Viele Änderungen

SPEC.md

@@ -3,15 +3,16 @@
 ## Project Overview
 - **Name**: Transcribe App
 - **Type**: Web Application (FastAPI + Vanilla JS)
-- **Core Functionality**: Voice recording/transcription with AI processing and email delivery
+- **Core Functionality**: Voice recording/transcription with AI-based protocol generation and email delivery
 - **Target Users**: Users who need to transcribe voice notes and receive processed text via email
 
 ## Tech Stack
 - **Backend**: FastAPI (Python)
-- **Speech-to-Text**: OpenAI Whisper API
-- **AI Processing**: OpenAI GPT-4o
+- **Speech-to-Text**: Azure AI Speech Fast Transcription API
+- **Meeting-Protokoll**: Claude Sonnet 4.6 via Azure Foundry
 - **Email**: SMTP with environment variables
 - **Frontend**: Vanilla HTML/CSS/JS
+- **Lokale Zwischenspeicherung im Browser**: IndexedDB für UI-State und lokale Audio-Datei bis zur Transkription
 
 ## UI/UX Specification
 
@@ -70,8 +71,8 @@
 - Steps displayed vertically:
   1. "Hochladen..." / "✓ Hochgeladen"
   2. "Transkribieren..." / "✓ Transkribiert"
-  3. "KI-Verarbeitung..." / "✓ Verarbeitet"
-  4. "E-Mail senden..." / "✓ Gesendet"
+  3. "Review..." / "✓ Bereit"
+  4. "Protokoll senden..." / "✓ Gesendet"
 - Each step with icon, timestamp
 
 #### Send Button
@@ -84,43 +85,51 @@
 ### Core Features
 
 #### 1. File Upload (Drag & Drop)
-- Accept only .mp3 files
-- Max file size: 25MB
-- Show progress during upload
+- Accept common audio formats (.mp3, .wav, .webm, .ogg, .flac)
+- Max file size configurable via environment variable
 - Validate file type before upload
+- Keep audio local in the browser until the user starts transcription
 
 #### 2. Voice Recording
 - Use MediaRecorder API (browser)
-- Record in webm/opus format (convert to mp3 on server)
+- Record in webm/opus format (convert to wav on server for Azure Speech)
 - Show recording duration timer
 - Playback recorded audio before submission
 
 #### 3. Transcription
-- Use OpenAI Whisper API
+- Use Azure AI Speech Fast Transcription API
 - Support German language
-- Return timestamped segments
+- Use speaker diarization when Azure returns speaker-separated phrases
+- Send audio directly in the `/transcribe` request and delete the temporary server file immediately after processing
 
-#### 4. AI Processing
-- Use OpenAI GPT-4o
-- Improve transcription:
-  - Fix grammar and spelling
-  - Add proper punctuation
-  - Structure into paragraphs
-  - Maintain original meaning
-
-#### 5. Email Delivery
-- Send processed text via SMTP
-- Include original and processed text
-- Email format: plain text
+#### 4. Meeting-Protokoll und Versand
+- Generate the final meeting summary from `prompt.txt`
+- Use Claude Sonnet 4.6 hosted in Azure Foundry
+- Queue document generation and email delivery in the background
+- Expose send-job status via API polling
+- Send formatted HTML email via SMTP
+- Attach full transcript as PDF
 
 ### API Endpoints
-- `POST /upload` - Upload audio file
-- `POST /transcribe` - Transcribe audio
-- `POST /process` - AI process text
-- `POST /send-email` - Send email
+- `POST /transcribe` - Upload and transcribe audio in one request
+- `POST /send` - Queue email send job
+- `GET /send-status/{job_id}` - Poll send job status
 
 ### Environment Variables
-- OPENAI_API_KEY
+- AZURE_SPEECH_ENDPOINT
+- AZURE_SPEECH_KEY
+- AZURE_SPEECH_API_VERSION
+- AZURE_SPEECH_DIARIZATION_ENABLED
+- AZURE_SPEECH_MAX_SPEAKERS
+- AZURE_SPEECH_PROFANITY_MODE
+- AZURE_SPEECH_LLM_PROMPT
+- AZURE_ANTHROPIC_BASE_URL
+- AZURE_ANTHROPIC_API_KEY
+- AZURE_ANTHROPIC_MODEL
+- AZURE_ANTHROPIC_MAX_TOKENS
+- AZURE_ANTHROPIC_TEMPERATURE
+- MAX_FILE_SIZE_MB
+- TEMP_DIR
 - SMTP_HOST
 - SMTP_PORT
 - SMTP_USER
@@ -137,4 +146,4 @@
 6. ✓ App shows step-by-step progress
 7. ✓ Final email is sent with processed text
 8. ✓ Clean black/white design throughout
-9. ✓ Responsive on mobile devices
+9. ✓ Responsive on mobile devices