@bojackduy/opencode-voice 0.10.0 → 0.10.1

package/lib/stt.js

@@ -2021,6 +2021,14 @@ export function registerSTT(api, kv, complete, prompts, opts, logger, deps = {})
   }
 
   api.lifecycle?.onDispose?.(() => {
+    // Unload mid-recording must not orphan the sox child: cancel kills it,
+    // bumps the pipeline generation (late transcriptions are dropped), and
+    // clears the sticky toast. processing is not owned by cancel, reset it
+    // here so a later registration starts from idle.
+    try {
+      cancelRecording(logger);
+    } catch {}
+    processing = false;
     if (streamController) {
       const c = streamController;
       streamController = null;

package/lib/tts.js

@@ -638,11 +638,23 @@ export function registerTTS(api, kv, complete, prompts, opts, logger, deps = {})
   let conversationActive = false;
   let liveNotesActive = false;
 
-  api.event.on("session.status", (event) => {
+  // Auto-mode session subscriptions. The unsubscribe handles are kept so
+  // plugin unload (resumed session / reload) removes them: without this a
+  // second registration doubled auto-TTS speech and stale handlers fired
+  // cross-session. api.event.on may return nothing (older host), so guard.
+  const eventUnsubs = [];
+  function subscribe(name, fn) {
+    try {
+      const unsub = api.event.on(name, fn);
+      if (typeof unsub === "function") eventUnsubs.push(unsub);
+    } catch {}
+  }
+
+  subscribe("session.status", (event) => {
     if (event.properties?.status?.type === "busy") wasBusy = true;
   });
 
-  api.event.on("session.idle", async (event) => {
+  subscribe("session.idle", async (event) => {
     if (conversationActive || liveNotesActive) {
       wasBusy = false;
       return;
@@ -674,7 +686,7 @@ export function registerTTS(api, kv, complete, prompts, opts, logger, deps = {})
     await speakWithSessionPrefix(sessionID, llmResult.text, "Ready for your input.");
   });
 
-  api.event.on("permission.asked", async (event) => {
+  subscribe("permission.asked", async (event) => {
     if (conversationActive || liveNotesActive) return;
     if (kv.get("tts.mode", "off") !== "on") return;
     await speakWithSessionPrefix(
@@ -683,7 +695,7 @@ export function registerTTS(api, kv, complete, prompts, opts, logger, deps = {})
     );
   });
 
-  api.event.on("question.asked", async (event) => {
+  subscribe("question.asked", async (event) => {
     if (conversationActive || liveNotesActive) return;
     if (kv.get("tts.mode", "off") !== "on") return;
     await speakWithSessionPrefix(
@@ -822,8 +834,20 @@ export function registerTTS(api, kv, complete, prompts, opts, logger, deps = {})
 
   // The Chatterbox sidecar holds a multi-hundred-MB torch model; leaving it
   // running past the session would strand that memory. Same lifecycle hook the
-  // STT and live-notes servers use.
+  // STT and live-notes servers use. Also drops the session subscriptions (no
+  // duplicate auto-speech after re-registration), kills owned piper/play
+  // audio, and invalidates in-flight normalization so a late completion can
+  // never speak after unload.
   api.lifecycle?.onDispose?.(() => {
+    for (const unsub of eventUnsubs) {
+      try {
+        unsub();
+      } catch {}
+    }
+    eventUnsubs.length = 0;
+    try {
+      stopSpeech();
+    } catch {}
     chatterboxClient?.stop();
     chatterboxClient = null;
   });

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@bojackduy/opencode-voice",
-  "version": "0.10.0",
+  "version": "0.10.1",
   "description": "Speech-to-text and text-to-speech for OpenCode. Record voice prompts with whisper-cpp, hear responses via Piper TTS, with LLM normalization through any OpenAI-compatible endpoint.",
   "keywords": [
     "opencode",