About
Parakeet is a family of automatic speech recognition (ASR) models developed by NVIDIA within the NeMo toolkit. Parakeet models combine a FastConformer encoder with various decoders: CTC, Transducer (RNN-T) and Token-and-Duration Transducer (TDT). The family includes variants with different parameter counts (e.g. 0.6B and 1.1B) and multilingual versions such as parakeet-tdt-0.6b-v3 covering 25 European languages. The models are optimized for high-throughput transcription on NVIDIA accelerators and released under permissive licenses.
