Parakeet-TDT-0.6B-v3

Parakeet-TDT-0.6B-v3 is NVIDIA's 600M-parameter automatic speech recognition model, pairing a FastConformer encoder with a TDT decoder. Released alongside the Granary open speech dataset, it transcribes 25 European languages plus Russian and Ukrainian with automatic language detection, punctuation, capitalization, and word-level timestamps, and it can process 24-minute audio segments in a single pass.

The weights ship under CC-BY-4.0, which clears commercial use, and the model is tuned for high-throughput, low-latency transcription at scale. Compared with its English-only v2 predecessor, v3 trades a little peak English accuracy for broad multilingual coverage. Caveats: the language set is European-focused, so non-European languages need other models, and best throughput assumes NVIDIA GPUs with the NeMo stack.