Description
NVIDIA Parakeet is an advanced automatic speech recognition (ASR) model featuring 600 million parameters, designed to deliver high-quality English transcription. It supports automatic punctuation, capitalization, and accurate word-level timestamp predictions, making it suitable for various applications such as conversational AI, voice assistants, and transcription services.
Read More
The model is based on the FastConformer architecture, integrating the TDT decoder, and is trained with full attention. This enables efficient transcription of audio segments up to 24 minutes in a single pass. It achieves an RTFx of 3380 on the HF-Open-ASR leaderboard with a batch size of 128, although performance may vary depending on dataset audio duration and batch size.
NVIDIA Parakeet is ready for commercial and non-commercial use, governed by the CC-BY-4.0 license. It is optimized to run on NVIDIA GPU-accelerated systems, leveraging hardware like GPU cores and software frameworks such as CUDA libraries for faster training and inference times compared to CPU-only solutions.
The model is intended for developers, researchers, academics, and industries building applications that require speech-to-text capabilities. It is trained using the NeMo toolkit, initialized from a FastConformer SSL checkpoint and fine-tuned on high-quality, human-transcribed data. The training dataset includes approximately 120,000 hours of English speech data from various sources.
NVIDIA Parakeet demonstrates robust performance across different signal-to-noise ratios and telephony-style audio formats. It is evaluated using the Huggingface Open ASR Leaderboard datasets, achieving a Word Error Rate (WER) of 6.05% on average. The model is designed to be noise robust and performs well across various types of audio inputs.
- CategoryAI Models & LLMs
- Pricing ModelOpen source
- DeploymentCloud
- PlatformsLinux
NVIDIA Parakeet ASR Model Highlights
600-million-parameter ASR model
Automatic punctuation and capitalization
Accurate word-level timestamp predictions
Robust performance on spoken numbers and song lyrics
FastConformer architecture with TDT decoder
Supports audio segments up to 24 minutes
RTFx of 3380 on HF-Open-ASR leaderboard
Optimized for NVIDIA GPU-accelerated systems
Commercial and non-commercial use
CC-BY-4.0 license
Getting Started with NVIDIA Parakeet ASR Model
Access page: Visit huggingface.co/nvidia/parakeet-tdt-0.6b-v2
Load model: Use NeMo toolkit to load pre-trained model
Configure environment: Install NVIDIA NeMo and PyTorch
Integrate: Use API or CLI for transcription tasks
Fine-tune: Customize model with additional datasets
NVIDIA Parakeet ASR Model's Use Cases
- Conversational AI
- Voice Assistants
- Transcription Services
- Subtitle Generation
- Voice Analytics
