@absolutejs/voice 0.0.14 → 0.0.16

package/dist/angular/index.js

@@ -379,7 +379,7 @@ var createVoiceStreamStore = () => {
       case "final":
         state = {
           ...state,
-          partial: "",
+          partial: action.transcript.text,
           turns: state.turns.map((turn) => turn)
         };
         break;

package/dist/client/index.js

@@ -375,7 +375,7 @@ var createVoiceStreamStore = () => {
       case "final":
         state = {
           ...state,
-          partial: "",
+          partial: action.transcript.text,
           turns: state.turns.map((turn) => turn)
         };
         break;

package/dist/index.js

@@ -236,12 +236,59 @@ var toVoiceSessionSummary = (session) => ({
 
 // src/turnDetection.ts
 var DEFAULT_SILENCE_MS = 700;
+var DEFAULT_SPEECH_THRESHOLD = 0.015;
+var toUint8Array = (audio) => {
+  if (audio instanceof ArrayBuffer) {
+    return new Uint8Array(audio);
+  }
+  return new Uint8Array(audio.buffer, audio.byteOffset, audio.byteLength);
+};
+var measureAudioLevel = (audio) => {
+  const bytes = toUint8Array(audio);
+  if (bytes.byteLength < 2) {
+    return 0;
+  }
+  const samples = new Int16Array(bytes.buffer, bytes.byteOffset, Math.floor(bytes.byteLength / 2));
+  if (samples.length === 0) {
+    return 0;
+  }
+  let sumSquares = 0;
+  for (const sample of samples) {
+    const normalized = sample / 32768;
+    sumSquares += normalized * normalized;
+  }
+  return Math.sqrt(sumSquares / samples.length);
+};
+var normalizeText = (value) => value.trim().replace(/\s+/g, " ");
+var mergeTranscriptTexts = (transcripts) => {
+  const merged = [];
+  for (const transcript of transcripts) {
+    const nextText = normalizeText(transcript.text);
+    if (!nextText) {
+      continue;
+    }
+    const previous = merged.at(-1);
+    if (!previous) {
+      merged.push(nextText);
+      continue;
+    }
+    if (nextText === previous || previous.includes(nextText)) {
+      continue;
+    }
+    if (nextText.includes(previous)) {
+      merged[merged.length - 1] = nextText;
+      continue;
+    }
+    merged.push(nextText);
+  }
+  return merged.join(" ").trim();
+};
 var buildTurnText = (transcripts, partialText) => {
-  const finalText = transcripts.map((transcript) => transcript.text.trim()).filter(Boolean).join(" ").trim();
+  const finalText = mergeTranscriptTexts(transcripts);
   if (finalText) {
     return finalText;
   }
-  return partialText.trim();
+  return normalizeText(partialText);
 };
 
