Nvidia Riva

Riva offers pretrained high-performance speech AI models available as gRPC-based microservices for low-latency streaming and high-throughput offline use cases, fully containerized to easily scale to hundreds and thousands of parallel streams. Riva benefits include:

• High Accuracy: pretrained automatic speech recognition (ASR) models are trained on thousands of hours of audio on NVIDIA supercomputers with support for English, Spanish, Mandarin, Hindi, Russian, Korean, Portuguese, German, and French.

• Human-Like Voices: state of the-art text-to-speech (TTS) models generate expressive voices, with two out-of-the-box professional female & male voices for US English.

• End-To-End Model Customization: best possible accuracy for different languages, accents, domains, vocabulary, and context is achievable by finetuning ASR flexible pipeline; desired voice and accent can be achieved by fine-tuning pitch, volume, and duration.

• Flexible Deployment: support in the cloud, data center, at the edge and on embedded.

• Real-Time Performance: far below the 300-milisecond threshold (100 for embedded) using powerful NVIDIA AI optimizations.

• Enterprise Support: broad platform support with priority notifications, and access to instructor-led workshops and NVIDIA AI experts.