Fast i-vector denoising using MAP estimation and a noise distributions database for robust speaker recognition

Abstract

Once the i-vector paradigm has been introduced in the field of speaker recognition, many techniques have been proposed to deal with additive noise within this framework. Due to the complexity of its effect in the i-vector space, a lot of effort has been put into dealing with noise in other domains (speech enhancement, feature compensation, robust i-vector extraction and robust scoring). As far as we know, there was no serious attempt to handle the noise problem directly in the i-vector space without relying on data distributions computed on a prior domain. The aim of this paper is twofold. First, it proposes a fullcovariance Gaussian modeling of the clean i-vectors and noise distribution in the i-vector space and introduces a technique to estimate a clean i-vector given the noisy version and the noise density function using the MAP approach. Based on NIST data, we show that it is possible to improve by up to 60% the baseline system performance. Second, in order to make this algorithm usable in a real application and reduce the computational time needed by i-MAP, we propose an extension that requires building a noise distribution database in the i-vector space in an off-line step and using it later in the test phase. We show that it is possible to achieve comparable results using this approach (up to 57% of relative EER improvement) with a sufficiently large noise distribution database.

Compensate multiple distortions for speaker recognition systems

“…In the previous generation of speaker recognition systems (i.e., i-vector) the mapping from distorted to clean vectors is explored broadly. i-MAP [14] and joint i-MAP [15] are two statistical techniques used for noise compensation in the i-vector framework.…”

Denoising x-vectors for Robust Speaker Recognition

“…I-MAP is a statistical denoising method that is applied in the ivector space. The main advantage of this method is that it uses both information about the relation between noisy and clean speech and the clean speech distribution [5]. A nonparametric algorithm without considering the relation between corrupted and clean i-vector was proposed by [4], that utilizes the joint distribution of corrupted and clean i-vectors to denoise corrupted i-vector with an MMSE estimator.…”

“…Applying denoising techniques at the speaker modeling level has been done successfully in the i-vector space [4,5,6]. In this paper we apply statistical denoising techniques on xvectors that works effectively in i-vector domain.…”

“…Although, we want to explore the effectiveness of DNN denoising techniques in x-vector domain. Our first attempt consists of using a statistical approach based on the use of maximum a posteriori (MAP), namely the i-MAP approach [5]. The i-MAP denoising technique has been used successfully in the i-vector space.…”

Cam: Context-Aware Masking for Robust Speaker Verification

“…The first methodology is to apply a denoising transformation to speaker embeddings. In [7,8,9], researchers use either statistical back end or neural network back end to transform noisy speaker embeddings into enhanced ones. A problem with this methodology is information loss.…”