NVIDIA Parakeet Model Test Results

[Rei-0011]

Model Used:
https://huggingface.co/nvidia/parakeet-tdt_ctc-0.6b-ja
Test Environment: vultr.com

3 vCPU
GPU: A100 20G VRAM
RAM: 30GB

Test Audio:
STARS-774.mp4 (7GB) → STARS-774.wav (884MB) conversion
ACC → PCM wav file
Processing Settings:
180-second chunks with 30-second overlap
Results:

Many segments were not recognized
Performance incomparable to Whisper
Significant improvements needed in future versions
This isn't just a comparison at the level of Whisper's small, medium, large models.
https://catalog.ngc.nvidia.com/orgs/nvidia/teams/riva/models/parakeet-rnnt-riva-1-1b-unified-ml-cs-universal
It's disappointing that I couldn't test this specific model.