Dimitra Vergyri

Standard keyword spotting based on Automatic Speech Recognition (ASR) cannot be used on low- and no-resource languages due to lack of annotated data and/or linguistic resources.  In recent years, query-by-example (QbE) has emerged as an alternate way to enroll and find spoken queries in large audio corpora,  yet  mismatched  and  unseen  acoustic  conditions  remain a  difficult  challenge  given  the  lack  of  enrollment  data.  This paper revisits two neural network architectures developed for noise  and  channel-robust  ASR,  and  applies  them  to  building a  state-of-art  multilingual  QbE  system.   By  applying  convolution in time or frequency across the spectrum, those convolutional bottlenecks learn more discriminative deep bottleneck features.  In conjunction with dynamic time warping (DTW), these features enable robust QbE systems.  We use the MediaEval 2014 QUESST data to evaluate robustness against language  and  channel  mismatches,  and  add  several  levels  of  artificial  noise  to  the  data  to  evaluate  performance  in  degraded acoustic environments.  We also assess performance on an Air Traffic Control QbE task with more realistic and higher levels of distortion in the push-to-talk domain.

Articulatory information can effectively model variability in speech and can improve speech recognition performance under varying acoustic conditions.  Learning speaker-independent articulatory models has always been challenging, as speaker-specific information in the articulatory and acoustic spaces increases the complexity of the speech-to-articulatory space inverse modeling, which is already an ill-posed problem due to its inherent nonlinearity and  non-uniqueness.  This  paper  investigates using deep neural networks (DNN) and convolutional neural networks (CNNs) for mapping speech data into its corresponding articulatory space.  Our results indicate that the CNN models perform better than their DNN counterparts for speech inversion.  In addition, we used the inverse models to generate articulatory trajectories from speech for three different standard speech recognition tasks.  To effectively model the articulatory features’ temporal modulations while retaining the acoustic features’ spatiotemporal signatures, we explored a joint modeling strategy to simultaneously learn both the acoustic and articulatory spaces.  The results from multiple speech recognition tasks indicate that articulatory features can improve recognition performance when the acoustic and articulatory spaces are jointly learned with one common objective function.

Speech recognition in varying background conditions is a challenging problem. Acoustic condition mismatch between training and evaluation data can significantly reduce recognition performance. For mismatched conditions, data-adaptation techniques are typically found to be useful, as they expose the acoustic model to the new data condition(s). Supervised adaptation techniques usually provide substantial performance improvement, but such gain is contingent on having labeled or transcribed data, which is often unavailable. The alternative is unsupervised adaptation, where feature-transform methods and model-adaptation techniques are typically explored. This work investigates robust features, feature-space maximum likelihood linear regression (fMLLR) transform, and deep convolutional nets to address the problem of unseen channel and noise conditions. In addition, the work investigates bottleneck (BN) features extracted from deep autoencoder (DAE) networks trained by using acoustic features extracted from the speech signal. We demonstrate that such representations not only produce robust systems but also that they can be used to perform data selection for unsupervised model adaptation. Our results indicate that the techniques presented in this paper significantly improve performance of speech recognition systems in unseen channel and noise conditions.

This work addresses lexical unit discovery for languages without (usable) written resources. Previous work has addressed this problem using entirely unsupervised methodologies.  Our approach in contrast investigates the use of linguistic and speaker knowledge which are often available even if text resources are not.  We create a framework that benefits from such resources, not assuming orthographic representations and avoiding generation of word-level transcriptions.  We adapt a universal phone recognizer to the target language and use it to convert audio into a searchable phone string for lexical unit discovery via fuzzy sub-string matching.  Linguistic knowledge is used to constrain phone recognition output and to constrain lexical unit discovery on the phone recognizer output.
Target language speakers are used to assist a linguist in creating phonetic transcriptions for the adaptation of acoustic and language models, by respeaking more clearly a small portion of the target language audio.  We also explore robust features and feature transform through deep auto-encoders for better phone recognition performance.
The proposed approach achieves lexical unit discovery performance comparable to state-of-the-art zero-resource methods.  Since the system is built on phonetic recognition, discovered units are immediately interpretable.  They can be used to automatically populate a pronunciation lexicon and enable iterative improvement through additional feedback from target language speakers.