Jabberjay 0.0.4

🦜 Synthetic Voice Detection

Jabberjay

🦜 Synthetic Voice Detection

Models

Vision Transformer

Name	Model	Dataset	Visualisation	Model
MattyB95/VIT-ASVspoof2019-ConstantQ-Synthetic-Voice-Detection	ViT	ASVspoof2019	ConstantQ	Hugging Face
MattyB95/VIT-ASVspoof2019-Mel_Spectrogram-Synthetic-Voice-Detection	ViT	ASVspoof2019	MelSpectrogram	Hugging Face
MattyB95/VIT-ASVspoof2019-MFCC-Synthetic-Voice-Detection	ViT	ASVspoof2019	MFCC	Hugging Face
MattyB95/VIT-ASVspoof5-ConstantQ-Synthetic-Voice-Detection	ViT	ASVspoof5	ConstantQ	Hugging Face
MattyB95/VIT-ASVspoof5-Mel_Spectrogram-Synthetic-Voice-Detection	ViT	ASVspoof5	MelSpectrogram	Hugging Face
MattyB95/VIT-ASVspoof5-MFCC-Synthetic-Voice-Detection	ViT	ASVspoof5	MFCC	Hugging Face
MattyB95/VIT-VoxCelebSpoof-ConstantQ-Synthetic-Voice-Detection	ViT	VoxCelebSpoof	ConstantQ	Hugging Face
MattyB95/VIT-VoxCelebSpoof-Mel_Spectrogram-Synthetic-Voice-Detection	ViT	VoxCelebSpoof	MelSpectrogram	Hugging Face
MattyB95/VIT-VoxCelebSpoof-MFCC-Synthetic-Voice-Detection	ViT	VoxCelebSpoof	MFCC	Hugging Face

Audio Spectrogram Transformer

Name	Model	Dataset	Model
MattyB95/AST-ASVspoof2019-Synthetic-Voice-Detection	AST	ASVspoof2019	Hugging Face
MattyB95/AST-ASVspoof5-Synthetic-Voice-Detection	AST	ASVspoof5	Hugging Face
MattyB95/AST-VoxCelebSpoof-Synthetic-Voice-Detection	AST	VoxCelebSpoof	Hugging Face

Other

Name	Paper	Codebase	Model
Classical	Placeholder	Placeholder	Placeholder
RawNet2	End-to-End anti-spoofing with RawNet2	rawnet2-antispoofing	pre_trained_DF_RawNet2.zip

Usage

Command Line Interface

usage: Jabberjay [-h] [-m {AST,Classical,RawNet2,VIT}]
                 [-d {ASVspoof2019,ASVspoof5,VoxCelebSpoof}]
                 [-vis {ConstantQ,MelSpectrogram,MFCC}] [-v]
                 audio

Python API

from Jabberjay.Utilities.enum_handler import Visualisation, Model, Dataset
from Jabberjay.jabberjay import Jabberjay

jabberjay = Jabberjay()

bonafide = jabberjay.load(filename="../res/bonafide/bonafide.flac")
spoof = jabberjay.load(filename="../res/spoof/spoof.flac")

jabberjay.detect(audio=bonafide, model=Model.VIT, visualisation=Visualisation.ConstantQ, dataset=Dataset.VoxCelebSpoof)
jabberjay.detect(audio=spoof, model=Model.VIT, visualisation=Visualisation.ConstantQ, dataset=Dataset.VoxCelebSpoof)

Contributing to Jabberjay

🌟 We value your contributions!

Whether you're fixing a bug, improving the documentation, or proposing a new feature, we're delighted to have you as part of the Jabberjay community. Your efforts help us make Synthetic Voice Detection even better for everyone.

We especially welcome and encourage additional models for speech deepfake (bonafide vs. spoof) detection, with the aim of making Jabberjay the one-stop shop for state-of-the-art models in the field.

We are truly grateful for your interest in improving Jabberjay. Your contributions, no matter how big or small, make our open-source community a vibrant place to learn, inspire, and create.

Let's make Jabberjay the best tool for Synthetic Voice Detection together! 🚀

Acknowledgement

This work was supported, in whole or in part, by the Bill & Melinda Gates Foundation [INV-001309].