AudioMosaic — ESC-50 (fold 1) Classifier

This is the AudioMosaic ViT-B/16 encoder fine-tuned on ESC-50 (fold 1) for single-label audio classification over 50 classes.

Metric	Value
Accuracy	97.25

Model Details

Architecture: ViT-B/16 with linear classifier head
Embedding dim: 768, Depth: 12, Heads: 12
Input: log-mel spectrogram of size 1024 × 128
Patch size: 16 × 16
Pooling: average over patch tokens
Classes: 50

Model Usage

import sys, torch
from huggingface_hub import snapshot_download

local_dir = snapshot_download("hanxunh/AudioMosaic-vit-b16-finetune-esc-split1")
sys.path.insert(0, local_dir)

from load_model import load_classifier
model = load_classifier(device="cuda")

# Forward a log-mel spectrogram batch of shape [B, 1, 1024, 128]
fbank = torch.randn(2, 1, 1024, 128).cuda()
with torch.no_grad():
    logits = model(fbank)              # [B, 50]
    pred = logits.argmax(dim=-1)       # predicted class id

The release contains:

model.safetensors — fine-tuned classifier weights
config.json — architecture hyperparameters
modeling.py — vendored model architecture (no need to install AudioMosaic)
load_model.py — convenience loader

Required dependencies: torch, timm, torchlibrosa, safetensors, huggingface_hub.

Citation

@inproceedings{huang2026audiomosaic,
  title={AudioMosaic: Contrastive Masked Audio Representation Learning},
  author={Hanxun Huang and Qizhou Wang and Xingjun Ma and Cihang Xie and Christopher Leckie and Sarah Erfani},
  booktitle={ICML},
  year={2026}
}