@absolutejs/voice 0.0.22-beta.596 → 0.0.22-beta.598

package/dist/core/semanticTurn.d.ts

@@ -1,10 +1,20 @@
-import type { Transcript } from "./types";
+import type { AudioFormat, Transcript } from "./types";
 export type VoiceSemanticTurnInput = {
     audioLevel?: number;
     lastFinalTranscript?: Transcript;
     partialText: string;
     silenceMs: number;
     transcripts: Transcript[];
+    /**
+     * The current turn's buffered user audio (PCM chunks, oldest→newest) and its
+     * format. Lets an AUDIO-based end-of-turn detector (e.g. a smart-turn / Whisper
+     * EOT model) judge completion from prosody — pitch, pace, trailing intonation —
+     * which a transcript-only judge fundamentally cannot see. Undefined when no
+     * audio was buffered for the turn (the runtime only stores chunks above the
+     * speech threshold).
+     */
+    turnAudio?: ReadonlyArray<Uint8Array>;
+    turnAudioFormat?: AudioFormat;
 };
 export type VoiceSemanticTurnVerdict = {
     confidence?: number;

package/dist/index.js

@@ -3757,6 +3757,8 @@ var getBufferedAudioDurationMs = (chunks) => chunks.reduce((total, chunk) => tot
 var STREAM_SENTENCE_BOUNDARY = /[.!?\u2026]['")\]]*\s/;
 var STREAM_CLAUSE_BOUNDARY = /[,;:]\s/g;
 var MAX_TTS_CHUNK_CHARS = 320;
+var STREAM_SENTENCE_END = /[.!?\u2026]['")\]]*$/;
+var STREAM_IDLE_FLUSH_MS = 350;
 var nextSpeakableBoundary = (buffer) => {
   const match = STREAM_SENTENCE_BOUNDARY.exec(buffer);
   return match ? match.index + match[0].length : -1;
@@ -4135,6 +4137,17 @@ var createVoiceSession = (options) => {
     pruneTurnAudio();
     return currentTurnAudio.map((audio) => audio.chunk);
   };
+  const turnAudioInputFormat = recordingConfig?.userInputFormat ?? options.realtimeInputFormat ?? DEFAULT_REALTIME_FORMAT;
+  const getTurnAudioForDetector = () => {
+    if (!options.semanticTurnDetector || currentTurnAudio.length === 0) {
+      return { turnAudio: undefined, turnAudioFormat: undefined };
+    }
+    const turnAudio = currentTurnAudio.map((audio) => {
+      const c = audio.chunk;
+      return c instanceof ArrayBuffer ? new Uint8Array(c) : new Uint8Array(c.buffer, c.byteOffset, c.byteLength);
+    });
+    return { turnAudio, turnAudioFormat: turnAudioInputFormat };
+  };
   const clearSilenceTimer = () => {
     if (!silenceTimer) {
       return;
@@ -4474,7 +4487,8 @@ var createVoiceSession = (options) => {
         lastFinalTranscript: transcripts.at(-1),
         partialText,
         silenceMs,
-        transcripts
+        transcripts,
+        ...getTurnAudioForDetector()
       }));
       endOfTurn = verdict.endOfTurn;
     } catch {
@@ -5259,7 +5273,8 @@ var createVoiceSession = (options) => {
         lastFinalTranscript: transcript,
         partialText: session.currentTurn.partialText,
         silenceMs: session.currentTurn.silenceStartedAt !== undefined ? Date.now() - session.currentTurn.silenceStartedAt : 0,
-        transcripts: session.currentTurn.transcripts
+        transcripts: session.currentTurn.transcripts,
+        ...getTurnAudioForDetector()
       }));
       if (verdict.endOfTurn) {
         clearSilenceTimer();
@@ -5448,6 +5463,7 @@ var createVoiceSession = (options) => {
     let charsSent = 0;
     let started = false;
     let streamed = false;
+    let idleFlushTimer = null;
     let sendChain = Promise.resolve();
     let ttsSessionRequest = null;
     const ttsStartedAt = Date.now();
@@ -5507,8 +5523,23 @@ var createVoiceSession = (options) => {
         }
       })();
     };
+    const clearIdleFlush = () => {
+      if (idleFlushTimer) {
+        clearTimeout(idleFlushTimer);
+        idleFlushTimer = null;
+      }
+    };
+    const flushOnIdle = () => {
+      idleFlushTimer = null;
+      const pending = buffer.trim();
+      if (pending && STREAM_SENTENCE_END.test(pending)) {
+        flush(buffer);
+        buffer = "";
+      }
+    };
     return {
       finish: async () => {
+        clearIdleFlush();
         if (buffer.trim()) {
           flush(buffer);
         }
@@ -5554,6 +5585,10 @@ var createVoiceSession = (options) => {
           flush(buffer.slice(0, cut));
           buffer = buffer.slice(cut);
         }
+        clearIdleFlush();
+        if (buffer.trim()) {
+          idleFlushTimer = setTimeout(flushOnIdle, STREAM_IDLE_FLUSH_MS);
+        }
       }
     };
   };

