DNN-Based Score Calibration With Multitask Learning for Noise Robust Speaker Verification

Abstract

This paper proposes and investigates several deep neural network (DNN)-based score compensation, transformation and calibration algorithms for enhancing the noise robustness of i-vector speaker verification systems. Unlike conventional calibration methods where the required score shift is a linear function of SNR or log-duration, the DNN approach learns the complex relationship between the score shifts and the combination of i-vector pairs and uncalibrated scores. Furthermore, with the flexibility of DNNs, it is possible to explicitly train a DNN to recover the clean scores without having to estimate the score shifts. To alleviate the overfitting problem, multi-task learning is applied to incorporate auxiliary information such as SNRs and speaker ID of training utterances into the DNN. Experiments on NIST 2012 SRE show that score calibration derived from multi-task DNNs can improve the performance of the conventional score-shift approch significantly, especially under noisy conditions.

i-Vector DNN Scoring and Calibration for Noise Robust Speaker Verification

“…In [25], we proposed to estimate the score shifts by multitask DNNs using noisy i-vector pairs and their corresponding PLDA scores as input. Moreover, instead of expressing the score shifts as a linear function of SNRs, we used the SNRs of training utterances as part of the target outputs and applied multi-task learning to guide the network to produce the ideal score shifts or clean scores.…”

“…Moreover, instead of expressing the score shifts as a linear function of SNRs, we used the SNRs of training utterances as part of the target outputs and applied multi-task learning to guide the network to produce the ideal score shifts or clean scores. In this paper, we extend the multitask DNNs in [25] in three respects. First, in addition to using SNR as target outputs, we also use utterance duration and same-speaker and different-speaker hypotheses as target outputs.…”

Rapid field trace detection of pesticide residue in food based on surface-enhanced Raman spectroscopy

“…It is suitable to choose WT method when the SNR is high, while CNN method performs well when the SNR is low. Besides, a certain amount of noise in SERS spectra can enhance the robustness and generalization of arithmetic [186]. Therefore, some noise of original SERS signal does not need to be removed in the process of qualitative training model.…”