@dtelecom/agents-js 0.1.14 → 0.1.16

package/dist/providers/index.d.mts

@@ -1,4 +1,4 @@
-import { h as STTPlugin, i as STTStreamOptions, S as STTStream, L as LLMPlugin, M as Message, e as LLMChunk, j as TTSPlugin } from '../types-BVMiP1bW.mjs';
+import { h as STTPlugin, i as STTStreamOptions, S as STTStream, L as LLMPlugin, M as Message, e as LLMChunk, j as TTSPlugin } from '../types-MPHcuMhp.mjs';
 import '@dtelecom/server-sdk-node';
 
 /**
@@ -200,4 +200,87 @@ declare class DeepgramTTS implements TTSPlugin {
     private ensureConnection;
 }
 
-export { CartesiaTTS, type CartesiaTTSOptions, DeepgramSTT, type DeepgramSTTOptions, DeepgramTTS, type DeepgramTTSOptions, OpenRouterLLM, type OpenRouterLLMOptions };
+/**
+ * DtelecomSTT — real-time streaming STT via dTelecom STT server (realtime-stt-m2).
+ *
+ * Protocol:
+ * - Connect to ws://<server>:<port> (address from options, no API key)
+ * - Send config: {"type":"config","language":"en"} (or "auto" for Parakeet auto-detect)
+ * - Wait for ready: {"type":"ready","client_id":"...","language":"en"}
+ * - Send audio as binary PCM16 16kHz mono frames
+ * - Receive transcriptions: {"type":"transcription","text":"...","is_final":true,"latency_ms":N}
+ * - Receive VAD events: {"type":"vad_event","event":"speech_start"|"speech_end"}
+ * - Keepalive via {"type":"ping"} / {"type":"pong"}
+ * - Mid-session reconfigure: send {"type":"config","language":"es","model":"whisper"} at any time
+ */
+
+interface DtelecomSTTOptions {
+    /** WebSocket server URL, e.g. "ws://192.168.1.100:8765" */
+    serverUrl: string;
+    /** Initial language (default: "auto" for Parakeet auto-detect) */
+    language?: string;
+    /** Force Whisper model even if Parakeet supports the language */
+    forceWhisper?: boolean;
+}
+declare class DtelecomSTT implements STTPlugin {
+    private readonly options;
+    constructor(options: DtelecomSTTOptions);
+    createStream(options?: STTStreamOptions): STTStream;
+}
+
+/**
+ * DtelecomTTS — real-time streaming TTS via dTelecom TTS server (realtime-tts-m2).
+ *
+ * Protocol:
+ * - Connect to ws://<server>:<port> (address from options, no API key)
+ * - Send config: {"config":{"voice":"af_heart","lang_code":"a","speed":1.0}}
+ * - Send text: {"text":"Hello world"} — uses config defaults
+ * - Send text with per-message override: {"text":"Hola","voice":"ef_dora","lang_code":"e","speed":1.0}
+ * - Receive: {"type":"generating","text":"..."} then binary PCM16 48kHz chunks, then {"type":"done"}
+ * - Cancel: {"type":"clear"} → {"type":"cleared"}
+ *
+ * Key differences from DeepgramTTS:
+ * - Single WebSocket connection (not per-language pool)
+ * - Per-message voice/language switching instead of separate connections
+ * - Server outputs 48kHz PCM16 (resampled from Kokoro's native 24kHz)
+ * - Uses SSML <lang> tags to route text segments to correct voice (same as DeepgramTTS)
+ */
+
+interface VoiceConfig {
+    voice: string;
+    langCode: string;
+}
+interface DtelecomTTSOptions {
+    /** WebSocket server URL, e.g. "ws://192.168.1.100:8766" */
+    serverUrl: string;
+    /** Voice config per language: { en: { voice: "af_heart", langCode: "a" }, es: { voice: "bf_emma", langCode: "b" } } */
+    voices: Record<string, VoiceConfig>;
+    /** Default language code (default: "en") */
+    defaultLanguage?: string;
+    /** Speech speed multiplier (default: 1.0) */
+    speed?: number;
+}
+declare class DtelecomTTS implements TTSPlugin {
+    private readonly serverUrl;
+    private readonly voices;
+    private readonly defaultLang;
+    private readonly speed;
+    private ws;
+    private connectPromise;
+    private flushState;
+    /** Default language code for untagged text (e.g. 'en'). */
+    get defaultLanguage(): string;
+    constructor(options: DtelecomTTSOptions);
+    /** Pre-connect WebSocket to TTS server. */
+    warmup(): Promise<void>;
+    /** Close WebSocket connection. */
+    close(): void;
+    /** Strip SSML lang tags from text for display/events. */
+    cleanText(text: string): string;
+    synthesize(text: string, signal?: AbortSignal): AsyncGenerator<Buffer>;
+    private synthesizeSegment;
+    /** Ensure a WebSocket connection exists and is open. */
+    private ensureConnection;
+}
+
+export { CartesiaTTS, type CartesiaTTSOptions, DeepgramSTT, type DeepgramSTTOptions, DeepgramTTS, type DeepgramTTSOptions, DtelecomSTT, type DtelecomSTTOptions, DtelecomTTS, type DtelecomTTSOptions, type VoiceConfig as DtelecomVoiceConfig, OpenRouterLLM, type OpenRouterLLMOptions };

package/dist/providers/index.d.ts

@@ -1,4 +1,4 @@
-import { h as STTPlugin, i as STTStreamOptions, S as STTStream, L as LLMPlugin, M as Message, e as LLMChunk, j as TTSPlugin } from '../types-BVMiP1bW.js';
+import { h as STTPlugin, i as STTStreamOptions, S as STTStream, L as LLMPlugin, M as Message, e as LLMChunk, j as TTSPlugin } from '../types-MPHcuMhp.js';
 import '@dtelecom/server-sdk-node';
 
