HyukII/audio-emotion-model

• Input: MFCC 13ch, length 100 → shape (B, 13, 100)
• Delta: (X - mean) / (std + 1e-8)
• Labels: see labels.json (index ↔ label 1:1)

Usage

import json, torch, numpy as np
from huggingface_hub import hf_hub_download
from importlib.machinery import SourceFileLoader

repo = "HyukII/audio-emotion-model"
w = hf_hub_download(repo, "pytorch_model.pth")
m = hf_hub_download(repo, "model.py")
lab = hf_hub_download(repo, "labels.json")

labels = json.load(open(lab, encoding="utf-8"))
Model = SourceFileLoader("amodel", m).load_module().PyTorchAudioModel

model = Model(num_labels=len(labels)).eval()
state = torch.load(w, map_location="cpu")
model.load_state_dict(state)
# x: tensor (1,13,100) → probs = softmax(model(x), dim=1)