| tags: | |
| - onnx | |
| - openvino | |
| - speech-recognition | |
| - npu | |
| - parakeet | |
| - nvidia | |
| - nemo | |
| language: en | |
| license: apache-2.0 | |
| base_model: nvidia/parakeet-tdt-0.6b-v3 | |
| # Parakeet TDT 0.6B v3 — ONNX (NPU-ready) | |
| ONNX export of [nvidia/parakeet-tdt-0.6b-v3](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3) for use with OpenVINO on Intel NPU. | |
| Includes the bundled NeMo mel spectrogram preprocessor (\) for a self-contained pipeline. | |
| ## Files | |
| | File | Size | Description | | |
| |------|------|-------------| | |
| | \ + \ | ~2.5 GB | Conformer encoder (runs on NPU) | | |
| | \ | 73 MB | TDT joint decoder (runs on CPU) | | |
| | \ | 141 KB | Mel spectrogram preprocessor (onnxruntime CPU) | | |
| | \ | 94 KB | 8193-token vocabulary | | |
| | \ | 97 B | Model metadata | | |
| ## Pipeline | |
| ## Performance (Intel Core Ultra / Meteor Lake NPU) | |
| | Metric | Value | | |
| |--------|-------| | |
| | Load time (cached) | 3.6s | | |
| | Transcribe 3s audio | 0.29s (RTF 0.095) | | |
| | WER (LibriSpeech test-clean) | 3.7% | | |
| | Max audio length | ~16s (MEL_FRAMES=1600) | | |
| ## Usage | |
| Used by [npu-whisper](https://github.com/goodsmileduck/npu-whisper) dictation engine: | |
| ## Credits | |
| - Original model: [nvidia/parakeet-tdt-0.6b-v3](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3) | |
| - ONNX export by: [istupakov/parakeet-tdt-0.6b-v3-onnx](https://huggingface.co/istupakov/parakeet-tdt-0.6b-v3-onnx) | |
| - Preprocessor from: [onnx-asr](https://pypi.org/project/onnx-asr/) package | |