Search

Abstract Phonetic analysis of speech, in general, requires the alignment of audio samples to its phonetic transcription. This could be done manually for a couple of files, but as the corpus grows large, it becomes infeasibly time-consuming. This paper describes the evolution process toward creating free resources for phonetic alignment in Brazilian Portuguese (BP) using Kaldi, a toolkit that achieves state of the art for open-source speech recognition, within a toolkit we call UFPAlign. The contributions of this work are then twofold: developing resources to perform forced alignment in BP, including the release of scripts to train acoustic models via Kaldi, as well as the resources themselves under open licenses; and bringing forth a comparison to other two phonetic aligners that provide resources for BP, namely EasyAlign and Montreal Forced Aligner (MFA), the latter being also Kaldi-based. Evaluation took place in terms of phone boundary and intersection over union metrics over a dataset of 385 hand-aligned utterances, and results show that Kaldi-based aligners perform better overall, and that UFPAlign models are more accurate than MFA’s. Furthermore, complex deep-learning-based approaches still do not improve performance compared to simpler models.

Kriging is a geostatistical interpolation technique that performs the prediction of observations in unknown locations through previously collected data. The modelling of the variogram is an essential step of the kriging process because it drives the accuracy of the interpolation model. The conventional method of variogram modelling consists of using specialized knowledge and in-depth study to determine which parameters are suitable for the theoretical variogram. However, this situation is not always possible, and, in this case, it becomes interesting to use an automatic process. Thus, this work aims to propose a new methodology to automate the estimation of theoretical variogram parameters of the kriging process. The proposed methodology is based on preprocessing techniques, data clustering, genetic algorithms, and the K-Nearest Neighbor classifier (KNN). The performance of the methodology was evaluated using two databases, and it was compared to other optimization techniques widely used in the literature. The impacts of the clustering step on the stationary hypothesis were also investigated with and without trend removal techniques. The results showed that, in this automated proposal, the clustering process increases the accuracy of the kriging prediction. However, it generates groups that might not be stationary. Genetic algorithms are easily configurable with the proposed heuristic when setting the variable ranges in comparison to other optimization techniques, and the KNN method is satisfactory in solving some problems caused by the clustering task and allocating unknown points into previously determined clusters.