articleschannelstagsspacestoolkit
NVIDIA Technical Blog

NVIDIA Speech AI Models Deliver Industry-Leading Accuracy and Performance

View Original
thumbnail

Table of Contents

Introduction

NVIDIA is leading the way in speech AI models with industry-leading accuracy and performance. The Parakeet and Canary AI model families are part of NVIDIA Riva, offering state-of-the-art automatic speech recognition (ASR) capabilities. These models are setting new benchmarks in accuracy, speed, and versatility.

NVIDIA Parakeet v2 Model Highlights

  • Parakeet v2 achieves a remarkable 6.05% word error rate (WER).
  • It offers blazing-fast inference capabilities, with an RTFx of 3386.02.
  • Features include accurate timestamps and song-to-lyrics transcription.
  • Parakeet v2 supports multilingual transcription in 25 languages with punctuation.

NVIDIA NeMo Canary Model Highlights

  • NeMo Canary models are also excelling on the Hugging Face ASR leaderboard.
  • These models offer high accuracy and performance for a variety of use cases.

NVIDIA Speech AI Model Details and Use Cases

  • Parakeet RNNT 1.1B: Accurate, multilingual transcription with punctuation in 25 languages.
  • Parakeet CTC 1.1B: Supports Silero Voice Activity Detector for noise robustness.
  • Parakeet CTC 06B: English model trained on over 35,000 hours of speech.
  • NVIDIA speech models are easy to deploy and enterprise-ready.

Deployment and Accessibility

  • Riva models are available through NVIDIA AI Enterprise, NVIDIA NGC, and as NVIDIA NIM microservices.
  • The latest research models can be accessed on Hugging Face for experimentation and development.

Conclusion

NVIDIA's Parakeet and NeMo Canary AI models are leading the field in speech recognition, offering superior accuracy, speed, and versatility. Whether for enterprise solutions, customer support, or media applications, these models provide the tools to elevate your products with clear and intelligent speech recognition capabilities. For more details on technical aspects, deployment guides, and access to the latest models, visit the NGC Catalog.