FastAudio: A Learnable Audio Front-End for Spoof Speech Detection

View on arXiv ← Back to list

Authors: Quchen Fu, Zhongwei Teng, Jules White, Maria Powell, Douglas C. Schmidt

Published: 2021-09-06 23:32:10+00:00

AI Summary

This paper proposes FastAudio, a learnable audio front-end for spoof speech detection. By replacing fixed filterbanks with a learnable layer, FastAudio achieves a 27% relative improvement in minimum tandem detection cost function (min t-DCF) compared to fixed front-ends on the ASVspoof 2019 dataset, outperforming other learnable front-ends.

Abstract

Voice assistants, such as smart speakers, have exploded in popularity. It is currently estimated that the smart speaker adoption rate has exceeded 35% in the US adult population. Manufacturers have integrated speaker identification technology, which attempts to determine the identity of the person speaking, to provide personalized services to different members of the same family. Speaker identification can also play an important role in controlling how the smart speaker is used. For example, it is not critical to correctly identify the user when playing music. However, when reading the user's email out loud, it is critical to correctly verify the speaker that making the request is the authorized user. Speaker verification systems, which authenticate the speaker identity, are therefore needed as a gatekeeper to protect against various spoofing attacks that aim to impersonate the enrolled user. This paper compares popular learnable front-ends which learn the representations of audio by joint training with downstream tasks (End-to-End). We categorize the front-ends by defining two generic architectures and then analyze the filtering stages of both types in terms of learning constraints. We propose replacing fixed filterbanks with a learnable layer that can better adapt to anti-spoofing tasks. The proposed FastAudio front-end is then tested with two popular back-ends to measure the performance on the LA track of the ASVspoof 2019 dataset. The FastAudio front-end achieves a relative improvement of 27% when compared with fixed front-ends, outperforming all other learnable front-ends on this task.

Key findings

FastAudio significantly outperforms fixed filterbanks and other learnable front-ends on the ASVspoof 2019 LA dataset, achieving a 27% relative improvement in min t-DCF. The learned filterbanks exhibit selectivity around formants and high-frequency regions, suggesting adaptation to relevant features for spoof detection. Shape constraints were found to be crucial for preventing overfitting.

Approach

FastAudio replaces the fixed filterbanks in traditional Short-Time Fourier Transform (STFT)-based audio front-ends with a learnable layer. This allows the system to better adapt to anti-spoofing tasks, improving performance on spoof speech detection.

Datasets

ASVspoof 2019 Logical Access (LA) dataset

Model(s)

X-vector and ECAPA-TDNN as backends; FastAudio as a learnable front-end. Comparisons were also made with other front-ends including CQT, LEAF, nnAudio, and TD-filterbanks.

Author countries

UNKNOWN

← Previous