nvidia

/

parakeet-tdt-0.6b-v3

like 796

Follow

NVIDIA 55.3k

I think I figured out the way to get this running on Jetson devices. (I think this scripts / flow will work on Thor or Spark as well).

Here is what you need.

1. Install UV

curl -LsSf https://astral.sh/uv/install.sh | sh

2. Setup venv

mkdir test_asr && cd test_asr
uv venv venv
source venv/bin/activate

3. Install all packages
   Here is the key magic, we have to make sure it points to the custom wheels pre-compiled for Jetson or you will have lots of headaches

#install torch from jetson wheels
uv pip install torch torchvision --index-url=https://pypi.jetson-ai-lab.io/jp6/cu126
#make sure it matches to the cuda version! 
uv pip install cuda-python==12.6 --reinstall --index-url=https://pypi.jetson-ai-lab.io/jp6/cu126
uv pip install sacrebleu
uv pip install nemo_toolkit[asr]

4. Run a test

#run this in a terminal to get the wav file
wget https://dldata-public.s3.us-east-2.amazonaws.com/2086-149220-0033.wav

#run these in Python 
import nemo.collections.asr as nemo_asr
asr_model = nemo_asr.models.ASRModel.from_pretrained(model_name="nvidia/parakeet-tdt-0.6b-v3")
output = asr_model.transcribe(['2086-149220-0033.wav'])
print(output[0].text)

Happy Jetson coding