fix(voice): stop microphone capture during startup

Author: waleedlatif1

apps/sim/hooks/use-speech-to-text.ts

@@ -223,6 +223,7 @@ export function useSpeechToText({
     if (startingRef.current) return false
     startingRef.current = true
     const generation = ++generationRef.current
+    setIsListening(true)
 
     try {
       let tokenData: Awaited<ReturnType<typeof requestJson<typeof speechTokenContract>>>
@@ -236,6 +237,7 @@ export function useSpeechToText({
         if (!mountedRef.current || generation !== generationRef.current) return false
         if (isApiClientError(err) && err.status === 402) {
           const isMemberLimit = (err.body as { scope?: string } | null)?.scope === 'member'
+          setIsListening(false)
           onUsageLimitExceededRef.current?.(err.message, isMemberLimit)
           return false
         }
@@ -262,6 +264,10 @@ export function useSpeechToText({
       }
 
       streamRef.current = stream
+      sessionTimerRef.current = setTimeout(() => {
+        logger.info('Voice input session reached max duration, stopping')
+        stopStreamingRef.current()
+      }, MAX_SESSION_MS)
 
       const params = new URLSearchParams({
         token,
@@ -349,19 +355,13 @@ export function useSpeechToText({
 
       sendIntervalRef.current = setInterval(flushAudioBuffer, CHUNK_SEND_INTERVAL_MS)
 
-      setIsListening(true)
-
-      sessionTimerRef.current = setTimeout(() => {
-        logger.info('Voice input session reached max duration, stopping')
-        stopStreamingRef.current()
-      }, MAX_SESSION_MS)
-
       return true
     } catch (error) {
       if (generation !== generationRef.current) return false
       logger.error('Failed to start speech streaming', error)
       cleanup()
       if (mountedRef.current) {
+        setIsListening(false)
         onErrorRef.current?.(classifyStartError(error))
       }
       return false
@@ -371,6 +371,14 @@ export function useSpeechToText({
   }, [cleanup, flushAudioBuffer])
 
   const stopStreaming = useCallback(() => {
+    if (startingRef.current) {
+      generationRef.current++
+      startingRef.current = false
+      cleanup()
+      setIsListening(false)
+      return
+    }
+
     if (sessionTimerRef.current) {
       clearTimeout(sessionTimerRef.current)
       sessionTimerRef.current = null
@@ -426,7 +434,7 @@ export function useSpeechToText({
     }
 
     setIsListening(false)
-  }, [flushAudioBuffer])
+  }, [cleanup, flushAudioBuffer])
 
   stopStreamingRef.current = stopStreaming
 
@@ -439,7 +447,7 @@ export function useSpeechToText({
   }, [cleanup])
 
   const toggleListening = useCallback(() => {
-    if (isListening) {
+    if (startingRef.current || isListening) {
       stopStreaming()
     } else {
       startStreaming()