AASIST

AASIST audio anti-spoofing (voice-deepfake detection) countermeasure from "AASIST: Audio Anti-Spoofing using Integrated Spectro-Temporal Graph Attention Networks" (Jung et al., ICASSP 2022). This is the official AASIST variant (not AASIST-L), using the upstream clovaai/aasist ASVspoof2019 LA pretrained checkpoint. The model takes a raw speech waveform and returns a score where higher = more bona fide.

• Code: https://github.com/clovaai/aasist
• Paper: https://arxiv.org/abs/2110.01200
• Parameters: 297,866 (0.298 M)
• Checkpoint: AASIST.pth

This repo is self-contained for inference: the network definition is in _net.py and the exact wrapper used to produce the Arena scores in aasist.py.

Architecture

AASIST operates directly on the raw waveform: a sinc-convolution front-end and a RawNet2-style residual encoder produce a spectro-temporal feature map, which is modelled by heterogeneous stacking graph attention layers over spectral and temporal sub-graphs with a learnable max/average readout, followed by a 2-class output (bona fide vs. spoof). The Arena score is the bona-fide logit.

Reproducing the Arena scores

Inference uses a deterministic first-64600-sample window (no random crop), matching the upstream data_utils.pad() used at eval. Audio is provided as float32 mono at 16 kHz (no resampling in the wrapper).

from aasist import AASIST
m = AASIST(); m.load()
scores = m.score_batch([wav], [16000])   # higher = more bona fide

Dataset	EER %	n_trials
ASVspoof2019_LA (in-domain)	0.83	71,237
ASVspoof2021_LA	12.35	181,566
ASVspoof2021_DF	17.04	611,829
InTheWild	43.01	31,779
CD-ADD	51.05	20,786

The in-domain ASVspoof2019 LA result reproduces the paper's reported EER (~0.83%).

License

MIT (inherited from clovaai/aasist; see LICENSE).