 // src/session.ts
@@ -269,11 +316,13 @@ var createVoiceSession = (options) => {
     timeout: options.reconnect.timeout ?? DEFAULT_RECONNECT_TIMEOUT
   };
   const turnDetection = {
-    silenceMs: options.turnDetection.silenceMs ?? DEFAULT_SILENCE_MS
+    silenceMs: options.turnDetection.silenceMs ?? DEFAULT_SILENCE_MS,
+    speechThreshold: options.turnDetection.speechThreshold ?? DEFAULT_SPEECH_THRESHOLD
   };
   let socket = options.socket;
   let sttSession = null;
   let silenceTimer = null;
+  let speechDetected = false;
   const clearSilenceTimer = () => {
     if (!silenceTimer) {
       return;
@@ -315,7 +364,9 @@ var createVoiceSession = (options) => {
     }
   };
   const scheduleSilenceCommit = () => {
-    clearSilenceTimer();
+    if (silenceTimer) {
+      return;
+    }
     silenceTimer = setTimeout(() => {
       api.commitTurn("silence");
     }, turnDetection.silenceMs);
@@ -338,7 +389,7 @@ var createVoiceSession = (options) => {
   const handlePartial = async (transcript) => {
     await writeSession((session) => {
       session.currentTurn.lastAudioAt = Date.now();
-      session.currentTurn.partialText = transcript.text;
+      session.currentTurn.partialText = buildTurnText(session.currentTurn.transcripts, transcript.text);
       session.lastActivityAt = Date.now();
       session.status = "active";
     });
@@ -346,7 +397,6 @@ var createVoiceSession = (options) => {
       transcript,
       type: "partial"
     });
-    scheduleSilenceCommit();
   };
   const handleFinal = async (transcript) => {
     await writeSession((session) => {
@@ -370,7 +420,6 @@ var createVoiceSession = (options) => {
       transcript,
       type: "final"
     });
-    scheduleSilenceCommit();
   };
   const ensureAdapter = async () => {
     if (sttSession) {
@@ -472,6 +521,7 @@ var createVoiceSession = (options) => {
         currentSession.status = "active";
         currentSession.turns = [...currentSession.turns, turn];
       });
+      speechDetected = false;
       logger.info("voice turn committed", {
         reason,
         sessionId: options.id,
@@ -505,6 +555,7 @@ var createVoiceSession = (options) => {
         type: "complete"
       });
       await closeAdapter("complete");
+      speechDetected = false;
       await options.route.onComplete({
         api,
         context: options.context,
@@ -575,6 +626,7 @@ var createVoiceSession = (options) => {
         session.reconnect.lastDisconnectAt = Date.now();
         session.status = "reconnecting";
       });
+      speechDetected = false;
     },
     fail: async (error) => {
       clearSilenceTimer();
@@ -589,6 +641,7 @@ var createVoiceSession = (options) => {
         type: "error"
       });
       await closeAdapter("failed");
+      speechDetected = false;
       await options.route.onError?.({
         api,
         context: options.context,
@@ -603,11 +656,22 @@ var createVoiceSession = (options) => {
         return;
       }
       const adapter = await ensureAdapter();
+      const audioLevel = measureAudioLevel(audio);
       await writeSession((currentSession) => {
         currentSession.currentTurn.lastAudioAt = Date.now();
         currentSession.lastActivityAt = Date.now();
         currentSession.status = "active";
       });
+      if (audioLevel >= turnDetection.speechThreshold) {
+        speechDetected = true;
+        clearSilenceTimer();
+      } else if (speechDetected) {
+        const currentSession = await readSession();
+        const hasTurnText = Boolean(buildTurnText(currentSession.currentTurn.transcripts, currentSession.currentTurn.partialText));
+        if (hasTurnText) {
+          scheduleSilenceCommit();
+        }
+      }
       await adapter.send(audio);
     },
     snapshot: async () => readSession()
@@ -776,7 +840,8 @@ var voice = (config) => {
         store: config.session,
         stt: config.stt,
         turnDetection: {
-          silenceMs: config.turnDetection?.silenceMs ?? 700
+          silenceMs: config.turnDetection?.silenceMs ?? 700,
+          speechThreshold: config.turnDetection?.speechThreshold ?? 0.015
         }
       });
       if (!current) {
@@ -811,7 +876,8 @@ var voice = (config) => {
         store: config.session,
         stt: config.stt,
         turnDetection: {
-          silenceMs: config.turnDetection?.silenceMs ?? 700
+          silenceMs: config.turnDetection?.silenceMs ?? 700,
+          speechThreshold: config.turnDetection?.speechThreshold ?? 0.015
         }
       });
       runtime.activeSessions.set(sessionId, session);

package/dist/react/index.js

@@ -379,7 +379,7 @@ var createVoiceStreamStore = () => {
       case "final":
         state = {
           ...state,
-          partial: "",
+          partial: action.transcript.text,
           turns: state.turns.map((turn) => turn)
         };
         break;

package/dist/svelte/index.js

@@ -376,7 +376,7 @@ var createVoiceStreamStore = () => {
       case "final":
         state = {
           ...state,
-          partial: "",
+          partial: action.transcript.text,
           turns: state.turns.map((turn) => turn)
         };
         break;

package/dist/turnDetection.d.ts

@@ -1,3 +1,5 @@
-import type { Transcript } from './types';
+import type { AudioChunk, Transcript } from './types';
 export declare const DEFAULT_SILENCE_MS = 700;
+export declare const DEFAULT_SPEECH_THRESHOLD = 0.015;
+export declare const measureAudioLevel: (audio: AudioChunk) => number;
 export declare const buildTurnText: (transcripts: Transcript[], partialText: string) => string;

package/dist/types.d.ts

@@ -211,6 +211,7 @@ export type VoicePluginConfig<TContext = unknown, TSession extends VoiceSessionR
     reconnect?: VoiceReconnectConfig;
     turnDetection?: {
         silenceMs?: number;
+        speechThreshold?: number;
     };
     logger?: VoiceLogger;
     htmx?: boolean | VoiceHTMXConfig<TSession, NoInfer<TResult>>;
@@ -224,6 +225,7 @@ export type CreateVoiceSessionOptions<TContext = unknown, TSession extends Voice
     reconnect: Required<VoiceReconnectConfig>;
     turnDetection: {
         silenceMs: number;
+        speechThreshold: number;
     };
     route: VoiceNormalizedRouteConfig<TContext, TSession, TResult>;
     logger?: VoiceLogger;

package/dist/vue/index.js

@@ -379,7 +379,7 @@ var createVoiceStreamStore = () => {
       case "final":
         state = {
           ...state,
-          partial: "",
+          partial: action.transcript.text,
           turns: state.turns.map((turn) => turn)
         };
         break;

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@absolutejs/voice",
-	"version": "0.0.14",
+	"version": "0.0.16",
 	"description": "Voice primitives and Elysia plugin for AbsoluteJS",
 	"repository": {
 		"type": "git",