Analysis of DNN Speech Signal Enhancement for Robust Speaker Recognition

Analysis of DNN Speech Signal Enhancement for Robust Speaker Recognition

Článek WoS

In this work, we present an analysis of a DNN-based autoencoder for speech enhancement, dereverberation and denoising. Thetarget application is a robust speaker verification (SV) system. We start our approach by carefully designing a data augmentationprocess to cover a wide range of acoustic conditions and to obtain rich training data for various components of our SV system.We augment several well-known databases used in SV with artificially noised and reverberated data and we use them to train adenoising autoencoder (mapping noisy and reverberated speech to its clean version) as well as an x-vector extractor which is cur-rently considered as state-of-the-art in SV. Later, we use the autoencoder as a preprocessing step for a text-independent SV sys-tem. We compare results achieved with autoencoder enhancement, multi-condition PLDA training and their simultaneous use.We present a detailed analysis with various conditions of NIST SRE 2010, 2016, PRISM and with re-transmitted data. We con-clude that the proposed preprocessing can significantly improve both i-vector and x-vector baselines and that this technique canbe used to build a robust SV system for various target domains.

In this work, we present an analysis of a DNN-based autoencoder for speech enhancement, dereverberation and denoising. Thetarget application is a robust speaker verification (SV) system. We start our approach by carefully designing a data augmentationprocess to cover a wide range of acoustic conditions and to obtain rich training data for various components of our SV system.We augment several well-known databases used in SV with artificially noised and reverberated data and we use them to train adenoising autoencoder (mapping noisy and reverberated speech to its clean version) as well as an x-vector extractor which is cur-rently considered as state-of-the-art in SV. Later, we use the autoencoder as a preprocessing step for a text-independent SV sys-tem. We compare results achieved with autoencoder enhancement, multi-condition PLDA training and their simultaneous use.We present a detailed analysis with various conditions of NIST SRE 2010, 2016, PRISM and with re-transmitted data. We con-clude that the proposed preprocessing can significantly improve both i-vector and x-vector baselines and that this technique canbe used to build a robust SV system for various target domains.

Speakerverification; Signalenhancement; Autoencoder; Neuralnetwork; Robustness; Embedding

Speakerverification; Signalenhancement; Autoencoder; Neuralnetwork; Robustness; Embedding

NOVOTNÝ, O.; PLCHOT, O.; GLEMBEK, O.; ČERNOCKÝ, J.; BURGET, L.

2020

09.06.2019

0885-2308

COMPUTER SPEECH AND LANGUAGE

2019

58

Spojené království Velké Británie a Severního Irska

403

421

19

https://www.sciencedirect.com/science/article/pii/S0885230818303607

novotny_elsevier_Journal_Paper_2019