komos
A model that predicts the overall audio quality of Korean speech on a 1–5 MOS scale.
SITE CAPTUREBased on information published on the service's website. Selection and review policy
WHAT IT DOES
What does it do?
KOMOS predicts the overall quality of Korean speech on a 1–5 MOS scale. It accepts an audio file path, a single-channel NumPy array, or a PyTorch tensor, and converts file input to single-channel 16kHz audio. The model can be moved to CUDA for GPU inference.
KEY FEATURES
Key features
- Predict MOS scores for Korean read and spontaneous speech
- Accept an audio file path or NumPy and PyTorch audio data
- Convert file input to single-channel 16kHz audio
- Run GPU inference with CUDA