{"version":3,"file":"VowelClient-DI7aJvGC.cjs","sources":["../../lib/vowel/managers/StateManager.ts","../../lib/vowel/managers/audio-processor.worklet.js?raw","../../lib/vowel/types/constants.ts","../../lib/vowel/managers/AudioManager.ts","../../lib/vowel/types/types.ts","../../lib/vowel/managers/ToolManager.ts","../../lib/vowel/utils/cdnLoader.ts","../../lib/vowel/managers/VADManager.ts","../../lib/vowel/vad/VADAdapter.ts","../../lib/vowel/vad/VADRegistry.ts","../../node_modules/onnxruntime-common/dist/esm/backend-impl.js","../../node_modules/onnxruntime-common/dist/esm/version.js","../../node_modules/onnxruntime-common/dist/esm/env-impl.js","../../node_modules/onnxruntime-common/dist/esm/env.js","../../node_modules/onnxruntime-common/dist/esm/tensor-conversion-impl.js","../../node_modules/onnxruntime-common/dist/esm/tensor-factory-impl.js","../../node_modules/onnxruntime-common/dist/esm/tensor-impl-type-mapping.js","../../node_modules/onnxruntime-common/dist/esm/tensor-utils-impl.js","../../node_modules/onnxruntime-common/dist/esm/tensor-impl.js","../../node_modules/onnxruntime-common/dist/esm/tensor.js","../../node_modules/onnxruntime-common/dist/esm/trace.js","../../node_modules/onnxruntime-common/dist/esm/inference-session-impl.js","../../node_modules/onnxruntime-common/dist/esm/inference-session.js","../../node_modules/onnxruntime-common/dist/esm/index.js","../../node_modules/@huggingface/transformers/dist/transformers.web.js","../../lib/vowel/vad/adapters/SmartTurnAdapter.ts","../../lib/vowel/vad/adapters/SileroVADAdapter.ts","../../lib/vowel/vad/adapters/SimpleVADAdapter.ts","../../lib/vowel/managers/EnhancedVADManager.ts","../../lib/vowel/providers/RealtimeProvider.ts","../../node_modules/retry/lib/retry_operation.js","../../node_modules/retry/lib/retry.js","../../node_modules/retry/index.js","../../node_modules/p-retry/index.js","../../node_modules/@google/genai/dist/web/index.mjs","../../lib/vowel/providers/GeminiRealtimeProvider.ts","../../node_modules/@openai/agents-realtime/dist/metadata.mjs","../../node_modules/@openai/agents-core/dist/utils/safeExecute.mjs","../../node_modules/zod/v4/core/core.js","../../node_modules/zod/v4/core/util.js","../../node_modules/zod/v4/core/errors.js","../../node_modules/zod/v4/core/parse.js","../../node_modules/zod/v4/core/regexes.js","../../node_modules/zod/v4/core/checks.js","../../node_modules/zod/v4/core/doc.js","../../node_modules/zod/v4/core/versions.js","../../node_modules/zod/v4/core/schemas.js","../../node_modules/zod/v4/core/registries.js","../../node_modules/zod/v4/core/api.js","../../node_modules/zod/v4/core/to-json-schema.js","../../node_modules/zod/v4/core/json-schema-processors.js","../../node_modules/zod/v4/classic/iso.js","../../node_modules/zod/v4/classic/errors.js","../../node_modules/zod/v4/classic/parse.js","../../node_modules/zod/v4/classic/schemas.js","../../node_modules/zod/v4/classic/compat.js","../../node_modules/openai/lib/parser.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/Options.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/util.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/Refs.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/errorMessages.mjs","../../node_modules/zod/v3/helpers/util.js","../../node_modules/zod/v3/ZodError.js","../../node_modules/zod/v3/helpers/errorUtil.js","../../node_modules/zod/v3/types.js","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/any.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/array.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/bigint.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/boolean.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/branded.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/catch.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/date.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/default.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/effects.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/enum.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/intersection.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/literal.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/string.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/record.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/map.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/nativeEnum.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/never.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/null.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/union.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/nullable.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/number.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/object.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/optional.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/pipeline.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/promise.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/set.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/tuple.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/undefined.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/unknown.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parsers/readonly.