package/dist/testing/index.js

@@ -5984,6 +5984,8 @@ var getBufferedAudioDurationMs = (chunks) => chunks.reduce((total, chunk) => tot
 var STREAM_SENTENCE_BOUNDARY = /[.!?\u2026]['")\]]*\s/;
 var STREAM_CLAUSE_BOUNDARY = /[,;:]\s/g;
 var MAX_TTS_CHUNK_CHARS = 320;
+var STREAM_SENTENCE_END = /[.!?\u2026]['")\]]*$/;
+var STREAM_IDLE_FLUSH_MS = 350;
 var nextSpeakableBoundary = (buffer) => {
   const match = STREAM_SENTENCE_BOUNDARY.exec(buffer);
   return match ? match.index + match[0].length : -1;
@@ -6362,6 +6364,17 @@ var createVoiceSession = (options) => {
     pruneTurnAudio();
     return currentTurnAudio.map((audio) => audio.chunk);
   };
+  const turnAudioInputFormat = recordingConfig?.userInputFormat ?? options.realtimeInputFormat ?? DEFAULT_REALTIME_FORMAT;
+  const getTurnAudioForDetector = () => {
+    if (!options.semanticTurnDetector || currentTurnAudio.length === 0) {
+      return { turnAudio: undefined, turnAudioFormat: undefined };
+    }
+    const turnAudio = currentTurnAudio.map((audio) => {
+      const c = audio.chunk;
+      return c instanceof ArrayBuffer ? new Uint8Array(c) : new Uint8Array(c.buffer, c.byteOffset, c.byteLength);
+    });
+    return { turnAudio, turnAudioFormat: turnAudioInputFormat };
+  };
   const clearSilenceTimer = () => {
     if (!silenceTimer) {
       return;
@@ -6701,7 +6714,8 @@ var createVoiceSession = (options) => {
         lastFinalTranscript: transcripts.at(-1),
         partialText,
         silenceMs,
-        transcripts
+        transcripts,
+        ...getTurnAudioForDetector()
       }));
       endOfTurn = verdict.endOfTurn;
     } catch {
@@ -7486,7 +7500,8 @@ var createVoiceSession = (options) => {
         lastFinalTranscript: transcript,
         partialText: session.currentTurn.partialText,
         silenceMs: session.currentTurn.silenceStartedAt !== undefined ? Date.now() - session.currentTurn.silenceStartedAt : 0,
-        transcripts: session.currentTurn.transcripts
+        transcripts: session.currentTurn.transcripts,
+        ...getTurnAudioForDetector()
       }));
       if (verdict.endOfTurn) {
         clearSilenceTimer();
@@ -7675,6 +7690,7 @@ var createVoiceSession = (options) => {
     let charsSent = 0;
     let started = false;
     let streamed = false;
+    let idleFlushTimer = null;
     let sendChain = Promise.resolve();
     let ttsSessionRequest = null;
     const ttsStartedAt = Date.now();
@@ -7734,8 +7750,23 @@ var createVoiceSession = (options) => {
         }
       })();
     };
+    const clearIdleFlush = () => {
+      if (idleFlushTimer) {
+        clearTimeout(idleFlushTimer);
+        idleFlushTimer = null;
+      }
+    };
+    const flushOnIdle = () => {
+      idleFlushTimer = null;
+      const pending = buffer.trim();
+      if (pending && STREAM_SENTENCE_END.test(pending)) {
+        flush(buffer);
+        buffer = "";
+      }
+    };
     return {
       finish: async () => {
+        clearIdleFlush();
         if (buffer.trim()) {
           flush(buffer);
         }
@@ -7781,6 +7812,10 @@ var createVoiceSession = (options) => {
           flush(buffer.slice(0, cut));
           buffer = buffer.slice(cut);
         }
+        clearIdleFlush();
+        if (buffer.trim()) {
+          idleFlushTimer = setTimeout(flushOnIdle, STREAM_IDLE_FLUSH_MS);
+        }
       }
     };
   };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@absolutejs/voice",
-  "version": "0.0.22-beta.596",
+  "version": "0.0.22-beta.598",
   "description": "Voice primitives and Elysia plugin for AbsoluteJS",
   "repository": {
     "type": "git",