Multi-style Training for South African Call Centre Audio

TL;DR

This paper investigates how multi-style training (MTR) can improve the robustness of speech recognition systems in mismatched and noisy real-world conditions, specifically for South African call center audio.

Contribution

It systematically evaluates the impact of various MTR styles on ASR performance under different testing conditions using a controlled LibriSpeech environment and real-world call center data.

Findings

Certain MTR styles significantly improve recognition accuracy in noisy conditions

MTR enhances robustness of DNN-HMM ASR systems against mismatched data

Performance varies depending on the style and test environment

Abstract

Mismatched data is a challenging problem for automatic speech recognition (ASR) systems. One of the most common techniques used to address mismatched data is multi-style training (MTR), a form of data augmentation that attempts to transform the training data to be more representative of the testing data; and to learn robust representations applicable to different conditions. This task can be very challenging if the test conditions are unknown. We explore the impact of different MTR styles on system performance when testing conditions are different from training conditions in the context of deep neural network hidden Markov model (DNN-HMM) ASR systems. A controlled environment is created using the LibriSpeech corpus, where we isolate the effect of different MTR styles on final system performance. We evaluate our findings on a South African call centre dataset that contains noisy, WAV49-encoded audio.