Onnx feature extractors
Collection
Various models in onnx format meant for audio feature extraction • 24 items • Updated
PNCC features
Power-normalized cepstral coefficient (PNCC) feature extractor exported to ONNX. It maps a 16 kHz waveform to per-frame PNCC features.
Using it
The ONNX graph takes input_values [1, samples] (float32, 16 kHz) and returns features [1, frames, coefficients].
import numpy as np, onnxruntime as ort
from huggingface_hub import hf_hub_download
sess = ort.InferenceSession(hf_hub_download("TigreGotico/pncc-onnx", "pncc-pncc13-fft512-filters40-onnx.onnx"))
feats = sess.run(None, {"input_values": np.zeros((1, 16000), np.float32)})[0]
Files:
pncc-pncc13-fft512-filters40-onnx.onnx(13 coefficients)pncc-pncc40-fft512-filters64-onnx.onnx(40 coefficients)