Aller au contenu
ModelPrésentéExterne
Télécharger

Lcnn

> Improved version available: caa-speech-detection-asvspoof2019/lcnn-v7-cqt — lcnnv7 (CQT + label-smoothing + cosine + grad-clip) achieves 3.26% eval EER (vs 8.43% here) and 0.4930 tandem min t-DCF.

Sourcehuggingface.co/caa-speech-detection-asvspoof2019/lcnn/tree/c99b2ac65c39e387cbbf1b7c64302274348a2028
TypeOtherVersionRevision c99b2acBClasse

Source du modèle

Type
Other
Taille du fichier
3,44 MB
Version
vRevision c99b2ac
Mis à jour
25 oct. 2026
Téléchargements
6
Source
Hugging Face

Description de la source

Improved version available: caa-speech-detection-asvspoof2019/lcnn-v7-cqt — lcnn_v7 (CQT + label-smoothing + cosine + grad-clip) achieves 3.26% eval EER (vs 8.43% here) and 0.4930 tandem min t-DCF.

LCNN — ASVspoof 2019 LA Countermeasure

Light CNN for binary classification of bonafide vs spoofed speech, trained on the ASVspoof 2019 Logical Access (LA) dataset.

This is one of three models compared in our study (LCNN, RawNet2, Wav2Vec 2.0) under identical training/evaluation conditions.

Architecture

  • 2D CNN over LFCC features
  • Reference: Lavrentyeva et al., "Audio Replay Attack Detection with Deep Learning Frameworks", Interspeech 2017
InputLFCC (60 coefficients, 512 FFT, 160 hop, ~4 s audio)
Channels[32, 48, 64, 128]
Kernel sizes[5, 5, 3, 3]
FC hidden64
Dropout0.3

See config.yaml for the full training/model configuration.

Training

  • Dataset: ASVspoof 2019 LA train split (~25k utterances)
  • Batch size: 128
  • Learning rate: 1e-4, cosine schedule
  • Gradient clipping: 1.0
  • Sample rate: 16 kHz mono
  • No data augmentation

Results

Baseline to beat: EER 8.09% (LFCC+GMM).

SplitEERtandem min t-DCFIn-the-Wild EER
Dev (baseline → improved)0.902% → 0.708%——
Eval8.43%——
Eval (improved: lcnn_v7)3.26%0.493033.41%

Trajectory and loss curves: see learning_curves.png and metrics.csv.

Note on the metric: t-DCF values use the normalized [0, 1] convention (0 = perfect, 1 = no better than trivial baseline). Tandem t-DCF uses ASV scores following the official ASVspoof 2019 formula.

Caveat on dev performance: dev shares attacks (A01–A06) with the training split. Eval-set performance against unseen attacks (A07–A19) is the meaningful generalisation metric.

Usage

import torch
# Load checkpoint
state = torch.load("best.pt", map_location="cpu")
# Plug into the LCNN model from the source repo:
# https://github.com/sebastiaoteixeira/caa-ai-generated-speech-detector

License

MIT

Sources

1 source
Hugging FacePrincipalVersion c99b2ac65c39e387cbbf1b7c64302274348a2028
huggingface.co/caa-speech-detection-asvspoof2019/lcnn/tree/c99b2ac65c39e387cbbf1b7c64302274348a2028

Vérifié 25 oct.

Artefacts du modèle

1 artefact

best.pt

pt · 3,44 MB · SHA-256 a9626bbc0ad3…a8cb · Hugging Face

Download file

Extraits de sources

3 extraits
Readme7 aoûthuggingface.co/caa-speech-detection-asvspoof2019/lcnn

--- license: mit tags: - audio - audio-classification - anti-spoofing - asvspoof - speech datasets: - asvspoof-2019-la library_name: pytorch --- > **Improved version available:** [caa-speech-detection-asvspoof2019/lcnn-v7-cqt](https://huggingface.co/caa-speech-detection-asvspoof2019/lcnn-v7-cqt) — lcnn_v7 (CQT + label-smoothing + cosine + grad-clip) achieves **3.26% eval EER** (vs 8.43% here) and **0.4930 tandem min t-DCF**. # LCNN — ASVspoof 2019 LA Countermeasure Light CNN for binary classification of bonafide vs spoofed speech, trained on the ASVspoof 2019 Logical Access (LA) dataset. This is one of three models compared in our study (LCNN, RawNet2, Wav2Vec 2.0) under identical training/evaluation conditions. ## Architecture - 2D CNN over LFCC features - Reference: Lavrentyeva et al., *"Audio Replay Attack Detection with Deep Learning Frameworks"*, Interspeech 2017 | | | |---|---| | Input | LFCC (60 coefficients, 512 FFT, 160 hop, ~4 s audio) | | Channels | `[32, 48, 64, 128]` | | Kernel sizes | `[5, 5, 3, 3]` | | FC hidden | 64 | | Dropout | 0.3 | See `config.yaml` for the full training/model configuration. ## Training - Dataset: ASVspoof 2019 LA train split (~25k utterances) - Batch size: 128 - Learning rate: 1e-4, cosine schedule - Gradient clipping: 1.0 - Sample rate: 16 kHz mono - No data augmentation ## Results Baseline to beat: **EER 8.09%** (LFCC+GMM). | Split | EER | tandem min t-DCF | In-the-Wild EER | |---|---|---|---| | Dev (baseline → improved) | 0.902% → 0.708% | — | — | | **Eval** | **8.43%** | — | — | | Eval (improved: lcnn_v7) | 3.26% | 0.4930 | 33.41% | Trajectory and loss curves: see `learning_curves.png` and `metrics.csv`. **Note on the metric**: t-DCF values use the normalized [0, 1] convention (0 = perfect, 1 = no better than trivial baseline). Tandem t-DCF uses ASV scores following the official ASVspoof 2019 formula. **Caveat on dev performance**: dev shares attacks (A01–A06) with the training split. Eval-set performance against unseen attacks (A07–A19) is the meaningful generalisation metric. ## Usage ```python import torch # Load checkpoint state = torch.load("best.pt", map_location="cpu") # Plug into the LCNN model from the source repo: # https://github.com/sebastiaoteixeira/caa-ai-generated-speech-detector ``` ## License MIT

Note25 oct.huggingface.co/caa-speech-detection-asvspoof2019/lcnn

Source context: 6 downloads · 0 likes · Pipeline audio-classification · Library pytorch · Repo caa-speech-detection-asvspoof2019/lcnn

Note7 aoûthuggingface.co/caa-speech-detection-asvspoof2019/lcnn

Source context: 4 downloads · 0 likes · Pipeline audio-classification · Library pytorch · Repo caa-speech-detection-asvspoof2019/lcnn