfig;
+    speaking?: SpeakingDetectionConfig;
+    output?: OutputGainConfig;
+    muteWhenSilent?: boolean;
+}
+interface SpeakingState {
+    speaking: boolean;
+    levelDb: number;
+}
+type SpeakingEvents = {
+    speakingChange: SpeakingState;
     error: Error;
 };
-/**
- * Handle to a running audio processing pipeline.
- */
 interface AudioPipelineHandle {
-    /**
-     * The processed MediaStreamTrack.
-     */
     readonly processedTrack: MediaStreamTrack;
-    /**
-     * Event emitter for VAD state and errors.
-     */
-    readonly events: Emitter<AudioPipelineEvents>;
-    /**
-     * Current VAD state.
-     */
-    readonly state: VADState;
-    /**
-     * Update configuration at runtime.
-     */
-    setConfig(config: Partial<AudioProcessingConfig>): void;
-    /**
-     * Stop processing and release resources.
-     */
+    readonly events: Emitter<SpeakingEvents>;
+    readonly state: SpeakingState;
+    setConfig(config: Partial<LivekitSpeakingOptions>): void;
     dispose(): void;
 }
-/**
- * Interface for a Noise Suppression Plugin.
- */
-interface NoiseSuppressionPlugin {
-    name: string;
-    createNode(context: AudioContext, config: AudioProcessingConfig["noiseSuppression"]): Promise<AudioNode>;
-}
-/**
- * Interface for a VAD Plugin.
- */
-interface VADPlugin {
-    name: string;
-    createNode(context: AudioContext, config: AudioProcessingConfig["vad"], onDecision: (probability: number) => void): Promise<AudioNode>;
+interface SpeakingController {
+    readonly speaking: boolean;
+    readonly levelDb: number;
+    onChange(listener: (state: SpeakingState) => void): () => void;
+    setConfig(config: Partial<LivekitSpeakingOptions>): void;
+    dispose(): void;
 }
 
-export type { AudioPipelineEvents, AudioPipelineHandle, AudioProcessingConfig, NoiseSuppressionPlugin, VADPlugin, VADState };
+export type { AudioPipelineHandle, LivekitSpeakingOptions, NoiseSuppressionConfig, OutputGainConfig, SpeakingController, SpeakingDetectionConfig, SpeakingEvents, SpeakingState };

package/dist/vad/vad-node.d.mts

@@ -1,11 +1,9 @@
-import { VADPlugin, AudioProcessingConfig } from '../types.mjs';
-import 'mitt';
-
-declare class EnergyVADPlugin implements VADPlugin {
-    name: string;
-    private workletNode;
-    createNode(context: AudioContext, config: AudioProcessingConfig["vad"], onDecision: (probability: number) => void): Promise<AudioNode>;
-    updateSpeakingState(isSpeaking: boolean): void;
+interface LevelDetectorNode {
+    node: AudioWorkletNode;
+    dispose: () => void;
 }
+declare function createLevelDetectorNode(context: AudioContext, onLevel: (levelDb: number) => void, options?: {
+    smoothing?: number;
+}): Promise<LevelDetectorNode>;
 
-export { EnergyVADPlugin };
+export { type LevelDetectorNode, createLevelDetectorNode };

package/dist/vad/vad-node.d.ts

@@ -1,11 +1,9 @@
-import { VADPlugin, AudioProcessingConfig } from '../types.js';
-import 'mitt';
-
-declare class EnergyVADPlugin implements VADPlugin {
-    name: string;
-    private workletNode;
-    createNode(context: AudioContext, config: AudioProcessingConfig["vad"], onDecision: (probability: number) => void): Promise<AudioNode>;
-    updateSpeakingState(isSpeaking: boolean): void;
+interface LevelDetectorNode {
+    node: AudioWorkletNode;
+    dispose: () => void;
 }
+declare function createLevelDetectorNode(context: AudioContext, onLevel: (levelDb: number) => void, options?: {
+    smoothing?: number;
+}): Promise<LevelDetectorNode>;
 
-export { EnergyVADPlugin };
+export { type LevelDetectorNode, createLevelDetectorNode };