NVIDIA ASR Transcription Toolkit

🎙️ Demo repository for the Build a Voice-Enabled AI Agent in Minutes session at Microsoft Reactor AI Apps & Agents Dev Days (February 10, 2026). Brought to you by Microsoft & NVIDIA.

Local audio transcription using NVIDIA ASR models via the NeMo framework, organized into five scenarios:

Quick Start

# Setup (Python 3.10-3.12 only)
py -3.12 -m venv venv && venv\Scripts\activate          # Windows
pip install torch --index-url https://download.pytorch.org/whl/cu121  # GPU
pip install -r requirements.txt && python fix_lhotse.py

# Scenario 1: Transcribe a file
python scenario1/transcribe.py audio.mp3

# Scenario 3: Multilingual
python scenario3/transcribe.py audio.mp3 es

# Scenario 4: Client-server
cd scenario4/AppHost && dotnet run

# Scenario 5: Voice Agent
cd scenario5
pip install -r requirements.txt
python app.py
# Open http://localhost:8000 in your browser

Documentation

Requirements

• Python 3.10–3.12 (3.13 not supported)
• NVIDIA GPU with CUDA recommended (CPU fallback available)
• ~2.5GB disk space for models

Slides & Link Inventory

The slides/ folder contains visual references from the Microsoft × NVIDIA session, along with curated link inventories for each slide.

References

• NVIDIA Parakeet Model — English, CC-BY-4.0
• NVIDIA Canary-1B Model — Multilingual, CC-BY-NC-4.0
• NeMo ASR Documentation
• NVIDIA FastPitch Model — English TTS, CC-BY-4.0
• NVIDIA HiFi-GAN Model — Vocoder, CC-BY-4.0
• TinyLlama-1.1B-Chat — LLM for Smart Mode, Apache-2.0
• Azure Deployment Example