Automatic Speech Recognition
NeMo
Core ML
PyTorch
speech
audio
Transducer
TDT
FastConformer
Conformer
NeMo
hf-asr-leaderboard
Instructions to use FluidInference/parakeet-tdt-0.6b-v3-coreml with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- NeMo
How to use FluidInference/parakeet-tdt-0.6b-v3-coreml with NeMo:
import nemo.collections.asr as nemo_asr asr_model = nemo_asr.models.ASRModel.from_pretrained("FluidInference/parakeet-tdt-0.6b-v3-coreml") transcriptions = asr_model.transcribe(["file.wav"]) - Notebooks
- Google Colab
- Kaggle
Commit History
-
Upload 5 files
Verified signatureafa992c -
Delete Encoder.mlmodelc
Verified signaturebca7f24
-
separated model support as well
Verified signatureIt turns out that some devices, particularly iphones can't support the fused melEncoder model we end up with this error: - The fused MelEncoder takes raw waveform as input with shape [1, 240000] (15 s × 16 kHz). In Core ML’s internal NCHW mapping this becomes N1D1C1H1W240000. A15’s ANE (E5) rejects any single dimension > 16384, so W=240000 triggers: “Invalid layer: Tensor dimensions … W240000 … W[1–16384]”.b10f709
-
Delete Encoder.mlmodelc
Verified signature9ea7176 -
Upload 5 files
Verified signaturecb15c32