@dtelecom/agents-js 0.1.4 → 0.1.6

package/dist/index.d.mts

@@ -1,8 +1,8 @@
 import * as _dtelecom_server_sdk_node from '@dtelecom/server-sdk-node';
 import { Room, AudioSource, RemoteAudioTrack, AudioFrame } from '@dtelecom/server-sdk-node';
 import { EventEmitter } from 'events';
-import { A as AgentConfig, a as AgentStartOptions, M as Message, L as LLMPlugin, P as PipelineOptions, b as AgentState, S as STTStream, T as TranscriptionResult } from './types-EvtHMokR.mjs';
-export { c as AgentEvents, d as AudioOutput, D as DataMessageHandler, e as LLMChunk, f as MemoryConfig, g as PipelineEvents, R as RespondMode, h as STTPlugin, i as STTStreamOptions, j as TTSPlugin } from './types-EvtHMokR.mjs';
+import { A as AgentConfig, a as AgentStartOptions, M as Message, L as LLMPlugin, P as PipelineOptions, b as AgentState, S as STTStream, T as TranscriptionResult } from './types-Bfz5YQwd.mjs';
+export { c as AgentEvents, d as AudioOutput, D as DataMessageHandler, e as LLMChunk, f as MemoryConfig, g as PipelineEvents, R as RespondMode, h as STTPlugin, i as STTStreamOptions, j as TTSPlugin } from './types-Bfz5YQwd.mjs';
 
 declare class VoiceAgent extends EventEmitter {
     private readonly config;
@@ -103,6 +103,8 @@ declare class Pipeline extends EventEmitter {
     private readonly nameVariants;
     private readonly beforeRespond?;
     private readonly memory?;
+    /** Strip provider-specific markup (e.g. SSML lang tags) for display. */
+    private cleanText;
     /** Active STT streams, keyed by participant identity */
     private sttStreams;
     private _processing;

package/dist/index.d.ts

@@ -1,8 +1,8 @@
 import * as _dtelecom_server_sdk_node from '@dtelecom/server-sdk-node';
 import { Room, AudioSource, RemoteAudioTrack, AudioFrame } from '@dtelecom/server-sdk-node';
 import { EventEmitter } from 'events';
-import { A as AgentConfig, a as AgentStartOptions, M as Message, L as LLMPlugin, P as PipelineOptions, b as AgentState, S as STTStream, T as TranscriptionResult } from './types-EvtHMokR.js';
-export { c as AgentEvents, d as AudioOutput, D as DataMessageHandler, e as LLMChunk, f as MemoryConfig, g as PipelineEvents, R as RespondMode, h as STTPlugin, i as STTStreamOptions, j as TTSPlugin } from './types-EvtHMokR.js';
+import { A as AgentConfig, a as AgentStartOptions, M as Message, L as LLMPlugin, P as PipelineOptions, b as AgentState, S as STTStream, T as TranscriptionResult } from './types-Bfz5YQwd.js';
+export { c as AgentEvents, d as AudioOutput, D as DataMessageHandler, e as LLMChunk, f as MemoryConfig, g as PipelineEvents, R as RespondMode, h as STTPlugin, i as STTStreamOptions, j as TTSPlugin } from './types-Bfz5YQwd.js';
 
 declare class VoiceAgent extends EventEmitter {
     private readonly config;
@@ -103,6 +103,8 @@ declare class Pipeline extends EventEmitter {
     private readonly nameVariants;
     private readonly beforeRespond?;
     private readonly memory?;
+    /** Strip provider-specific markup (e.g. SSML lang tags) for display. */
+    private cleanText;
     /** Active STT streams, keyed by participant identity */
     private sttStreams;
     private _processing;

