NVIDIA Releases Open-Source Parakeet-TDT-0.6B-v2 for Fast Speech Recognition

NVIDIA has launched Parakeet-TDT-0.6B-v2, a new open-source automatic speech recognition (ASR) model, now available on Hugging Face. The model, which boasts 600 million parameters, can transcribe 60 minutes of audio in just one second, setting a new benchmark in the AI speech industry with its exceptional speed and performance.

Parakeet-TDT-0.6B-v2 is released under the Creative Commons CC-BY-4.0 license, allowing developers and businesses to utilize it freely. This model uses a combination of Fast Conformer encoder and TDT decoder architecture, optimized for NVIDIA GPUs, and supports hardware like A100, H100, T4, and V100 boards. It also performs well on systems with as little as 2GB of RAM, enabling broad deployment across various enterprises.

The model achieves a Word Error Rate (WER) of 6.05% on the Hugging Face Open ASR Leaderboard, outperforming many existing open ASR models. It supports features such as punctuation, capitalization, and detailed word-level timestamping, making it suitable for applications like transcription services, voice assistants, and conversational AI platforms.

Developers can access Parakeet-TDT-0.6B-v2 via Hugging Face or through NVIDIA's NeMo toolkit, with installation instructions and integration guidance readily available to facilitate experimentation and deployment.