@speechos/core 0.2.0 → 0.2.2

package/dist/audio-capture.d.cts

@@ -0,0 +1,130 @@
+/**
+ * Audio capture module for SpeechOS WebSocket integration.
+ *
+ * Provides MediaRecorder-based audio capture with:
+ * - Format detection for cross-browser compatibility
+ * - Buffering for instant start (audio captured before connection is ready)
+ * - Atomic buffer swap pattern to prevent chunk reordering
+ */
+/**
+ * Supported audio formats with their MIME types and whether
+ * Deepgram needs explicit encoding parameters.
+ */
+export interface AudioFormat {
+    /** MIME type for MediaRecorder */
+    mimeType: string;
+    /** Short identifier for the format */
+    format: 'webm' | 'mp4' | 'pcm';
+    /** Whether Deepgram needs encoding/sample_rate params */
+    needsEncodingParams: boolean;
+}
+/**
+ * Detect the best supported audio format for the current browser.
+ *
+ * IMPORTANT: Safari must use MP4/AAC. Its WebM/Opus implementation is buggy
+ * and produces truncated/incomplete audio.
+ */
+export declare function getSupportedAudioFormat(): AudioFormat;
+/**
+ * Callback for receiving audio chunks.
+ */
+export type AudioChunkCallback = (chunk: Blob) => void;
+/**
+ * Audio capture manager with buffering support.
+ *
+ * Usage:
+ * 1. Create instance with onChunk callback
+ * 2. Call start() - immediately begins capturing
+ * 3. Call setReady() when connection is established - flushes buffer
+ * 4. Call stop() when done
+ */
+export declare class AudioCapture {
+    private mediaStream;
+    private recorder;
+    private buffer;
+    private isReady;
+    private isRecording;
+    private onChunk;
+    private audioFormat;
+    private deviceId;
+    /**
+     * Time slice for MediaRecorder in milliseconds.
+     *
+     * Safari requires a larger timeslice (1000ms) to properly flush its internal
+     * audio buffers. Smaller values cause Safari to drop or truncate audio data.
+     * See: https://community.openai.com/t/whisper-problem-with-audio-mp4-blobs-from-safari/
+     *
+     * Other browsers (Chrome, Firefox, Edge) work well with smaller timeslices
+     * which provide lower latency for real-time transcription.
+     */
+    private static readonly TIME_SLICE_MS;
+    private static readonly SAFARI_TIME_SLICE_MS;
+    /**
+     * @param onChunk - Callback for receiving audio chunks
+     * @param deviceId - Optional audio device ID (empty string or undefined for system default)
+     */
+    constructor(onChunk: AudioChunkCallback, deviceId?: string);
+    /**
+     * Get the appropriate timeslice for the current browser.
+     * Safari needs a larger timeslice to avoid dropping audio data.
+     */
+    private getTimeSlice;
+    /**
+     * Get the timeslice being used (in milliseconds).
+     * Useful for callers that need to wait for audio processing.
+     */
+    getTimeSliceMs(): number;
+    /**
+     * Get the audio format being used.
+     */
+    getFormat(): AudioFormat;
+    /**
+     * Start capturing audio immediately.
+     *
+     * Audio chunks will be buffered until setReady() is called.
+     */
+    start(): Promise<void>;
+    /**
+     * Handle an audio chunk with atomic buffer swap pattern.
+     *
+     * If not ready: buffer the chunk.
+     * If ready: send directly via callback.
+     */
+    private handleChunk;
+    /**
+     * Mark the capture as ready (connection established).
+     *
+     * This flushes any buffered chunks and switches to direct mode.
+     * Uses atomic swap to prevent chunk reordering.
+     */
+    setReady(): void;
+    /**
+     * Stop capturing audio and wait for final chunk.
+     *
+     * Uses requestData() before stop() to force the MediaRecorder to flush
+     * any buffered audio immediately. This is critical for Safari which
+     * may hold audio data in internal buffers.
+     *
+     * Safari requires an additional delay after stopping to ensure all audio
+     * from its internal encoding pipeline has been fully processed and emitted.
+     */
+    stop(): Promise<void>;
+    /**
+     * Check if currently recording.
+     */
+    get recording(): boolean;
+    /**
+     * Check if ready (connection established, direct mode active).
+     */
+    get ready(): boolean;
+    /**
+     * Get the number of buffered chunks waiting to be sent.
+     */
+    get bufferedChunks(): number;
+}
+/**
+ * Factory function to create an AudioCapture instance.
+ * @param onChunk - Callback for receiving audio chunks
+ * @param deviceId - Optional audio device ID (empty string or undefined for system default)
+ */
+export declare function createAudioCapture(onChunk: AudioChunkCallback, deviceId?: string): AudioCapture;