package/dist/index.js

@@ -698,20 +698,33 @@ var AudioOutput = class {
    * Start sparse silence keepalive to prevent the SFU from dropping the track.
    * With Opus DTX enabled, the encoder handles silence natively — we only need
    * an occasional packet to keep the SSRC alive.
+   *
+   * Waits for the RTP transport to be ready before sending — no frames are
+   * wasted before DTLS is connected.
    */
   startSilence() {
     if (this.silenceInterval) return;
-    log3.debug("Starting silence keepalive (sparse, 3s interval)");
-    const immediate = new import_server_sdk_node2.AudioFrame(SILENCE, SAMPLE_RATE, CHANNELS, SAMPLES_PER_FRAME);
-    this.source.captureFrame(immediate).catch(() => {
+    const startKeepalive = () => {
+      log3.debug("Transport ready \u2014 sending initial silence + starting 3s keepalive");
+      this.sendSilenceFrame();
+      this.silenceInterval = setInterval(() => {
+        if (!this._playing && !this._responding && !this._stopped) {
+          this.sendSilenceFrame();
+        }
+      }, 3e3);
+    };
+    if (this.source.ready) {
+      startKeepalive();
+    } else {
+      log3.debug("Waiting for transport before starting silence keepalive...");
+      this.source.onReady = () => startKeepalive();
+    }
+  }
+  sendSilenceFrame() {
+    const frame = new import_server_sdk_node2.AudioFrame(SILENCE, SAMPLE_RATE, CHANNELS, SAMPLES_PER_FRAME);
+    this.source.captureFrame(frame).catch((err) => {
+      log3.warn("Failed to send silence frame:", err);
     });
-    this.silenceInterval = setInterval(() => {
-      if (!this._playing && !this._responding && !this._stopped) {
-        const f = new import_server_sdk_node2.AudioFrame(SILENCE, SAMPLE_RATE, CHANNELS, SAMPLES_PER_FRAME);
-        this.source.captureFrame(f).catch(() => {
-        });
-      }
-    }, 3e3);
   }
   /**
    * Write a PCM16 buffer to the audio output.
@@ -1143,6 +1156,10 @@ var Pipeline = class extends import_events.EventEmitter {
   nameVariants;
   beforeRespond;
   memory;
+  /** Strip provider-specific markup (e.g. SSML lang tags) for display. */
+  cleanText(text) {
+    return this.tts?.cleanText ? this.tts.cleanText(text) : text;
+  }
   /** Active STT streams, keyed by participant identity */
   sttStreams = /* @__PURE__ */ new Map();
   _processing = false;
@@ -1396,7 +1413,7 @@ var Pipeline = class extends import_events.EventEmitter {
                   tFirstAudioPlayed = t;
                   this.setAgentState("speaking");
                 }
-                this.emit("sentence", sentence);
+                this.emit("sentence", this.cleanText(sentence));
               });
               continue;
             }
@@ -1424,7 +1441,7 @@ var Pipeline = class extends import_events.EventEmitter {
       if (fullResponse.trim()) {
         this.context.addAgentTurn(fullResponse.trim());
         this.memory?.storeTurn("assistant", fullResponse.trim(), true);
-        this.emit("response", fullResponse.trim());
+        this.emit("response", this.cleanText(fullResponse.trim()));
       }
       await sleep2(AUDIO_DRAIN_MS);
       this.setAgentState("idle");
@@ -1465,13 +1482,13 @@ var Pipeline = class extends import_events.EventEmitter {
       this.setAgentState("thinking");
       await this.synthesizeAndPlay(text, signal, () => {
         this.setAgentState("speaking");
-        this.emit("sentence", text);
+        this.emit("sentence", this.cleanText(text));
       });
       if (!signal.aborted) {
         await this.audioOutput.writeSilence(40);
         this.context.addAgentTurn(text);
         this.memory?.storeTurn("assistant", text, true);
-        this.emit("response", text);
+        this.emit("response", this.cleanText(text));
       }
       await sleep2(AUDIO_DRAIN_MS);
       this.setAgentState("idle");