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/parseDef.mjs","../../node_modules/openai/_vendor/zod-to-json-schema/zodToJsonSchema.mjs","../../node_modules/openai/lib/ResponsesParser.mjs","../../node_modules/openai/lib/transform.mjs","../../node_modules/openai/helpers/zod.mjs","../../node_modules/@openai/agents-core/dist/errors.mjs","../../node_modules/@openai/agents-core/dist/utils/zodCompat.mjs","../../node_modules/@openai/agents-core/dist/utils/typeGuards.mjs","../../node_modules/@openai/agents-core/dist/utils/zodJsonSchemaCompat.mjs","../../node_modules/@openai/agents-core/dist/utils/tools.mjs","../../node_modules/@openai/agents-core/dist/utils/abortSignals.mjs","../../node_modules/ms/index.js","../../node_modules/debug/src/common.js","../../node_modules/debug/src/browser.js","../../node_modules/@openai/agents-core/dist/config.mjs","../../node_modules/@openai/agents-core/dist/logger.mjs","../../node_modules/@openai/agents-core/dist/utils/smartString.mjs","../../node_modules/@openai/agents-core/dist/tooling.mjs","../../node_modules/@openai/agents-core/dist/toolIdentity.mjs","../../node_modules/@openai/agents-core/dist/toolGuardrail.mjs","../../node_modules/@openai/agents-core/dist/tool.mjs","../../node_modules/@openai/agents-core/dist/mcp.mjs","../../node_modules/@openai/agents-core/dist/shims/shims-browser.mjs","../../node_modules/@openai/agents-core/dist/utils/base64.mjs","../../node_modules/@openai/agents-core/dist/utils/binary.mjs","../../node_modules/@openai/agents-core/dist/runner/items.mjs","../../node_modules/@openai/agents-core/dist/types/protocol.mjs","../../node_modules/@openai/agents-core/dist/result.mjs","../../node_modules/@openai/agents-core/dist/tracing/context.mjs","../../node_modules/@openai/agents-core/dist/tracing/processor.mjs","../../node_modules/@openai/agents-core/dist/tracing/utils.mjs","../../node_modules/@openai/agents-core/dist/tracing/spans.mjs","../../node_modules/@openai/agents-core/dist/tracing/traces.mjs","../../node_modules/@openai/agents-core/dist/tracing/provider.mjs","../../node_modules/@openai/agents-core/dist/tracing/createSpans.mjs","../../node_modules/@openai/agents-core/dist/tracing/index.mjs","../../node_modules/@openai/agents-core/dist/lifecycle.mjs","../../node_modules/@openai/agents-core/dist/defaultModel.mjs","../../node_modules/@openai/agents-core/dist/usage.mjs","../../node_modules/@openai/agents-core/dist/runContext.mjs","../../node_modules/@openai/agents-core/dist/handoff.mjs","../../node_modules/@openai/agents-core/dist/events.mjs","../../node_modules/@openai/agents-core/dist/guardrail.mjs","../../node_modules/@openai/agents-core/dist/providers.mjs","../../node_modules/@openai/agents-core/dist/agentToolSourceRegistry.mjs","../../node_modules/@openai/agents-core/dist/items.mjs","../../node_modules/@openai/agents-core/dist/runner/toolUseTracker.mjs","../../node_modules/@openai/agents-core/dist/runner/steps.mjs","../../node_modules/@openai/agents-core/dist/utils/serialize.mjs","../../node_modules/@openai/agents-core/dist/runner/toolSearch.mjs","../../node_modules/@openai/agents-core/dist/runState.mjs","../../node_modules/@openai/agents-core/dist/runner/constants.mjs","../../node_modules/@openai/agents-core/dist/runner/conversation.mjs","../../node_modules/@openai/agents-core/dist/runner/guardrails.mjs","../../node_modules/@openai/agents-core/dist/runner/modelSettingsMerge.mjs","../../node_modules/@openai/agents-core/dist/runner/modelSettings.mjs","../../node_modules/@openai/agents-core/dist/runner/modelRetry.mjs","../../node_modules/@openai/agents-core/dist/runner/modelOutputs.mjs","../../node_modules/@openai/agents-core/dist/runner/streaming.mjs","../../node_modules/@openai/agents-core/dist/memory/session.mjs","../../node_modules/@openai/agents-core/dist/runner/sessionPersistence.mjs","../../node_modules/@openai/agents-core/dist/utils/messages.mjs","../../node_modules/@openai/agents-core/dist/agentToolRunConfig.mjs","../../node_modules/@openai/agents-core/dist/agentToolRunResults.mjs","../../node_modules/@openai/agents-core/dist/helpers/message.mjs","../../node_modules/@openai/agents-core/dist/utils/toolGuardrails.mjs","../../node_modules/@openai/agents-core/dist/runner/approvalRejection.mjs","../../node_modules/@openai/agents-core/dist/runner/toolExecution.mjs","../../node_modules/@openai/agents-core/dist/runner/mcpApprovals.mjs","../../node_modules/@openai/agents-core/dist/runner/turnResolution.mjs","../../node_modules/@openai/agents-core/dist/runner/tracing.mjs","../../node_modules/@openai/agents-core/dist/runner/modelPreparation.mjs","../../node_modules/@openai/agents-core/dist/runner/turnPreparation.mjs","../../node_modules/@openai/agents-core/dist/runner/runLoop.mjs","../../node_modules/@openai/agents-core/dist/runner/errorHandlers.mjs","../../node_modules/@openai/agents-core/dist/run.mjs","../../node_modules/@openai/agents-core/dist/agentToolInput.mjs","../../node_modules/@openai/agents-core/dist/agent.mjs","../../node_modules/@openai/agents-core/dist/mcpServers.mjs","../../node_modules/@openai/agents-core/dist/index.mjs","../../node_modules/@openai/agents-realtime/dist/utils.mjs","../../node_modules/@openai/agents-realtime/dist/realtimeAgent.mjs","../../node_modules/@openai/agents-realtime/dist/guardrail.mjs","../../node_modules/@openai/agents-realtime/dist/clientMessages.mjs","../../node_modules/@openai/agents-realtime/dist/items.mjs","../../node_modules/@openai/agents-realtime/dist/logger.mjs","../../node_modules/@openai/agents-realtime/dist/openaiRealtimeEvents.mjs","../../node_modules/@openai/agents-realtime/dist/openaiRealtimeBase.mjs","../../node_modules/@openai/agents-realtime/dist/responseCreateSequencer.mjs","../../node_modules/@openai/agents-realtime/dist/openaiRealtimeWebRtc.mjs","../../node_modules/@openai/agents-realtime/dist/shims/shims-browser.mjs","../../node_modules/@openai/agents-realtime/dist/openaiRealtimeWebsocket.mjs","../../node_modules/@openai/agents-realtime/dist/tool.mjs","../../node_modules/@openai/agents-realtime/dist/realtimeSession.mjs","../../lib/vowel/providers/OpenAIRealtimeProvider.ts","../../lib/vowel/providers/GrokRealtimeWebSocketTransport.ts","../../lib/vowel/providers/normalizeRealtimeEvent.ts","../../lib/vowel/providers/grokNormalizingWebSocket.ts","../../lib/vowel/providers/WebSocketRealtimeProviderBase.ts","../../lib/vowel/providers/GrokRealtimeProvider.ts","../../lib/vowel/utils/vowel-prime-urls.ts","../../lib/vowel/providers/VowelPrimeRealtimeProvider.ts","../../lib/vowel/providers/VowelCoreRealtimeProvider.ts","../../lib/vowel/types/providers.ts","../../lib/vowel/providers/RealtimeProviderFactory.ts","../../lib/vowel/utils/connectionIdentity.ts","../../lib/vowel/managers/SessionManager.ts","../../lib/vowel/managers/TypingSoundManager.ts","../../lib/vowel/managers/FloatingCursorManager.ts","../../lib/vowel/core/action-notifier.ts","../../lib/vowel/ui/border-glow.ts","../../lib/vowel/ui/FloatingActionPillManager.ts","../../lib/vowel/utils/darkMode.ts","../../lib/vowel/utils/deprecation.ts","../../lib/vowel/webmcp/conversion.ts","../../lib/vowel/webmcp/index.ts","../../lib/vowel/core/VowelClient.ts"],"sourcesContent":["/**\n * @fileoverview State Manager - Handles voice session state and listener notifications\n * \n * This file contains the `StateManager` class which provides reactive state management\n * for the Vowel client. It maintains the current session state, tracks speaking states\n * (user, AI thinking, AI speaking), manages transcripts, and notifies subscribers of\n * state changes.\n * \n * Responsibilities:\n * - Maintaining session connection state\n * - Tracking user and AI speaking states\n * - Managing conversation transcripts\n * - Notifying listeners of state changes\n * - Providing immutable state access\n * \n * @module @vowel.to/client/managers\n * @author vowel.to\n * @license Proprietary\n */\n\n/**\n * Voice session state\n */\nexport interface VoiceSessionState {\n isConnecting: boolean;\n isConnected: boolean;\n isDisconnecting: boolean; // True when disconnection process is in progress\n isResuming: boolean; // True when resuming a session from navigation\n status: string;\n transcripts: Array<{\n role: \"user\" | \"assistant\";\n text: string;\n timestamp: Date;\n }>;\n // Speaking state tracking\n isUserSpeaking: boolean; // User is actively speaking (detected by client-side VAD)\n isAIThinking: boolean; // AI is processing/generating response (tool calls, waiting for first response)\n isToolExecuting: boolean; // AI is executing a tool (different shade of yellow indicator)\n isAISpeaking: boolean; // AI is delivering audio response\n isHibernated: boolean; // Session is hibernated (STT stream closed, waiting for wake)\n // Error tracking\n error?: {\n message: string;\n details?: string | object;\n timestamp: Date;\n } | null;\n}\n\n/**\n * State change listener function\n */\nexport type StateChangeListener = (state: VoiceSessionState) => void;\n\n/**\n * State Manager class\n * Manages voice session state and notifies listeners of changes\n */\nexport class StateManager {\n private state: VoiceSessionState;\n private listeners: Set = new Set();\n\n constructor(initialState?: Partial) {\n this.state = {\n isConnecting: false,\n isConnected: false,\n isDisconnecting: false,\n isResuming: false,\n status: \"\",\n transcripts: [],\n isUserSpeaking: false,\n isAIThinking: false,\n isToolExecuting: false,\n isAISpeaking: false,\n isHibernated: false,\n error: null,\n ...initialState,\n };\n }\n\n /**\n * Get current state (returns a copy)\n */\n getState(): VoiceSessionState {\n return { ...this.state };\n }\n\n /**\n * Update state and notify listeners\n */\n updateState(updates: Partial): void {\n console.log(\"📊 [StateManager] State update:\", updates);\n this.state = { ...this.state, ...updates };\n this.notifyListeners();\n }\n\n /**\n * Add a transcript entry\n */\n addTranscript(transcript: {\n role: \"user\" | \"assistant\";\n text: string;\n timestamp: Date;\n }): void {\n this.state.transcripts.push(transcript);\n this.notifyListeners();\n }\n\n /**\n * Clear transcript history\n */\n clearTranscripts(): void {\n this.state.transcripts = [];\n this.notifyListeners();\n }\n\n /**\n * Subscribe to state changes\n * @returns Unsubscribe function\n */\n subscribe(listener: StateChangeListener): () => void {\n this.listeners.add(listener);\n return () => this.listeners.delete(listener);\n }\n\n /**\n * Notify all listeners of state change\n */\n private notifyListeners(): void {\n const stateCopy = this.getState();\n this.listeners.forEach((listener) => listener(stateCopy));\n }\n\n /**\n * Set user speaking state\n */\n setUserSpeaking(isSpeaking: boolean): void {\n if (this.state.isUserSpeaking !== isSpeaking) {\n console.log(`🎤 [StateManager] User speaking: ${isSpeaking}`);\n this.state.isUserSpeaking = isSpeaking;\n this.notifyListeners();\n }\n }\n\n /**\n * Set AI thinking state\n */\n setAIThinking(isThinking: boolean): void {\n if (this.state.isAIThinking !== isThinking) {\n console.log(`🧠 [StateManager] AI thinking: ${isThinking}`);\n this.state.isAIThinking = isThinking;\n this.notifyListeners();\n }\n }\n\n /**\n * Set tool executing state\n */\n setToolExecuting(isExecuting: boolean): void {\n if (this.state.isToolExecuting !