address comments

Author: icecrasher321

apps/docs/content/docs/en/execution/costs.mdx

@@ -139,12 +139,12 @@ When configured, workflows use your key instead of Sim's hosted keys. If removed
 
 Voice input uses ElevenLabs Scribe v2 Realtime for speech-to-text transcription. It is available in the Mothership chat and in deployed chat voice mode.
 
-| Item | Cost | Notes |
-|------|------|-------|
-| Voice session | ~5 credits ($0.024) | Flat charge per session (up to 3 minutes) |
-| Session limit | 3 minutes | After 3 minutes, the session ends automatically. Start a new session to continue. |
+| Context | Cost per session | Max duration |
+|---------|-----------------|--------------|
+| Mothership (workspace) | ~5 credits ($0.024) | 3 minutes |
+| Deployed chat (voice mode) | ~2 credits ($0.008) | 1 minute |
 
-Voice sessions are billed when the session starts (when you click the mic button or enter voice mode). Multi-turn voice conversations in deployed chat reuse the same session billing — reconnections within a single voice call are not charged again.
+Each voice session is billed when it starts. In deployed chat voice mode, each conversation turn (speak → agent responds → speak again) is a separate session. Multi-turn conversations are billed per turn.
 
 <Callout type="info">
   Voice input requires `ELEVENLABS_API_KEY` to be configured. When the key is not set, voice input controls are hidden.

apps/sim/app/api/speech/token/route.ts

@@ -19,8 +19,8 @@ export const dynamic = 'force-dynamic'
 const ELEVENLABS_TOKEN_URL = 'https://api.elevenlabs.io/v1/single-use-token/realtime_scribe'
 
 const VOICE_SESSION_COST_PER_MIN = 0.008
-const VOICE_SESSION_MAX_MINUTES = 3
-const VOICE_SESSION_COST = VOICE_SESSION_COST_PER_MIN * VOICE_SESSION_MAX_MINUTES
+const WORKSPACE_SESSION_MAX_MINUTES = 3
+const CHAT_SESSION_MAX_MINUTES = 1
 
 const STT_TOKEN_RATE_LIMIT = {
   maxTokens: 30,
@@ -144,14 +144,17 @@ export async function POST(request: NextRequest) {
     const data = await response.json()
 
     if (billingUserId) {
+      const maxMinutes = chatId ? CHAT_SESSION_MAX_MINUTES : WORKSPACE_SESSION_MAX_MINUTES
+      const sessionCost = VOICE_SESSION_COST_PER_MIN * maxMinutes
+
       await recordUsage({
         userId: billingUserId,
         entries: [
           {
             category: 'fixed',
             source: 'voice-input',
-            description: `Voice input session (${VOICE_SESSION_MAX_MINUTES} min)`,
-            cost: VOICE_SESSION_COST * getCostMultiplier(),
+            description: `Voice input session (${maxMinutes} min)`,
+            cost: sessionCost * getCostMultiplier(),
           },
         ],
       }).catch((err) => {

apps/sim/app/chat/components/voice-interface/voice-interface.tsx

@@ -10,7 +10,7 @@ import { arrayBufferToBase64, floatTo16BitPCM } from '@/lib/speech/audio'
 import {
   CHUNK_SEND_INTERVAL_MS,
   ELEVENLABS_WS_URL,
-  MAX_SESSION_MS,
+  MAX_CHAT_SESSION_MS,
   SAMPLE_RATE,
 } from '@/lib/speech/config'
 
@@ -321,7 +321,7 @@ export function VoiceInterface({
       stopSendingAudio()
       closeWebSocket()
       updateState('idle')
-    }, MAX_SESSION_MS)
+    }, MAX_CHAT_SESSION_MS)
   }, [connectWebSocket, updateState, startSendingAudio, stopSendingAudio, closeWebSocket])
 
   const stopListening = useCallback(() => {

apps/sim/app/workspace/[workspaceId]/home/components/user-input/user-input.tsx

@@ -243,6 +243,7 @@ export function UserInput({
     isListening,
     isSupported: isSttSupported,
     toggleListening: rawToggle,
+    resetTranscript,
   } = useSpeechToText({ onTranscript: handleTranscript })
 
   const toggleListening = useCallback(() => {
@@ -432,13 +433,15 @@ export function UserInput({
     )
     setValue('')
     valueRef.current = ''
+    sttPrefixRef.current = ''
+    resetTranscript()
     currentFiles.clearAttachedFiles()
     currentContext.clearContexts()
 
     if (textareaRef.current) {
       textareaRef.current.style.height = 'auto'
     }
-  }, [onSubmit, textareaRef])
+  }, [onSubmit, textareaRef, resetTranscript])
 
   const handleKeyDown = useCallback(
     (e: React.KeyboardEvent<HTMLTextAreaElement>) => {

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

@@ -26,6 +26,7 @@ interface UseSpeechToTextReturn {
   isSupported: boolean
   permissionState: PermissionState
   toggleListening: () => void
+  resetTranscript: () => void
 }
 
 export function useSpeechToText({
@@ -334,18 +335,29 @@ export function useSpeechToText({
       streamRef.current = null
     }
 
-    setTimeout(() => {
-      if (wsRef.current) {
-        wsRef.current.close()
-        wsRef.current = null
-      }
-    }, 2000)
+    const wsToClose = wsRef.current
+    wsRef.current = null
+    if (wsToClose) {
+      setTimeout(() => {
+        if (
+          wsToClose.readyState === WebSocket.OPEN ||
+          wsToClose.readyState === WebSocket.CONNECTING
+        ) {
+          wsToClose.close()
+        }
+      }, 2000)
+    }
 
     setIsListening(false)
   }, [flushAudioBuffer])
 
   stopStreamingRef.current = stopStreaming
 
+  const resetTranscript = useCallback(() => {
+    committedTextRef.current = ''
+    isFirstChunkRef.current = true
+  }, [])
+
   const toggleListening = useCallback(() => {
     if (isListening) {
       stopStreaming()
@@ -367,5 +379,6 @@ export function useSpeechToText({
     isSupported,
     permissionState,
     toggleListening,
+    resetTranscript,
   }
 }

apps/sim/lib/speech/config.ts

@@ -4,6 +4,7 @@ export const ELEVENLABS_WS_URL = 'wss://api.elevenlabs.io/v1/speech-to-text/real
 export const SAMPLE_RATE = 16000
 export const CHUNK_SEND_INTERVAL_MS = 250
 export const MAX_SESSION_MS = 3 * 60 * 1000
+export const MAX_CHAT_SESSION_MS = 1 * 60 * 1000
 
 /**
  * Whether a speech-to-text provider is configured.