Improving Unsupervised Sparsespeech Acoustic Models with Categorical Reparameterization

TL;DR

This paper enhances the Sparsespeech unsupervised acoustic model by integrating categorical reparameterization via Gumbel-Softmax, enabling controllable sparsity and improved speech recognition performance on Libri-Light data.

Contribution

It introduces a novel categorical reparameterization technique to Sparsespeech, allowing for efficient sampling and sparsity control post-training, leading to better ASR results.

Findings

Up to 31.4% relative ABX error reduction on 600h data.

Significant improvements with larger training data (6000h).

Effective sparsity control after training.

Abstract

The Sparsespeech model is an unsupervised acoustic model that can generate discrete pseudo-labels for untranscribed speech. We extend the Sparsespeech model to allow for sampling over a random discrete variable, yielding pseudo-posteriorgrams. The degree of sparsity in this posteriorgram can be fully controlled after the model has been trained. We use the Gumbel-Softmax trick to approximately sample from a discrete distribution in the neural network and this allows us to train the network efficiently with standard backpropagation. The new and improved model is trained and evaluated on the Libri-Light corpus, a benchmark for ASR with limited or no supervision. The model is trained on 600h and 6000h of English read speech. We evaluate the improved model using the ABX error measure and a semi-supervised setting with 10h of transcribed speech. We observe a relative improvement of up to 31.4% on ABX error rates across speakers on the test set with the improved Sparsespeech model on 600h of speech data and further improvements when we scale the model to 6000h.