End-to-end recurrent denoising autoencoder embeddings for speaker identification

Speech -in-the-wild- is a handicap for speaker recognition systems due to the variability induced by real-life conditions, such as environmental noise and the emotional state of the speaker. Taking advantage of the principles of representation learning, we aim to design a recurrent denoising autoencoder that extracts robust speaker embeddings from noisy spectrograms to perform speaker identification. The end-to-end proposed architecture uses a feedback loop to encode information regarding the speaker into low-dimensional representations extracted by a spectrogram denoising autoencoder. We employ data augmentation techniques by additively corrupting clean speech with real-life environmental noise in a database containing real stressed speech. Our study presents that the joint optimization of both the denoiser and speaker identification modules outperforms independent optimization of both components under stress and noise distortions as well as handcrafted features.The authors would like to thank the rest of the members of the UC3M4Safety for their support and NVIDIA Corporation for the donation of a TITAN Xp. This work has been partially supported by the Dept. of Research and Innovation of Madrid Regional Authority (EMPATIA-CM Y2018/TCS-5046) and the Dept. of Education and Research of Madrid Regional Authority with a European Social Fund for the Pre-doctoral Research Staff grant for Research Activities, within the CAM Youth Employment Programme (PEJD-2019-PRE/TIC-16295)