== isExecuting) {\n console.log(`🔧 [StateManager] Tool executing: ${isExecuting}`);\n this.state.isToolExecuting = isExecuting;\n this.notifyListeners();\n }\n }\n\n /**\n * Set AI speaking state\n */\n setAISpeaking(isSpeaking: boolean): void {\n if (this.state.isAISpeaking !== isSpeaking) {\n console.log(`🔊 [StateManager] AI speaking: ${isSpeaking}`);\n this.state.isAISpeaking = isSpeaking;\n this.notifyListeners();\n }\n }\n\n /**\n * Set hibernated state\n */\n setHibernated(isHibernated: boolean): void {\n if (this.state.isHibernated !== isHibernated) {\n console.log(`💤 [StateManager] Hibernated: ${isHibernated}`);\n this.state.isHibernated = isHibernated;\n this.notifyListeners();\n }\n }\n\n /**\n * Clear error state\n */\n clearError(): void {\n if (this.state.error !== null) {\n this.state.error = null;\n this.notifyListeners();\n }\n }\n\n /**\n * Reset state to initial values\n */\n reset(): void {\n this.state = {\n isConnecting: false,\n isConnected: false,\n isDisconnecting: false,\n isResuming: false,\n status: \"\",\n transcripts: [],\n isUserSpeaking: false,\n isAIThinking: false,\n isToolExecuting: false,\n isAISpeaking: false,\n isHibernated: false,\n error: null,\n };\n this.notifyListeners();\n }\n\n /**\n * Export conversation state for restoration\n * Returns a serializable object containing transcripts\n * \n * @param options - Export options\n * @param options.maxTurns - Maximum number of conversation turns to include (default: all)\n * @returns Serializable state object\n * \n * @example\n * ```ts\n * const state = vowel.exportState({ maxTurns: 10 });\n * localStorage.setItem('vowel-state', JSON.stringify(state));\n * ```\n */\n exportState(options?: { maxTurns?: number }): VoiceSessionState {\n const state = this.getState();\n \n // Truncate transcripts if maxTurns specified\n if (options?.maxTurns && state.transcripts.length > options.maxTurns) {\n const startIndex = state.transcripts.length - options.maxTurns;\n state.transcripts = state.transcripts.slice(startIndex);\n }\n \n return state;\n }\n\n /**\n * Import conversation state from a previous session\n * Merges transcripts into current state\n * \n * @param savedState - Previously exported state\n * \n * @example\n * ```ts\n * const savedState = JSON.parse(localStorage.getItem('vowel-state'));\n * vowel.importState(savedState);\n * ```\n */\n importState(savedState: Partial): void {\n if (savedState.transcripts) {\n console.log(`📥 [StateManager] Importing ${savedState.transcripts.length} transcript(s)`);\n this.state.transcripts = [...savedState.transcripts];\n this.notifyListeners();\n }\n }\n}\n\n","export default \"/**\\n * @fileoverview Audio Worklet Processor for microphone input\\n * \\n * This AudioWorklet runs on a separate audio rendering thread, providing\\n * better performance and lower latency than the deprecated ScriptProcessorNode.\\n * It processes microphone input, converts to PCM16 format, and sends to main thread.\\n * \\n * @module @vowel.to/client/managers\\n * @author vowel.to\\n * @license Proprietary\\n */\\n\\n/**\\n * Base64 encoding table\\n * btoa() is not available in AudioWorklet context, so we implement our own\\n */\\nconst base64Chars = 'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/';\\n\\n/**\\n * Encode Uint8Array to base64 string\\n * Custom implementation for AudioWorklet context\\n * @param {Uint8Array} bytes - Bytes to encode\\n * @returns {string} Base64 encoded string\\n */\\nfunction encodeBase64(bytes) {\\n let result = '';\\n const len = bytes.length;\\n \\n for (let i = 0; i < len; i += 3) {\\n const byte1 = bytes[i];\\n const byte2 = i + 1 < len ? bytes[i + 1] : 0;\\n const byte3 = i + 2 < len ? bytes[i + 2] : 0;\\n \\n const encoded1 = byte1 >> 2;\\n const encoded2 = ((byte1 & 0x03) << 4) | (byte2 >> 4);\\n const encoded3 = ((byte2 & 0x0f) << 2) | (byte3 >> 6);\\n const encoded4 = byte3 & 0x3f;\\n \\n result += base64Chars[encoded1];\\n result += base64Chars[encoded2];\\n result += i + 1 < len ? base64Chars[encoded3] : '=';\\n result += i + 2 < len ? base64Chars[encoded4] : '=';\\n }\\n \\n return result;\\n}\\n\\n/**\\n * VowelAudioProcessor - processes microphone audio for streaming to Gemini Live\\n * Runs on the audio rendering thread for optimal performance\\n */\\nclass VowelAudioProcessor extends AudioWorkletProcessor {\\n /**\\n * @param {Object} options - Processor options\\n */\\n constructor(options) {\\n super();\\n \\n // Get sample rates from options\\n this.inputSampleRate = options.processorOptions?.inputSampleRate || 16000;\\n this.targetSampleRate = options.processorOptions?.targetSampleRate || 16000;\\n this.needsResampling = this.inputSampleRate !== this.targetSampleRate;\\n \\n // Buffer for accumulating samples before sending\\n // Configurable via processorOptions.bufferSize (see AUDIO_CAPTURE_CONFIG in constants.ts)\\n // 2048 = smaller chunks, more responsive; 4096 = larger chunks, original behavior\\n this.bufferSize = options.processorOptions?.bufferSize ?? 