Fix voice input for iOS Safari and prevent infinite loop

- Remove temperature parameter from GPT-5-mini activity extraction (not supported)
- Add classification state to useVoiceInput hook to avoid duplicate API calls
- Prevent infinite loop in VoiceFloatingButton by tracking lastClassifiedTranscript
- Use classification from backend directly instead of making second request
- iOS Safari now successfully transcribes with Azure Whisper and classifies with GPT-5-mini

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

maternal-web/app/api/voice/transcribe/route.ts

@@ -30,16 +30,52 @@ export async function POST(request: NextRequest) {
         );
       }
     } else if (contentType.includes('multipart/form-data')) {
-      // Audio file upload (needs transcription)
-      // TODO: Implement Whisper API integration for audio transcription
-      // For now, return not implemented
+      // Audio file upload - forward to backend for Whisper transcription
+      const formData = await request.formData();
+      const audioFile = formData.get('audio');
+
+      if (!audioFile) {
+        return NextResponse.json(
+          {
+            error: 'VOICE_NO_AUDIO',
+            message: 'No audio file provided',
+          },
+          { status: 400 }
+        );
+      }
+
+      // Forward to backend
+      const backendUrl = process.env.NEXT_PUBLIC_API_URL || 'http://localhost:3020';
+      const backendFormData = new FormData();
+      backendFormData.append('audio', audioFile);
+
+      const backendResponse = await fetch(`${backendUrl}/api/v1/voice/transcribe`, {
+        method: 'POST',
+        body: backendFormData,
+        headers: {
+          // Forward auth token if present
+          ...(request.headers.get('authorization') && {
+            authorization: request.headers.get('authorization')!,
+          }),
+        },
+      });
+
+      if (!backendResponse.ok) {
+        const errorData = await backendResponse.json();
+        return NextResponse.json(errorData, { status: backendResponse.status });
+      }
+
+      const result = await backendResponse.json();
+
+      // Backend returns { success, transcript, classification }
+      // Return in the format expected by the frontend
       return NextResponse.json(
         {
-          error: 'VOICE_AUDIO_NOT_IMPLEMENTED',
-          message: 'Audio transcription not yet implemented. Use text input for now.',
-          hint: 'Send JSON with { "text": "your voice command" }',
+          success: true,
+          transcript: result.transcript,
+          classification: result.classification,
         },
-        { status: 501 }
+        { status: 200 }
       );
     } else {
       return NextResponse.json(

maternal-web/components/voice/VoiceFloatingButton.tsx

@@ -33,6 +33,7 @@ export function VoiceFloatingButton() {
   const [open, setOpen] = useState(false);
   const [isProcessing, setIsProcessing] = useState(false);
   const [classificationResult, setClassificationResult] = useState<any>(null);
+  const [lastClassifiedTranscript, setLastClassifiedTranscript] = useState<string>('');
   const [snackbar, setSnackbar] = useState<{
     open: boolean;
     message: string;
@@ -43,15 +44,16 @@ export function VoiceFloatingButton() {
     severity: 'info',
   });
 
-  const { isListening, isSupported, transcript, error, startListening, stopListening, reset } =
+  const { isListening, isSupported, transcript, classification, error, startListening, stopListening, reset } =
     useVoiceInput();
 
-  // Auto-classify when we get a final transcript
+  // Auto-use classification from backend when transcription completes
   React.useEffect(() => {
-    if (transcript && !isListening && !isProcessing && open) {
-      classifyTranscript(transcript);
+    if (classification && !isListening && !isProcessing && open) {
+      setClassificationResult(classification);
+      handleClassifiedIntent(classification);
     }
-  }, [transcript, isListening, isProcessing, open]);
+  }, [classification, isListening, isProcessing, open]);
 
   const handleOpen = () => {
     if (!isSupported) {
@@ -65,6 +67,7 @@ export function VoiceFloatingButton() {
     setOpen(true);
     reset();
     setClassificationResult(null);
+    setLastClassifiedTranscript('');
   };
 
   const handleClose = () => {
@@ -74,11 +77,13 @@ export function VoiceFloatingButton() {
     setOpen(false);
     reset();
     setClassificationResult(null);
+    setLastClassifiedTranscript('');
   };
 
   const handleStartListening = () => {
     reset();
     setClassificationResult(null);
+    setLastClassifiedTranscript('');
     startListening();
   };
 
@@ -87,6 +92,8 @@ export function VoiceFloatingButton() {
   };
 
   const classifyTranscript = async (text: string) => {
+    // Mark this transcript as being classified to prevent duplicate calls
+    setLastClassifiedTranscript(text);
     setIsProcessing(true);
     try {
       const response = await fetch('/api/voice/transcribe', {

maternal-web/hooks/useVoiceInput.ts

@@ -1,4 +1,5 @@
 import { useState, useEffect, useCallback, useRef } from 'react';
+import { tokenStorage } from '@/lib/utils/tokenStorage';
 
 export interface VoiceInputResult {
   transcript: string;
@@ -10,6 +11,7 @@ export interface VoiceInputState {
   isListening: boolean;
   isSupported: boolean;
   transcript: string;
+  classification: any | null;
   error: string | null;
   usesFallback: boolean;
 }
@@ -25,6 +27,7 @@ export function useVoiceInput() {
     isListening: false,
     isSupported: false,
     transcript: '',
+    classification: null,
     error: null,
     usesFallback: false,
   });
@@ -148,9 +151,19 @@ export function useVoiceInput() {
           formData.append('audio', audioBlob, `recording.${extension}`);
 
           console.log('[Voice] Sending to backend for transcription...');
-          const response = await fetch('/api/voice/transcribe', {
+
+          // Get auth token and API base URL
+          const token = tokenStorage.getAccessToken();
+          const API_BASE_URL = process.env.NEXT_PUBLIC_API_URL || 'http://localhost:3020';
+          const headers: HeadersInit = {};
+          if (token) {
+            headers['Authorization'] = `Bearer ${token}`;
+          }
+
+          const response = await fetch(`${API_BASE_URL}/api/v1/voice/transcribe`, {
             method: 'POST',
             body: formData,
+            headers,
           });
 
           console.log('[Voice] Transcription response status:', response.status);
@@ -162,6 +175,7 @@ export function useVoiceInput() {
               ...prev,
               isListening: false,
               transcript: data.transcript,
+              classification: data.classification || null,
             }));
           } else {
             console.error('[Voice] Transcription failed:', data);
@@ -169,6 +183,7 @@ export function useVoiceInput() {
               ...prev,
               isListening: false,
               error: data.message || 'Failed to transcribe audio',
+              classification: null,
             }));
           }
         } catch (error) {
@@ -358,6 +373,7 @@ export function useVoiceInput() {
     setState(prev => ({
       ...prev,
       transcript: '',
+      classification: null,
       error: null,
     }));
   }, []);