Explainability of neural network prediction is essential to understand feature importance and gain interpretable insight into neural network performance. In this work, model explanations are fed back to the feed-forward training to help the model generalize better. To this extent, a custom weighted loss where the weights are generated by considering the Euclidean distances between true LIME (Local Interpretable Model-Agnostic Explanations) explanations and model-predicted LIME explanations is proposed. Also, in practical training scenarios, developing a solution that can help the model learn sequentially without losing information on previous data distribution is imperative due to the unavailability of all the training data at once. Thus, the framework known as XAI-Increment incorporates the custom weighted loss developed with elastic weight consolidation (EWC), to maintain performance in sequential testing sets. Finally, the training procedure involving the custom weighted loss shows around 1% accuracy improvement compared to the traditional loss based training for the keyword spotting task on the Google Speech Commands dataset and also shows low loss of information when coupled with EWC in the incremental learning setup.
ASR model deployment environment is ever-changing, and the incoming speech can be switched across different domains during a session. This brings a challenge for effective domain adaptation when only target domain text data is available, and our objective is to obtain obviously improved performance on the target domain while the performance on the general domain is less undermined. In this paper, we propose an adaptive LM fusion approach called internal language model estimation based adaptive domain adaptation (ILME-ADA). To realize such an ILME-ADA, an interpolated log-likelihood score is calculated based on the maximum of the scores from the internal LM and the external LM (ELM) respectively. We demonstrate the efficacy of the proposed ILME-ADA method with both RNN-T and LAS modeling frameworks employing neural network and n-gram LMs as ELMs respectively on two domain specific (target) test sets. The proposed method can achieve significantly better performance on the target test sets while it gets minimal performance degradation on the general test set, compared with both shallow and ILME-based LM fusion methods.
The advances in language-based Artificial Intelligence (AI) technologies applied to build educational applications can present AI for social-good opportunities with a broader positive impact. Across many disciplines, enhancing the quality of mathematics education is crucial in building critical thinking and problem-solving skills at younger ages. Conversational AI systems have started maturing to a point where they could play a significant role in helping students learn fundamental math concepts. This work presents a task-oriented Spoken Dialogue System (SDS) built to support play-based learning of basic math concepts for early childhood education. The system has been evaluated via real-world deployments at school while the students are practicing early math concepts with multimodal interactions. We discuss our efforts to improve the SDS pipeline built for math learning, for which we explore utilizing MathBERT representations for potential enhancement to the Natural Language Understanding (NLU) module. We perform an end-to-end evaluation using real-world deployment outputs from the Automatic Speech Recognition (ASR), Intent Recognition, and Dialogue Manager (DM) components to understand how error propagation affects the overall performance in real-world scenarios.
End-to-end intent classification using speech has numerous advantages compared to the conventional pipeline approach using automatic speech recognition (ASR), followed by natural language processing modules. It attempts to predict intent from speech without using an intermediate ASR module. However, such end-to-end framework suffers from the unavailability of large speech resources with higher acoustic variation in spoken language understanding. In this work, we exploit the scope of the transformer distillation method that is specifically designed for knowledge distillation from a transformer based language model to a transformer based speech model. In this regard, we leverage the reliable and widely used bidirectional encoder representations from transformers (BERT) model as a language model and transfer the knowledge to build an acoustic model for intent classification using the speech. In particular, a multilevel transformer based teacher-student model is designed, and knowledge distillation is performed across attention and hidden sub-layers of different transformer layers of the student and teacher models. We achieve an intent classification accuracy of 99.10% and 88.79% for Fluent speech corpus and ATIS database, respectively. Further, the proposed method demonstrates better performance and robustness in acoustically degraded condition compared to the baseline method.
Unsupervised models of representations based on Contrastive Predictive Coding (CPC)[1] are primarily used in spoken language modelling in that they encode phonetic information. In this study, we ask what other types of information are present in CPC speech representations. We focus on three categories: phone class, gender and language, and compare monolingual and bilingual models. Using qualitative and quantitative tools, we find that both gender and phone class information are present in both types of models. Language information, however, is very salient in the bilingual model only, suggesting CPC models learn to discriminate languages when trained on multiple languages. Some language information can also be retrieved from monolingual models, but it is more diffused across all features. These patterns hold when analyses are carried on the discrete units from a downstream clustering model. However, although there is no effect of the number of target clusters on phone class and language information, more gender information is encoded with more clusters. Finally, we find that there is some cost to being exposed to two languages on a downstream phoneme discrimination task.
