--- library_name: pytorch tags: - audio - spoofing-detection - anti-spoofing - wav2vec2 - aasist license: apache-2.0 pipeline_tag: audio-classification model-index: - name: spectra_aasist results: - task: type: Speech Antispoofing dataset: name: ASVspoof19_LA type: ASVspoof19_LA metrics: - name: Equal Error Rate type: Equal Error Rate value: 0.159 - task: type: Speech Antispoofing dataset: name: ASVspoof21_LA type: ASVspoof21_LA metrics: - name: Equal Error Rate type: Equal Error Rate value: 5.164 - task: type: Speech Antispoofing dataset: name: ASVspoof21_DF type: ASVspoof21_DF metrics: - name: Equal Error Rate type: Equal Error Rate value: 2.568 - task: type: Speech Antispoofing dataset: name: ASVspoof5 type: ASVspoof5 metrics: - name: Equal Error Rate type: Equal Error Rate value: 14.056 - task: type: Speech Antispoofing dataset: name: ADD2022 type: ADD2022 metrics: - name: Equal Error Rate type: Equal Error Rate value: 15.205 - task: type: Speech Antispoofing dataset: name: In-the-Wild type: In-the-Wild metrics: - name: Equal Error Rate type: Equal Error Rate value: 1.461 - task: type: Speech Antispoofing dataset: name: AD2R1 type: AD2R1 metrics: - name: Equal Error Rate type: Equal Error Rate value: 0.939 - task: type: Speech Antispoofing dataset: name: AD2R2 type: AD2R2 metrics: - name: Equal Error Rate type: Equal Error Rate value: 1.802 - task: type: Speech Antispoofing dataset: name: AD3R1 type: AD3R1 metrics: - name: Equal Error Rate type: Equal Error Rate value: 6.502 - task: type: Speech Antispoofing dataset: name: AD3R2 type: AD3R2 metrics: - name: Equal Error Rate type: Equal Error Rate value: 14.481 --- ## Model Card: Spectra-AASIST (anti-spoofing / bonafide vs spoof) `Spectra-AASIST` is a model for **speech spoofing detection** (binary classification: `bonafide` vs `spoof`) from **raw audio waveforms**. Architecture: SSL encoder (`Wav2Vec2`) → MLP projection → `AASIST` 2-class classifier. - **Input**: waveform \(float32\), shape `(batch, num_samples)` (typically 16 kHz). - **Output**: logits of shape `(batch, 2)`, where **index 0 = spoof**, **index 1 = bonafide**. On first run, the model will automatically download the SSL encoder `facebook/wav2vec2-xls-r-300m` via `transformers`. ## Evaluation Results | Model | ASVspoof19 LA | ASVspoof21 LA | ASVspoof21 DF | ASVspoof5 | ADD2022 | In-the-Wild | AD2R1 | AD2R2 | AD3R1 | AD2R2 | |-----------|--------|--------|--------|--------|--------|--------|--------|--------|--------|--------| | [Res2TCNGuard](https://github.com/mtuciru/Res2TCNGuard) | 7.487 | 19.130 | 19.883 | 37.620 | 49.538 | 49.246 | 34.683 | 35.343 | 48.051 | 39.558 | | [AASIST3](https://huggingface.co/lab260/AASIST3) | 27.585 | 37.407 | 33.099 | 41.001 | 47.192 | 39.626 | 36.581 | 37.351 | 41.333 | 44.278 | | [XSLS](https://github.com/QiShanZhang/SLSforASVspoof-2021-DF) | 0.231 | 7.714 | 4.220 | 17.688 | 33.951 | 7.453 | 14.386 | 15.743 | 19.368 | 21.095 | | [TCM-ADD](https://github.com/ductuantruong/tcm_add) | **0.152** | 6.655 | 3.444 | 19.505 | 35.252 | 7.767 | 16.951 | 17.688 | 21.913 | 18.627 | | [DF Arena 1B](https://huggingface.co/Speech-Arena-2025/DF_Arena_1B_V_1) | 43.793 | 40.137 | 42.994 | 35.333 | 42.139 | 17.598 | 12.442 | 13.292 | 33.381 | 43.42 | | [Spectra-0](https://huggingface.co/lab260/spectra_0) | 0.181 | 6.475 | 5.410 | 14.426 | **14.716** | 1.026 | 1.578 | 2.372 | 6.535 | 15.154 | | Spectra-AASIST | 0.159 | 5.164 | 2.568 | 14.056 | 15.205 | 1.461 | 0.939 | **1.802** | **6.427** | **12.968** | | **[Spectra-AASIST3](https://huggingface.co/lab260/Spectra-AASIST3)** | 0.723 | **4.506** | **1.998** | **13.82** | 15.187 | **0.961** | **0.727** | 1.806 | 6.502 | 14.481 | ## Quickstart ### Clone from Hugging Face This repository is hosted on Hugging Face Hub: `https://huggingface.co/lab260/spectra_aasist`. ```bash git lfs install git clone https://huggingface.co/lab260/spectra_aasist cd spectra_aasist ``` ### Install dependencies ```bash pip install -U torch torchaudio transformers huggingface_hub safetensors soundfile ``` ### Single-file inference (example preprocessing) ```python import random import torch import torchaudio import soundfile as sf from model import spectra_aasist def pad_random(x: torch.Tensor, max_len: int = 64600) -> torch.Tensor: # x: (num_samples,) or (1, num_samples) if x.ndim > 1: x = x.squeeze() x_len = x.shape[0] if x_len >= max_len: start = random.randint(0, x_len - max_len) return x[start:start + max_len] num_repeats = int(max_len / x_len) + 1 return x.repeat(num_repeats)[:max_len] def load_audio_mono(path: str) -> torch.Tensor: audio, sr = sf.read(path, dtype="float32") audio = torch.from_numpy(audio) if audio.ndim > 1: # (num_samples, channels) -> mono audio = audio.mean(dim=1) if sr != 16000: audio = torchaudio.functional.resample(audio, sr, 16000) return audio device = "cuda" if torch.cuda.is_available() else "cpu" model = spectra_aasist.from_pretrained(pretrained_model_name_or_path=".").eval().to(device) audio = load_audio_mono("path/to/audio.wav") audio = torchaudio.functional.preemphasis(audio.unsqueeze(0)) # (1, T) audio = pad_random(audio.squeeze(0), 64600).unsqueeze(0) # (1, 64600) with torch.inference_mode(): logits = model(audio.to(device)) # (1, 2) score_spoof = logits[0, 0].item() score_bonafide = logits[0, 1].item() print({"score_bonafide": score_bonafide, "score_spoof": score_spoof}) ``` ## Threshold-based classification (and how to tune it) In `model.py`, the `SpectraAASIST` class provides `classify()` with a **default threshold** chosen as an “optimal” value for the original setting: - **Default threshold**: `-1.140625` (it thresholds `logit_bonafide = logits[:, 1]`) - **Note**: this threshold **may not be optimal** on a different dataset/domain. It’s recommended to tune the threshold on your dataset using **EER** (Equal Error Rate) or a target FAR/FRR. Example: ```python with torch.inference_mode(): pred = model.classify(audio.to(device), threshold=-1.140625) # 1=bonafide, 0=spoof ``` ### Tuning the threshold via EER (typical workflow) 1) Run the model on a labeled set and collect scores for both classes. 2) Compute EER and the threshold ## Limitations and notes - This is a **pre-release** model. - Significantly stronger models are planned for **Q3–Q4 2026** — stay tuned. ## License MIT (see the `license` field in the model repo header). ## Contacts TG channel: https://t.me/korallll_ai email: k.n.borodin@mtuci.ru website: https://lab260.ru/