An Analysis of the Short Utterance Problem for Speaker Characterization

Ignacio Viñals,Antonio Miguel,Alfonso Ortega,Eduardo Lleida

doi:10.3390/app9183697

Abstract

Speaker characterization has always been conditioned by the length of the evaluated utterances. Despite performing well with large amounts of audio, significant degradations in performance are obtained when short utterances are considered. In this work we present an analysis of the short utterance problem providing an alternative point of view. From our perspective the performance in the evaluation of short utterances is highly influenced by the phonetic similarity between enrollment and test utterances. Both enrollment and test should contain similar phonemes to properly discriminate, being degraded otherwise. In this study we also interpret short utterances as incomplete long utterances where some acoustic units are either unbalanced or just missing. These missing units are responsible for the speaker representations to be unreliable. These unreliable representations are biased with respect to the reference counterparts, obtained from long utterances. These undesired shifts increase the intra-speaker variability, causing a significant loss of performance. According to our experiments, short utterances (3–60 s) can perform as accurate as if long utterances were involved by just reassuring the phonetic distributions. This analysis is determined by the current embedding extraction approach, based on the accumulation of local short-time information. Thus it is applicable to most of the state-of-the-art embeddings, including traditional i-vectors and Deep Neural Network (DNN) xvectors.

Highlights

Speaker recognition is the area of speech technologies that allows the automatic recognition of the speaker’s identity given some portions of his/her speech
The KL2 distance increases for both target and non-target trials as long as we move from the Long-Long experiment to the Short-Short Balanced and the Short-Short Random experiment
We study the whole set of trials at once, providing the evaluation metrics, the Equal Error Rate (EER) and the minimum Decision Cost Function

Summary

Introduction

Speaker recognition is the area of speech technologies that allows the automatic recognition of the speaker’s identity given some portions of his/her speech. Its goal is the proper characterization of the speaker, isolating singular characteristics of his/her voice and making possible accurate comparisons among different speakers. The traditional strategy to tackle this challenge consists of the right characterization of the involved speakers, enrollment and test and a fair comparison of hypotheses (target and non-target) afterwards. This characterization must exploit the singularities in the voice of speakers regardless of the message content and the acoustic conditions such as noise or reverberation. First we illustrate the phoneme dependent estimation error due to limited data For this reason we estimate the posterior distribution of the embeddings for multiple utterances only differing the number of samples. The embeddings should not suffer any bias but its uncertainty should get larger as long as the utterances contain less data

Methods

Results

Conclusion