2048;\\n this.buffer = new Float32Array(this.bufferSize);\\n this.bufferIndex = 0;\\n \\n console.log(`🎤 AudioWorklet initialized: ${this.inputSampleRate}Hz → ${this.targetSampleRate}Hz (resampling: ${this.needsResampling}), buffer: ${this.bufferSize} samples`);\\n }\\n \\n /**\\n * Resample audio from input sample rate to target sample rate\\n * Uses linear interpolation for downsampling\\n * \\n * @param {Float32Array} inputData - Input audio samples\\n * @returns {Float32Array} Resampled audio\\n */\\n resampleAudio(inputData) {\\n if (!this.needsResampling) {\\n return inputData;\\n }\\n \\n const ratio = this.inputSampleRate / this.targetSampleRate;\\n const outputLength = Math.round(inputData.length / ratio);\\n const output = new Float32Array(outputLength);\\n \\n for (let i = 0; i < outputLength; i++) {\\n const position = i * ratio;\\n const index = Math.floor(position);\\n const fraction = position - index;\\n \\n if (index + 1 < inputData.length) {\\n // Linear interpolation\\n const sample1 = inputData[index];\\n const sample2 = inputData[index + 1];\\n output[i] = sample1 + (sample2 - sample1) * fraction;\\n } else {\\n output[i] = inputData[index];\\n }\\n }\\n \\n return output;\\n }\\n \\n /**\\n * Process audio samples from microphone\\n * Called automatically by the audio rendering thread\\n * \\n * @param {Float32Array[][]} inputs - Audio input buffers\\n * @param {Float32Array[][]} outputs - Audio output buffers (unused)\\n * @param {Record} parameters - Parameter values (unused)\\n * @returns {boolean} true to keep processor alive\\n */\\n process(inputs, outputs, parameters) {\\n const input = inputs[0];\\n \\n // No input data available\\n if (!input || !input[0]) {\\n return true;\\n }\\n\\n let pcmData = input[0]; // First channel (mono)\\n \\n // Resample if needed (e.g., 48kHz → 24kHz on Safari/Firefox)\\n if (this.needsResampling) {\\n pcmData = this.resampleAudio(pcmData);\\n }\\n \\n // Accumulate samples into buffer\\n for (let i = 0; i < pcmData.length; i++) {\\n this.buffer[this.bufferIndex++] = pcmData[i];\\n \\n // Buffer full? Send it!\\n if (this.bufferIndex >= this.bufferSize) {\\n this.sendBuffer();\\n this.bufferIndex = 0;\\n }\\n }\\n\\n return true; // Keep processor alive\\n }\\n \\n /**\\n * Convert accumulated buffer to PCM16 and send to main thread\\n */\\n sendBuffer() {\\n // Use configured bufferSize to ensure we always send full chunks\\n const samplesToSend = this.bufferSize;\\n \\n // Convert Float32Array to Int16Array for PCM16 encoding\\n const int16 = new Int16Array(samplesToSend);\\n for (let i = 0; i < samplesToSend; i++) {\\n // Convert float32 -1 to 1 to int16 -32768 to 32767\\n // Clamp to prevent overflow\\n let sample = Math.max(-1, Math.min(1, this.buffer[i]));\\n // Use asymmetric scaling: -32768 for negative, +32767 for positive\\n // This matches the Int16 range and prevents distortion\\n int16[i] = sample < 0 ? sample * 0x8000 : sample * 0x7fff;\\n }\\n\\n // Encode to base64 using custom encoder (btoa not available in AudioWorklet)\\n const uint8 = new Uint8Array(int16.buffer);\\n const base64Data = encodeBase64(uint8);\\n\\n // Send encoded audio to main thread\\n this.port.postMessage({\\n type: 'audio',\\n data: base64Data\\n });\\n }\\n}\\n\\n// Register the processor\\nregisterProcessor('vowel-audio-processor', VowelAudioProcessor);\\n\\n\"","/**\n * Vowel SDK Constants\n *\n * Note: This library uses HTTP endpoints for all API communication.\n * Convex is no longer used as of v0.2.0.\n */\n\n/**\n * Vowel Platform API URL\n *\n * This is the HTTP API endpoint for the Vowel platform.\n * All Vowel clients connect to this API for voice agent token generation.\n *\n * This URL is managed by the Vowel platform and should not be changed by developers.\n * Configured via VITE_VOWEL_PLATFORM_API_URL environment variable at build time.\n */\nexport const VOWEL_PLATFORM_API_URL = import.meta.env.VITE_VOWEL_PLATFORM_API_URL || 'https://wooden-herring-934.convex.site';\n\n/**\n * Vowel Platform Token Endpoint\n * \n * HTTP endpoint for generating ephemeral voice API tokens.\n * Supports multi-provider backend (Gemini Live, OpenAI Realtime).\n */\nexport const VOWEL_TOKEN_ENDPOINT = `${VOWEL_PLATFORM_API_URL}/vowel/api/generateToken`;\n\n/**\n * Voice agent configuration defaults\n */\nexport const DEFAULT_VOICE_CONFIG = {\n model: 'gemini-live-2.5-flash-preview',\n voice: 'Orus',\n language: 'en-US',\n} as const;\n\n/**\n * Audio configuration constants\n */\nexport const AUDIO_CONFIG = {\n inputSampleRate: 16000,\n outputSampleRate: 24000,\n encoding: 'LINEAR16',\n bufferSize: 256,\n} as const;\n\n/**\n * Audio capture chunking configuration for microphone streaming.\n *\n * Controls how audio is buffered and sent to the server. Smaller chunks =\n * more frequent sends, lower latency, better responsiveness for short phrases\n * (e.g. \"yes\", \"no\", \"okay\"). Larger chunks = fewer sends, slightly less overhead.\n *\n * @see AUDIO_CAPTURE_CONFIG.bufferSize - Samples per chunk at output sample rate (24kHz)\n */\n/**\n * Disable RTC loopback path for audio routing.\n * \n * When true, microphone audio is sent directly to the provider without going through\n * an RTCPeerConnection loopback, and TTS audio is played directly via \n * outputContext.destination instead of through the RTC loopback/audio element.\n * \n * This is disabled by default because Brave browser and Grok playback have unreliable\n * or broken audio through the RTC loopback path. The loopback requires specific\n * WebRTC/AudioWorklet behavior that these browsers don't handle correctly.\n * \n * Set to false to re-enable the RTC loopback path for echo cancellation in browsers\n * that properly support it (Chrome, Edge).\n */\nexport const DISABLE_RTC_LOOPBACK = true;\n\nexport const AUDIO_CAPTURE_CONFIG = {\n /**\n * Number of samples to accumulate before sending each audio chunk.