NeMo

NVIDIA NeMo is a framework for researchers and PyTorch developers building speech AI, and in its current form it concentrates on automatic speech recognition, text-to-speech, and speech-focused large language models. It leverages pretrained checkpoints published on Hugging Face, with recent releases including multilingual ASR models with controllable latency, unified offline and streaming recognition, and multilingual TTS, letting developers choose their own point on the latency and accuracy trade-off. Speech LLM work extends to full-duplex, interruptible voice conversation models. NeMo installs with pip into systems running recent Python and PyTorch versions and is designed to slot into an existing PyTorch environment rather than replace it, with CUDA acceleration optional but expected for serious training. The framework is released under the Apache 2.0 license and is used across academic speech research and production voice pipelines built on NVIDIA hardware.