package/dist/audio-capture.d.ts

@@ -0,0 +1,130 @@
+/**
+ * Audio capture module for SpeechOS WebSocket integration.
+ *
+ * Provides MediaRecorder-based audio capture with:
+ * - Format detection for cross-browser compatibility
+ * - Buffering for instant start (audio captured before connection is ready)
+ * - Atomic buffer swap pattern to prevent chunk reordering
+ */
+/**
+ * Supported audio formats with their MIME types and whether
+ * Deepgram needs explicit encoding parameters.
+ */
+export interface AudioFormat {
+    /** MIME type for MediaRecorder */
+    mimeType: string;
+    /** Short identifier for the format */
+    format: 'webm' | 'mp4' | 'pcm';
+    /** Whether Deepgram needs encoding/sample_rate params */
+    needsEncodingParams: boolean;
+}
+/**
+ * Detect the best supported audio format for the current browser.
+ *
+ * IMPORTANT: Safari must use MP4/AAC. Its WebM/Opus implementation is buggy
+ * and produces truncated/incomplete audio.
+ */
+export declare function getSupportedAudioFormat(): AudioFormat;
+/**
+ * Callback for receiving audio chunks.
+ */
+export type AudioChunkCallback = (chunk: Blob) => void;
+/**
+ * Audio capture manager with buffering support.
+ *
+ * Usage:
+ * 1. Create instance with onChunk callback
+ * 2. Call start() - immediately begins capturing
+ * 3. Call setReady() when connection is established - flushes buffer
+ * 4. Call stop() when done
+ */
+export declare class AudioCapture {
+    private mediaStream;
+    private recorder;
+    private buffer;
+    private isReady;
+    private isRecording;
+    private onChunk;
+    private audioFormat;
+    private deviceId;
+    /**
+     * Time slice for MediaRecorder in milliseconds.
+     *
+     * Safari requires a larger timeslice (1000ms) to properly flush its internal
+     * audio buffers. Smaller values cause Safari to drop or truncate audio data.
+     * See: https://community.openai.com/t/whisper-problem-with-audio-mp4-blobs-from-safari/
+     *
+     * Other browsers (Chrome, Firefox, Edge) work well with smaller timeslices
+     * which provide lower latency for real-time transcription.
+     */
+    private static readonly TIME_SLICE_MS;
+    private static readonly SAFARI_TIME_SLICE_MS;
+    /**
+     * @param onChunk - Callback for receiving audio chunks
+     * @param deviceId - Optional audio device ID (empty string or undefined for system default)
+     */
+    constructor(onChunk: AudioChunkCallback, deviceId?: string);
+    /**
+     * Get the appropriate timeslice for the current browser.
+     * Safari needs a larger timeslice to avoid dropping audio data.
+     */
+    private getTimeSlice;
+    /**
+     * Get the timeslice being used (in milliseconds).
+     * Useful for callers that need to wait for audio processing.
+     */
+    getTimeSliceMs(): number;
+    /**
+     * Get the audio format being used.
+     */
+    getFormat(): AudioFormat;
+    /**
+     * Start capturing audio immediately.
+     *
+     * Audio chunks will be buffered until setReady() is called.
+     */
+    start(): Promise<void>;
+    /**
+     * Handle an audio chunk with atomic buffer swap pattern.
+     *
+     * If not ready: buffer the chunk.
+     * If ready: send directly via callback.
+     */
+    private handleChunk;
+    /**
+     * Mark the capture as ready (connection established).
+     *
+     * This flushes any buffered chunks and switches to direct mode.
+     * Uses atomic swap to prevent chunk reordering.
+     */
+    setReady(): void;
+    /**
+     * Stop capturing audio and wait for final chunk.
+     *
+     * Uses requestData() before stop() to force the MediaRecorder to flush
+     * any buffered audio immediately. This is critical for Safari which
+     * may hold audio data in internal buffers.
+     *
+     * Safari requires an additional delay after stopping to ensure all audio
+     * from its internal encoding pipeline has been fully processed and emitted.
+     */
+    stop(): Promise<void>;
+    /**
+     * Check if currently recording.
+     */
+    get recording(): boolean;
+    /**
+     * Check if ready (connection established, direct mode active).
+     */
+    get ready(): boolean;
+    /**
+     * Get the number of buffered chunks waiting to be sent.
+     */
+    get bufferedChunks(): number;
+}
+/**
+ * Factory function to create an AudioCapture instance.
+ * @param onChunk - Callback for receiving audio chunks
+ * @param deviceId - Optional audio device ID (empty string or undefined for system default)
+ */
+export declare function createAudioCapture(onChunk: AudioChunkCallback, deviceId?: string): AudioCapture;

