Parakeet-TDT-ExecuTorch-XNNPACK

Pre-exported ExecuTorch .pte file for Parakeet TDT 0.6B with XNNPACK backend (CPU). A fast speech-to-text model with word-level timestamps.

This export is unquantized (fp32, 2.4 GB). For a quantized version, see the export guide.

Installation

git clone https://github.com/pytorch/executorch/ ~/executorch
cd ~/executorch && ./install_executorch.sh
make parakeet-cpu

Download

pip install huggingface_hub
huggingface-cli download younghan-meta/Parakeet-TDT-ExecuTorch-XNNPACK --local-dir ~/parakeet

Run

cmake-out/examples/models/parakeet/parakeet_runner \
    --model_path ~/parakeet/model.pte \
    --tokenizer_path ~/parakeet/tokenizer.model \
    --audio_path ~/parakeet/poem.wav

Optional flags:

• --timestamps segment -- timestamp granularity: none|token|word|segment|all (default: segment)

Export Command

pip install "nemo_toolkit[asr]"
python examples/models/parakeet/export_parakeet_tdt.py \
    --backend xnnpack --output-dir ./parakeet_xnnpack

For quantized export (smaller, faster):

python examples/models/parakeet/export_parakeet_tdt.py \
    --backend xnnpack --qlinear_encoder 8da4w --qlinear 8da4w --qembedding 8w

More Info

• Official ExecuTorch Parakeet guide
• Original model