\n * At 24kHz: 2048 samples ≈ 85ms, 4096 samples ≈ 170ms.\n *\n * - **2048** (default): Smaller chunks, ~2x more frequent. Matches Vowel Engine demo\n * behavior at 48kHz input. Better for short phrases and server VAD.\n * - **4096**: Larger chunks, original behavior. Slightly less overhead.\n */\n bufferSize: 2048,\n} as const;\n\n","/**\n * @fileoverview Audio Manager - Handles all audio-related functionality for voice sessions\n * \n * This file contains the `AudioManager` class which manages the complete audio pipeline\n * for voice interactions. It handles microphone input capture, audio encoding for streaming\n * to Gemini Live, audio decoding from AI responses, and playback through the browser's\n * audio system.\n * \n * Responsibilities:\n * - Microphone permission and stream setup\n * - Audio context initialization and management\n * - Real-time audio encoding (PCM16 format)\n * - Audio streaming to Gemini Live session\n * - AI audio response decoding and playback\n * - Audio cleanup and resource management\n * - Speaking state detection and callbacks\n * \n * @module @vowel.to/client/managers\n * @author vowel.to\n * @license Proprietary\n */\n\nimport type { RealtimeProvider } from \"../providers/RealtimeProvider\";\nimport type { EnhancedVADManager } from \"./EnhancedVADManager\";\n// Import worklet as raw JavaScript string for inline bundling\n// Using .js file to avoid TypeScript syntax issues in AudioWorklet context\n// @ts-ignore - Vite will handle this\nimport audioWorkletCode from './audio-processor.worklet.js?raw';\nimport { getOperatingSystem } from \"../utils/device-detection\";\nimport { AUDIO_CAPTURE_CONFIG, DISABLE_RTC_LOOPBACK } from \"../types/constants\";\n\n/**\n * Audio context and node references\n */\ninterface AudioRefs {\n inputContext: AudioContext | null;\n outputContext: AudioContext | null;\n inputNode: GainNode | null;\n outputNode: GainNode | null;\n workletNode: AudioWorkletNode | null;\n sourceNode: MediaStreamAudioSourceNode | null;\n mediaStream: MediaStream | null;\n // RTC loopback for echo cancellation\n rtcConnection: RTCPeerConnection | null; // CLIENT side (sends mic, receives TTS)\n rtcLoopbackConnection: RTCPeerConnection | null; // SERVER side (receives mic, sends TTS)\n micLoopbackStream: MediaStream | null; // Mic audio from SERVER receiver side\n ttsLoopbackStream: MediaStream | null; // TTS audio from CLIENT receiver side\n loopbackAudioElement: HTMLAudioElement | null;\n mediaStreamDestination: MediaStreamAudioDestinationNode | null;\n}\n\n/**\n * Audio Manager configuration\n */\nexport interface AudioManagerConfig {\n /**\n * Called when AI speaking state changes\n */\n onAISpeakingChange?: (isSpeaking: boolean) => void;\n}\n\n/**\n * Audio Manager class\n * Handles microphone setup, audio streaming, and playback\n */\nexport class AudioManager {\n private refs: AudioRefs = {\n inputContext: null,\n outputContext: null,\n inputNode: null,\n outputNode: null,\n workletNode: null,\n sourceNode: null,\n mediaStream: null,\n // RTC loopback for echo cancellation\n rtcConnection: null, // CLIENT side\n rtcLoopbackConnection: null, // SERVER side\n micLoopbackStream: null, // Mic from SERVER receiver\n ttsLoopbackStream: null, // TTS from CLIENT receiver\n loopbackAudioElement: null,\n mediaStreamDestination: null,\n };\n\n private nextStartTime = 0;\n private audioSources: Set = new Set();\n private typingSoundSources: Set = new Set();\n private providerRef: RealtimeProvider | null = null;\n private config: AudioManagerConfig;\n private isAISpeaking: boolean = false;\n private playbackGeneration: number = 0; // Invalidates queued/in-flight audio from interrupted turns\n private provisionalInterruptPaused = false;\n private provisionalInterruptAudioQueue: Array = [];\n private provisionalInterruptResumeTimer: ReturnType | null = null;\n private isMuted: boolean = false;\n private selectedDeviceId: string | null = null;\n private currentDevice: MediaDeviceInfo | null = null;\n private enhancedVADManager: EnhancedVADManager | null = null;\n private frameTimestamp: number = 0;\n private clientVADAudioBuffer: string[] = []; // Buffer original provider-rate audio chunks when client VAD is active\n private rollingAudioBuffer: string[] = []; // Rolling buffer for pre-speech audio capture\n private rollingAudioBufferMaxSize: number = 150; // ~1.5 seconds at 100 chunks/sec (10ms chunks)\n private isCurrentlySpeaking: boolean = false; // Track speech state to detect transitions\n\n constructor(config: AudioManagerConfig = {}) {\n this.config = config;\n }\n\n private async createOutputAudioContext(sampleRate: number): Promise {\n this.refs.outputContext = new (window.AudioContext ||\n (window as any).webkitAudioContext)({\n sampleRate,\n });\n\n // Log audio output device info\n if (typeof navigator !