Riva beta is openly available and includes fully optimized pipelines for deploying real-time AI speech applications such as transcription and virtual assistants.

Riva highlights include:

  • World class Automatic speech recognition (ASR), and text-to-speech (TTS) models in the NVIDIA NGC™ catalog trained on thousands of hours of speech data on NVIDIA supercomputer
  • Integration with the NVIDIA TAO Toolkit to finetune on custom data
  • Fully optimized deep learning pipelines that can scale to hundreds of thousands of real-time streams
  • End-to-end workflow and tools to deploy services using single command
Download Now

Riva Studio is Available in Early Access

Riva Studio is an Integrated Development Environment (IDE) for authoring, training, and deploying multimodal virtual assistants.

Apply for Early Access

Introductory Resources

Quick Start Guide

Step-by-step guide to deploy pretrained models as services on a local workstation and interact with them through a client.

Get started

Introductory Blog

Learn about the architecture, key features and components in Riva that help you build speech services.

Read blog

Introductory Webinar

Build a sample Riva application for transcription and named entity recognition.

Watch webinar

Riva Samples

Virtual Assistant

Sample demonstrating a simple but complete real-time, domain-specific conversational AI app.

Try sample

Virtual Assistant (with Rasa)

App showing how to integrate the Rasa Dialog Manager with Riva speech services.

Try sample


Sample showing transcription and named entity recognition that is fine-tuned on biomedical and clinical language

Try sample

Additional Riva Resources

Fine-Tuning with TAO Toolkit

You can quickly start with NVIDIA’s free pre-trained models and fine-tune them using the TAO Toolkit in Riva. NGC contains the following models and example notebooks for speech recognition and natural language understanding.

Download TAO Toolkit pre-trained models from NGC.


Build Real-Time Apps with Riva Services

Ethical AI

NVIDIA’s platforms and application frameworks enable developers to build a wide array of AI applications. Consider potential algorithmic bias when choosing or creating the models being deployed. Work with the model’s developer to ensure that it meets the requirements for the relevant industry and use case; that the necessary instruction and documentation are provided to understand error rates, confidence intervals, and results; and that the model is being used under the conditions and in the manner intended.