package/dist/backend.d.cts

@@ -0,0 +1,41 @@
+/**
+ * Backend abstraction for voice sessions.
+ *
+ * Provides a unified interface for voice backends.
+ * Currently always uses WebSocket backend.
+ */
+import type { CommandDefinition, CommandResult, VoiceSessionOptions } from './types.js';
+/**
+ * Voice backend interface - common methods between backends
+ */
+export interface VoiceBackend {
+    startVoiceSession(options?: VoiceSessionOptions): Promise<void>;
+    stopVoiceSession(): Promise<string>;
+    requestEditText(originalText: string): Promise<string>;
+    requestCommand(commands: CommandDefinition[]): Promise<CommandResult | null>;
+    disconnect(): Promise<void>;
+    isConnected(): boolean;
+    /** Get the last input text (transcript) from a command result */
+    getLastInputText?(): string | undefined;
+    prefetchToken?(): Promise<unknown>;
+    startAutoRefresh?(): void;
+    stopAutoRefresh?(): void;
+    invalidateTokenCache?(): void;
+}
+/**
+ * Get the active voice backend.
+ * Always returns WebSocket backend (LiveKit is legacy).
+ *
+ * @returns The websocket backend
+ */
+export declare function getBackend(): VoiceBackend;
+/**
+ * Check if the current backend is LiveKit.
+ * @deprecated Always returns false - LiveKit is legacy
+ */
+export declare function isLiveKitBackend(): boolean;
+/**
+ * Check if the current backend is WebSocket.
+ * @deprecated Always returns true - WebSocket is the only backend
+ */
+export declare function isWebSocketBackend(): boolean;

package/dist/backend.d.ts

@@ -0,0 +1,41 @@
+/**
+ * Backend abstraction for voice sessions.
+ *
+ * Provides a unified interface for voice backends.
+ * Currently always uses WebSocket backend.
+ */
+import type { CommandDefinition, CommandResult, VoiceSessionOptions } from './types.js';
+/**
+ * Voice backend interface - common methods between backends
+ */
+export interface VoiceBackend {
+    startVoiceSession(options?: VoiceSessionOptions): Promise<void>;
+    stopVoiceSession(): Promise<string>;
+    requestEditText(originalText: string): Promise<string>;
+    requestCommand(commands: CommandDefinition[]): Promise<CommandResult | null>;
+    disconnect(): Promise<void>;
+    isConnected(): boolean;
+    /** Get the last input text (transcript) from a command result */
+    getLastInputText?(): string | undefined;
+    prefetchToken?(): Promise<unknown>;
+    startAutoRefresh?(): void;
+    stopAutoRefresh?(): void;
+    invalidateTokenCache?(): void;
+}
+/**
+ * Get the active voice backend.
+ * Always returns WebSocket backend (LiveKit is legacy).
+ *
+ * @returns The websocket backend
+ */
+export declare function getBackend(): VoiceBackend;
+/**
+ * Check if the current backend is LiveKit.
+ * @deprecated Always returns false - LiveKit is legacy
+ */
+export declare function isLiveKitBackend(): boolean;
+/**
+ * Check if the current backend is WebSocket.
+ * @deprecated Always returns true - WebSocket is the only backend
+ */
+export declare function isWebSocketBackend(): boolean;