== 'undefined' && navigator.mediaDevices?.enumerateDevices) {\n try {\n const devices = await navigator.mediaDevices.enumerateDevices();\n const audioOutputs = devices.filter(d => d.kind === 'audiooutput');\n console.log(`🔊 [AudioManager] Available audio output devices:`);\n audioOutputs.forEach((device, i) => {\n console.log(` ${i + 1}. ${device.label || 'Unknown'} (deviceId: ${device.deviceId})`);\n });\n const defaultOutput = audioOutputs.find(d => d.deviceId === 'default') || audioOutputs[0];\n if (defaultOutput) {\n console.log(` → Using: ${defaultOutput.label || defaultOutput.deviceId}`);\n }\n } catch (e) {\n console.log(`🔊 [AudioManager] Could not enumerate audio devices:`, e);\n }\n }\n\n if (this.refs.outputContext.state === 'suspended') {\n await this.refs.outputContext.resume();\n console.log(`✅ Output AudioContext resumed @ ${sampleRate}Hz`);\n }\n\n this.refs.outputNode = this.refs.outputContext.createGain();\n this.refs.mediaStreamDestination = this.refs.outputContext.createMediaStreamDestination();\n this.refs.outputNode.connect(this.refs.mediaStreamDestination);\n this.nextStartTime = this.refs.outputContext.currentTime;\n }\n\n private async ensureOutputAudioContext(sampleRate: number): Promise {\n if (\n this.refs.outputContext?.sampleRate === sampleRate &&\n this.refs.outputNode &&\n this.refs.mediaStreamDestination\n ) {\n return;\n }\n\n if (this.audioSources.size > 0) {\n this.stopAllAudio();\n }\n\n if (this.refs.outputContext) {\n try {\n await this.refs.outputContext.close();\n } catch (error) {\n console.warn(\"⚠️ Failed to close previous output AudioContext:\", error);\n }\n }\n\n this.refs.outputContext = null;\n this.refs.outputNode = null;\n this.refs.mediaStreamDestination = null;\n\n await this.createOutputAudioContext(sampleRate);\n console.log(`✅ Output audio context initialized @ ${sampleRate}Hz`);\n }\n\n /**\n * Initialize audio contexts for input and output\n * Also loads the AudioWorklet processor module\n * \n * CRITICAL for iOS: This must be called synchronously within a user gesture handler.\n * iOS Safari requires AudioContext creation and resume() to happen within the same\n * user gesture event handler. Any async operations between gesture and resume()\n * will cause the AudioContext to remain suspended.\n */\n async initAudio(): Promise {\n // Don't specify sampleRate for input - let browser use hardware default\n // This avoids sample rate mismatch errors on Safari/Firefox\n // The AudioWorklet will handle resampling to 16kHz for Gemini\n this.refs.inputContext = new (window.AudioContext ||\n (window as any).webkitAudioContext)();\n\n await this.createOutputAudioContext(24000);\n\n // CRITICAL: Resume AudioContext immediately after creation (within user gesture)\n // iOS Safari requires this to happen synchronously in the user gesture handler\n // If we wait, the context will remain suspended and getUserMedia will fail\n try {\n if (this.refs.inputContext.state === 'suspended') {\n await this.refs.inputContext.resume();\n console.log(\"✅ Input AudioContext resumed (iOS compatibility)\");\n }\n if (this.refs.outputContext?.state === 'suspended') {\n await this.refs.outputContext.resume();\n console.log(\"✅ Output AudioContext resumed (iOS compatibility)\");\n }\n } catch (error) {\n console.warn(\"⚠️ Failed to resume AudioContext immediately:\", error);\n // Continue anyway - resume() will be called again in setupMicrophone()\n }\n\n this.refs.inputNode = this.refs.inputContext.createGain();\n console.log(\"✅ MediaStreamDestination created (RTC loopback will be set up in setupMicrophone)\");\n\n // Load AudioWorklet processor module\n // The worklet runs on a separate audio rendering thread for better performance\n try {\n // Check AudioWorklet support\n if (!this.refs.inputContext.audioWorklet) {\n throw new Error(\n \"AudioWorklet is not supported in this browser. \" +\n \"AudioWorklet requires iOS 14.5+ or modern desktop browsers.\"\n );\n }\n\n let workletUrl: string;\n let needsCleanup = false;\n \n // Check if running in a Chrome extension context\n const isExtension = typeof chrome !== 'undefined' && chrome.runtime && chrome.runtime.id;\n const isIOS = getOperatingSystem() === 'ios';\n \n if (isExtension) {\n // In extension: use web_accessible_resource\n workletUrl = chrome.runtime.getURL('audio-processor.worklet.js');\n console.log('🔌 Loading audio worklet from extension:', workletUrl);\n } else {\n // In regular web page: use blob URL\n // iOS Safari supports blob URLs for AudioWorklet (iOS 14.5+)\n const blob = new Blob([audioWorkletCode], { type: 'application/javascript' });\n workletUrl = URL.createObjectURL(blob);\n needsCleanup = true;\n console.log(`🌐 Loading audio worklet from blob${isIOS ? ' (iOS)' : ''}`);\n }\n \n // Load the worklet module\n // On iOS, this must happen within the user gesture handler chain\n // (which it does, since initAudio() is called from startSession() which is called from user gesture)\n await this.refs.inputContext.audioWorklet.addModule(workletUrl);\n \n // Clean up the blob URL if we created one\n // Note: We can revoke immediately after addModule() resolves - the browser keeps a reference\n if (needsCleanup) {\n URL.revokeObjectURL(workletUrl);\n }\n \n console.log(\"✅ Audio contexts initialized with AudioWorklet\");\n } catch (error: any) {\n console.error(\"❌ Failed to load AudioWorklet:\", error);\n const isIOS = getOperatingSystem() === 'ios';\n \n if (isIOS) {\n // Provide iOS-specific error message\n if (error.message?.includes('not supported') || error.name === 'NotSupportedError') {\n throw new Error(\n \"AudioWorklet is not supported on this iOS version. \" +\n \"Please update to iOS 14.5 or later, or use a different browser.\"\n );\n } else if (error.message?.includes('user gesture') || error.name === 'InvalidStateError') {\n throw new Error(\n \"AudioWorklet initialization failed. On iOS, startSession() must be called \" +\n \"directly from a user gesture handler (click/touch event).