 /**
@@ -200,4 +200,87 @@ declare class DeepgramTTS implements TTSPlugin {
     private ensureConnection;
 }
 
-export { CartesiaTTS, type CartesiaTTSOptions, DeepgramSTT, type DeepgramSTTOptions, DeepgramTTS, type DeepgramTTSOptions, OpenRouterLLM, type OpenRouterLLMOptions };
+/**
+ * DtelecomSTT — real-time streaming STT via dTelecom STT server (realtime-stt-m2).
+ *
+ * Protocol:
+ * - Connect to ws://<server>:<port> (address from options, no API key)
+ * - Send config: {"type":"config","language":"en"} (or "auto" for Parakeet auto-detect)
+ * - Wait for ready: {"type":"ready","client_id":"...","language":"en"}
+ * - Send audio as binary PCM16 16kHz mono frames
+ * - Receive transcriptions: {"type":"transcription","text":"...","is_final":true,"latency_ms":N}
+ * - Receive VAD events: {"type":"vad_event","event":"speech_start"|"speech_end"}
+ * - Keepalive via {"type":"ping"} / {"type":"pong"}
+ * - Mid-session reconfigure: send {"type":"config","language":"es","model":"whisper"} at any time
+ */
+
+interface DtelecomSTTOptions {
+    /** WebSocket server URL, e.g. "ws://192.168.1.100:8765" */
+    serverUrl: string;
+    /** Initial language (default: "auto" for Parakeet auto-detect) */
+    language?: string;
+    /** Force Whisper model even if Parakeet supports the language */
+    forceWhisper?: boolean;
+}
+declare class DtelecomSTT implements STTPlugin {
+    private readonly options;
+    constructor(options: DtelecomSTTOptions);
+    createStream(options?: STTStreamOptions): STTStream;
+}
+
+/**
+ * DtelecomTTS — real-time streaming TTS via dTelecom TTS server (realtime-tts-m2).
+ *
+ * Protocol:
+ * - Connect to ws://<server>:<port> (address from options, no API key)
+ * - Send config: {"config":{"voice":"af_heart","lang_code":"a","speed":1.0}}
+ * - Send text: {"text":"Hello world"} — uses config defaults
+ * - Send text with per-message override: {"text":"Hola","voice":"ef_dora","lang_code":"e","speed":1.0}
+ * - Receive: {"type":"generating","text":"..."} then binary PCM16 48kHz chunks, then {"type":"done"}
+ * - Cancel: {"type":"clear"} → {"type":"cleared"}
+ *
+ * Key differences from DeepgramTTS:
+ * - Single WebSocket connection (not per-language pool)
+ * - Per-message voice/language switching instead of separate connections
+ * - Server outputs 48kHz PCM16 (resampled from Kokoro's native 24kHz)
+ * - Uses SSML <lang> tags to route text segments to correct voice (same as DeepgramTTS)
+ */
+
+interface VoiceConfig {
+    voice: string;
+    langCode: string;
+}
+interface DtelecomTTSOptions {
+    /** WebSocket server URL, e.g. "ws://192.168.1.100:8766" */
+    serverUrl: string;
+    /** Voice config per language: { en: { voice: "af_heart", langCode: "a" }, es: { voice: "bf_emma", langCode: "b" } } */
+    voices: Record<string, VoiceConfig>;
+    /** Default language code (default: "en") */
+    defaultLanguage?: string;
+    /** Speech speed multiplier (default: 1.0) */
+    speed?: number;
+}
+declare class DtelecomTTS implements TTSPlugin {
+    private readonly serverUrl;
+    private readonly voices;
+    private readonly defaultLang;
+    private readonly speed;
+    private ws;
+    private connectPromise;
+    private flushState;
+    /** Default language code for untagged text (e.g. 'en'). */
+    get defaultLanguage(): string;
+    constructor(options: DtelecomTTSOptions);
+    /** Pre-connect WebSocket to TTS server. */
+    warmup(): Promise<void>;
+    /** Close WebSocket connection. */
+    close(): void;
+    /** Strip SSML lang tags from text for display/events. */
+    cleanText(text: string): string;
+    synthesize(text: string, signal?: AbortSignal): AsyncGenerator<Buffer>;
+    private synthesizeSegment;
+    /** Ensure a WebSocket connection exists and is open. */
+    private ensureConnection;
+}
+
+export { CartesiaTTS, type CartesiaTTSOptions, DeepgramSTT, type DeepgramSTTOptions, DeepgramTTS, type DeepgramTTSOptions, DtelecomSTT, type DtelecomSTTOptions, DtelecomTTS, type DtelecomTTSOptions, type VoiceConfig as DtelecomVoiceConfig, OpenRouterLLM, type OpenRouterLLMOptions };