package/dist/config.d.cts

@@ -1,30 +1,35 @@
 /**
- * Configuration management for SpeechOS SDK
+ * Configuration management for SpeechOS Core SDK
  */
-import type { SpeechOSConfig } from "./types.js";
+import type { SpeechOSCoreConfig } from "./types.js";
 /**
  * Default host - can be overridden by SPEECHOS_HOST env var at build time
  */
 export declare const DEFAULT_HOST: string;
 /**
- * Default configuration values
+ * Configuration with defaults applied (all fields required internally)
  */
-export declare const defaultConfig: Required<SpeechOSConfig>;
+interface ResolvedConfig {
+    apiKey: string;
+    userId: string;
+    host: string;
+    debug: boolean;
+}
 /**
  * Validates and merges user config with defaults
  * @param userConfig - User-provided configuration
  * @returns Validated and merged configuration
  */
-export declare function validateConfig(userConfig?: SpeechOSConfig): Required<SpeechOSConfig>;
+export declare function validateConfig(userConfig: SpeechOSCoreConfig): ResolvedConfig;
 /**
  * Get the current configuration
  */
-export declare function getConfig(): Required<SpeechOSConfig>;
+export declare function getConfig(): ResolvedConfig;
 /**
  * Set the current configuration
  * @param config - Configuration to set
  */
-export declare function setConfig(config: SpeechOSConfig): void;
+export declare function setConfig(config: SpeechOSCoreConfig): void;
 /**
  * Reset configuration to defaults
  */
@@ -34,3 +39,4 @@ export declare function resetConfig(): void;
  * @param userId - The user identifier to set
  */
 export declare function updateUserId(userId: string): void;
+export {};

package/dist/config.d.ts

@@ -1,30 +1,35 @@
 /**
- * Configuration management for SpeechOS SDK
+ * Configuration management for SpeechOS Core SDK
  */
-import type { SpeechOSConfig } from "./types.js";
+import type { SpeechOSCoreConfig } from "./types.js";
 /**
  * Default host - can be overridden by SPEECHOS_HOST env var at build time
  */
 export declare const DEFAULT_HOST: string;
 /**
- * Default configuration values
+ * Configuration with defaults applied (all fields required internally)
  */
-export declare const defaultConfig: Required<SpeechOSConfig>;
+interface ResolvedConfig {
+    apiKey: string;
+    userId: string;
+    host: string;
+    debug: boolean;
+}
 /**
  * Validates and merges user config with defaults
  * @param userConfig - User-provided configuration
  * @returns Validated and merged configuration
  */
-export declare function validateConfig(userConfig?: SpeechOSConfig): Required<SpeechOSConfig>;
+export declare function validateConfig(userConfig: SpeechOSCoreConfig): ResolvedConfig;
 /**
  * Get the current configuration
  */
-export declare function getConfig(): Required<SpeechOSConfig>;
+export declare function getConfig(): ResolvedConfig;
 /**
  * Set the current configuration
  * @param config - Configuration to set
  */
-export declare function setConfig(config: SpeechOSConfig): void;
+export declare function setConfig(config: SpeechOSCoreConfig): void;
 /**
  * Reset configuration to defaults
  */
@@ -34,3 +39,4 @@ export declare function resetConfig(): void;
  * @param userId - The user identifier to set
  */
 export declare function updateUserId(userId: string): void;
+export {};