\"\n );\n }\n }\n \n throw new Error(`Failed to initialize audio worklet: ${error.message || error}`);\n }\n }\n\n /**\n * Set up RTC loopback connection for echo cancellation\n * \n * CRITICAL: This routes BOTH microphone input AND TTS output through the SAME\n * RTCPeerConnection loopback, simulating a real WebRTC peer-to-peer call.\n * This is essential for Chrome's Acoustic Echo Cancellation (AEC) to work.\n * \n * Architecture:\n * - Mic → RTC Connection (sender) → RTC Loopback Connection (receiver) → MediaStream → Web Audio API → Server\n * - TTS → Web Audio API → MediaStreamDestination → RTC Connection (sender) → RTC Loopback Connection (receiver) → Audio Element → Speakers\n * \n * Both streams go through the SAME RTCPeerConnection, allowing Chrome's AEC to\n * correlate input/output and cancel echo.\n * \n * @param micStream - The microphone MediaStream to add to the loopback\n * \n * Reference: https://gist.github.com/alexciarlillo/4b9f75516f93c10d7b39282d10cd17bc\n */\n private async setupRTCLoopback(micStream: MediaStream): Promise {\n try {\n console.log(\"🔄 Setting up RTC loopback for echo cancellation...\");\n console.log(\" Architecture: Mic + TTS → SAME RTC loopback → Output (simulating WebRTC peer call)\");\n\n if (!this.refs.mediaStreamDestination) {\n throw new Error(\"MediaStreamDestination not initialized\");\n }\n\n // Create two RTCPeerConnections for loopback (simulating LiveKit client ↔ server)\n // rtcConnection = CLIENT side (sends mic → server, receives TTS ← server)\n // rtcLoopbackConnection = SERVER side (receives mic ← client, sends TTS → client)\n this.refs.rtcConnection = new RTCPeerConnection();\n this.refs.rtcLoopbackConnection = new RTCPeerConnection();\n this.refs.micLoopbackStream = new MediaStream(); // SERVER receives mic from CLIENT\n this.refs.ttsLoopbackStream = new MediaStream(); // CLIENT receives TTS from SERVER\n\n console.log(\"✅ RTCPeerConnection pair created (simulating LiveKit client ↔ server)\");\n\n // Set up ICE candidate exchange between the two connections\n this.refs.rtcConnection.onicecandidate = (e) => {\n if (e.candidate && this.refs.rtcLoopbackConnection) {\n this.refs.rtcLoopbackConnection.addIceCandidate(new RTCIceCandidate(e.candidate));\n }\n };\n\n this.refs.rtcLoopbackConnection.onicecandidate = (e) => {\n if (e.candidate && this.refs.rtcConnection) {\n this.refs.rtcConnection.addIceCandidate(new RTCIceCandidate(e.candidate));\n }\n };\n\n // SERVER side: Receive mic audio from CLIENT\n // This simulates the server receiving microphone audio from the client\n // All audio tracks received by rtcLoopbackConnection are mic audio (from client)\n this.refs.rtcLoopbackConnection.ontrack = (e) => {\n if (e.streams[0] && this.refs.micLoopbackStream) {\n e.streams[0].getAudioTracks().forEach((track) => {\n this.refs.micLoopbackStream!.addTrack(track);\n console.log(`✅ [SERVER] Received mic track from CLIENT: ${track.label || 'audio'}`);\n });\n }\n };\n\n // CLIENT side: Receive TTS audio from SERVER\n // This simulates the client receiving TTS audio from the server\n // All audio tracks received by rtcConnection are TTS audio (from server)\n this.refs.rtcConnection.ontrack = (e) => {\n if (e.streams[0] && this.refs.ttsLoopbackStream) {\n e.streams[0].getAudioTracks().forEach((track) => {\n this.refs.ttsLoopbackStream!.addTrack(track);\n console.log(`✅ [CLIENT] Received TTS track from SERVER: ${track.label || 'audio'}`);\n });\n }\n };\n\n // CLIENT SIDE: Add microphone stream to rtcConnection (client → server)\n // This simulates the client sending mic audio to the server\n console.log(\"🎤 [CLIENT] Adding microphone stream to RTC connection (client → server)...\");\n micStream.getTracks().forEach((track) => {\n this.refs.rtcConnection!.addTrack(track, micStream);\n });\n console.log(\"✅ Microphone stream added (client sending to server)\");\n\n // SERVER SIDE: Add TTS output stream to rtcLoopbackConnection (server → client)\n // This simulates the server sending TTS audio to the client\n console.log(\"🔊 [SERVER] Adding TTS output stream to RTC loopback connection (server → client)...\");\n const outputStream = this.refs.mediaStreamDestination.stream;\n outputStream.getTracks().forEach((track) => {\n this.refs.rtcLoopbackConnection!.addTrack(track, outputStream);\n });\n console.log(\"✅ TTS output stream added (server sending to client)\");\n\n // Create offer/answer for loopback\n // CLIENT side (rtcConnection) creates offer - wants to send mic and receive TTS\n // SERVER side (rtcLoopbackConnection) creates answer - wants to receive mic and send TTS\n const offerOptions = {\n offerToReceiveAudio: true, // CLIENT wants to receive TTS from server\n offerToReceiveVideo: false,\n };\n\n const offer = await this.refs.rtcConnection.createOffer(offerOptions);\n await this.refs.rtcConnection.setLocalDescription(offer);\n\n await this.refs.rtcLoopbackConnection.setRemoteDescription(offer);\n const answer = await this.refs.rtcLoopbackConnection.createAnswer();\n await this.refs.rtcLoopbackConnection.setLocalDescription(answer);\n\n await this.refs.rtcConnection.setRemoteDescription(answer);\n\n // Wait for ICE connection to be established\n const iceSuccess = await this.waitForICEConnection();\n if (!iceSuccess) {\n console.warn(\"⚠️ ICE connection not established, AEC may not work\");\n }\n\n // Create audio element to play TTS from CLIENT receiver side\n // CRITICAL: Chrome's AEC requires the output to be connected to an unmuted audio element\n // According to https://github.com/twilio/twilio-video.js/issues/323 and the RTC loopback hack,\n // the stream MUST be fed into an