Audio Transcription Pipeline

This project provides an automated audio transcription pipeline using Azure services. Audio files are retrieved using youtube-dl, in an m4a format.

Scenarios

• Upload audio files to Azure Blob Storage, automatically transcribe them using Azure Speech Service with optional speaker diarization, and load transcripts into Azure AI Foundry agents for interactive Q&A.
• Get transcripts and then use those to populate the Researcher Agent to get a comprehensive overview.

How It Works

┌─────────────┐     ┌──────────────────┐     ┌─────────────────┐
│ Audio File  │────▶│ Blob Trigger     │────▶│ Speech Service  │
│ (+ metadata)│     │ (Function)       │     │ (Transcription) │
└─────────────┘     └──────────────────┘     └─────────────────┘
                                                       │
                          destinationContainerUrl      │
                    (Speech writes JSON directly)      │
                                                       ▼
┌─────────────┐     ┌──────────────────┐     ┌─────────────────┐
│ AI Agent    │◀────│ Blob Trigger     │◀────│ Transcript JSON │
│ (Q&A ready) │     │ (Function)       │     │ + .txt file     │
└─────────────┘     └──────────────────┘     └─────────────────┘

Architecture Notes

• Local Development: Uses polling to wait for transcription completion
• Azure Deployment: Uses destinationContainerUrl - Speech Service writes JSON directly to the transcripts container
• The blob trigger parses the JSON, extracts the transcript text, and saves a .txt file with the original audio filename
• The .txt file is then uploaded to Azure AI Foundry for agent-based Q&A

Notes

• Azure Speech Service supports: WAV MP3 OGG/OPUS FLAC AMR WEBM (NOT m4a - use make convert-audio)
• Transcripts are saved as .txt files with the same name as the source audio file.
• diarization: true enables speaker separation, false for single speaker
• topic: Groups transcripts under the same AI agent (e.g., "project-planning")

Quick Start

Run make help to see all available commands.