Unsupervised Linear Discriminant Analysis for Supporting DPGMM Clustering in the Zero Resource Scenario

By Michael Heck, Sakriani Watiasri Sakti, Satoshi Nakamura. 2016

36 citations recorded by the source corpus.

Abstract In this work we make use of unsupervised linear discriminant analysis (LDA) to support acoustic unit discovery in a zero resource scenario. The idea is to automatically find a mapping of feature vectors into a subspace that is more suitable for Dirichlet process Gaussian mixture model (DPGMM) based clustering, without the need of supervision. Supervised acoustic modeling typically makes use of feature transformations such as LDA to minimize intra-class discriminability, to maximize inter-class discriminability and to extract relevant informations from high-dimensional features spanning larger contexts. The need of class labels makes it difficult to use this technique in a zero resource setting where the classes and even their amount are unknown. To overcome this issue we use a first iteration of DPGMM clustering on standard features to generate labels for the data, that serve as basis for learning a proper transformation. A second clustering operates on the transformed features. The application of unsupervised LDA demonstrably leads to better clustering results given the unsupervised data. We show that the improved input features consistently outperform our baseline input features.

Subfields: Artificial Intelligence, Representation & Self-Supervised Learning, Probabilistic & Bayesian Methods, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Representation & Self-Supervised Learning

In Probabilistic & Bayesian Methods

In Signal Processing

Citations in the corpus

Unsupervised Speech Representation Learning Using WaveNet Autoencoders (2019, methodology): “Nevertheless, the multilingual prequantized features yield accuracies comparable to [58].”
Feature optimized DPGMM clustering for unsupervised subword modeling: A contribution to zerospeech 2017 (2017, methodology): “To tackle this demanding task, we expanded our previous work that we started in the aftermath of the first zero resource speech challenge [7, 8, 9], which in turn drew from the findings of Chen et al.”
Multilingual bottle-neck feature learning from untranscribed speech (2017, unknown): “[32, 33] to improve DPGMM posteriorgrams.”
Early phonetic learning without phonetic categories: Insights from large-scale simulations on realistic input (2021, methodology): “Following previous work (60, 65), model initialization was performed by partitioning training points uniformly at random into 10 clusters, and the hyperparameters were set as follows: α to 1, μ0 to the average of all input features vectors, λ to 1, λ0 to the inverse of the covariance of all input fe”
Measuring the perceptual availability of phonological features during language acquisition using unsupervised binary stochastic autoencoders (2019, background): “…al., 2015; Agenbag and Niesler, 2015; Chen et al., 2015; Baljekar et al., 2015; Räsänen et al., 2015; Lyzinski et al., 2015; Zeghidour et al., 2016; Heck et al., 2016; Srivastava and Shrivastava, 2016; Kamper et al., 2017b; Chen et al., 2017; Yuan et al., 2017; Heck et al., 2017; Shibata et al.,…”
ABX-Discriminability Measures and Applications. (Mesures de discriminabilité ABX et applications) (2016, background): “9% ABX error rate in the across-speaker task on the Buckeye corpus) were largely above the raw acoustic features baseline, but remained inferior to the results with DPGMM from [106, 107] and those obtained with NN trained based on top-down information from STD [101–103].”
Virtual Phone Discovery for Speech Synthesis Without Text (2019, methodology): “were used for Dirichlet process Gaussian mixture model clustering for acoustic segment modeling [7].”
An empirical evaluation of zero resource acoustic unit discovery (2017, unknown): “imilar LDA strategy but with target labels acquired by the ﬁrst-pass acoustic unit tokenizations, which is considered as a self-supervised fashion to solve the unknown label problem. Previous work in [5] also exploits such unsupervised LDA to support Dirichlet process Gaussian mixture model (DPGMM) ”
Optimizing DPGMM Clustering in Zero Resource Setting Based on Functional Load (2018, methodology): “For comparison, we use the same parameter setting as [2, 3] for DPGMM sampling; we use officially provided voice activity detection segmentations (2.”