NVIDIA NeMo Speech
A generative AI framework for researchers and developers building automatic speech recognition, text-to-speech, speaker diarization and speech translation models, with tools to create, customize and deploy new models from existing code and pretrained checkpoints.
NVIDIA NeMo Speech is a generative AI framework built for researchers and PyTorch developers working on speech models, including automatic speech recognition, text-to-speech and speech-oriented large language models. It helps create, customize and deploy new AI models by building on existing code and pretrained model checkpoints.
Pretrained checkpoints released through the project include multilingual streaming ASR models with controllable latency, a unified offline-and-streaming English ASR model, multilingual text-to-speech models, and a full-duplex conversational voice model built on a language model backbone with a speech and text-to-speech decoder. Speech recognition and translation checkpoints cover dozens of languages.
The project recently split from the broader NeMo toolkit to focus specifically on audio, speech and multimodal language models, and is written in Python under the Apache 2.0 license.