Audio denoising has been explored for decades using both traditional and deep learning-based methods. However, these methods are still limited to either manually added artificial noise or lower denoised audio quality. To overcome these challenges, we collect a large-scale natural noise bird sound dataset. We are the first to transfer the audio denoising problem into an image segmentation problem and propose a deep visual audio denoising (DVAD) model. With a total of 14,120 audio images, we develop an audio ImageMask tool and propose to use a few-shot generalization strategy to label these images. Extensive experimental results demonstrate that the proposed model achieves state-of-the-art performance. We also show that our method can be easily generalized to speech denoising, audio separation, audio enhancement, and noise estimation.
Automatic Speech Recognition (ASR) systems have become ubiquitous. They can be found in a variety of form factors and are increasingly important in our daily lives. As such, ensuring that these systems are equitable to different subgroups of the population is crucial. In this paper, we introduce, AequeVox, an automated testing framework for evaluating the fairness of ASR systems. AequeVox simulates different environments to assess the effectiveness of ASR systems for different populations. In addition, we investigate whether the chosen simulations are comprehensible to humans. We further propose a fault localization technique capable of identifying words that are not robust to these varying environments. Both components of AequeVox are able to operate in the absence of ground truth data. We evaluated AequeVox on speech from four different datasets using three different commercial ASRs. Our experiments reveal that non-native English, female and Nigerian English speakers generate 109%, 528.5% and 156.9% more errors, on average than native English, male and UK Midlands speakers, respectively. Our user study also reveals that 82.9% of the simulations (employed through speech transformations) had a comprehensibility rating above seven (out of ten), with the lowest rating being 6.78. This further validates the fairness violations discovered by AequeVox. Finally, we show that the non-robust words, as predicted by the fault localization technique embodied in AequeVox, show 223.8% more errors than the predicted robust words across all ASRs.
A vast majority of the world's 7,000 spoken languages are predicted to become extinct within this century, including the endangered language of Ladin from the Italian Alps. Linguists who work to preserve a language's phonetic and phonological structure can spend hours transcribing each minute of speech from native speakers. To address this problem in the context of Ladin, our paper presents the first analysis of speech representations and machine learning models for classifying 32 phonemes of Ladin. We experimented with a novel dataset of the Fascian dialect of Ladin, collected from native speakers in Italy. We created frame-level and segment-level speech feature extraction approaches and conducted extensive experiments with 8 different classifiers trained on 9 different speech representations. Our speech representations ranged from traditional features (MFCC, LPC) to features learned with deep neural network models (autoencoders, LSTM autoencoders, and WaveNet). Our highest-performing classifier, trained on MFCC representations of speech signals, achieved an 86% average accuracy across all Ladin phonemes. We also obtained average accuracies above 77% for all Ladin phoneme subgroups examined. Our findings contribute insights for learning discriminative Ladin phoneme representations and demonstrate the potential for leveraging machine learning and speech signal processing to preserve Ladin and other endangered languages.
Streaming voice conversion (VC) is the task of converting the voice of one person to another in real-time. Previous streaming VC methods use phonetic posteriorgrams (PPGs) extracted from automatic speech recognition (ASR) systems to represent speaker-independent information. However, PPGs lack the prosody and vocalization information of the source speaker, and streaming PPGs contain undesired leaked timbre of the source speaker. In this paper, we propose to use intermediate bottleneck features (IBFs) to replace PPGs. VC systems trained with IBFs retain more prosody and vocalization information of the source speaker. Furthermore, we propose a non-streaming teacher guidance (TG) framework that addresses the timbre leakage problem. Experiments show that our proposed IBFs and the TG framework achieve a state-of-the-art streaming VC naturalness of 3.85, a content consistency of 3.77, and a timbre similarity of 3.77 under a future receptive field of 160 ms which significantly outperform previous streaming VC systems.
Human speech production encompasses physiological processes that naturally react to physic stress. Stress caused by physical activity (PA), e.g., running, may lead to significant changes in a person's speech. The major changes are related to the aspects of pitch level, speaking rate, pause pattern, and breathiness. The extent of change depends presumably on physical fitness and well-being of the person, as well as intensity of PA. The general wellness of a person is further related to his/her physical literacy (PL), which refers to a holistic description of engagement in PA. This paper presents the development of a Cantonese speech database that contains audio recordings of speech before and after physical exercises of different intensity levels. The corpus design and data collection process are described. Preliminary results of acoustical analysis are presented to illustrate the impact of PA on pitch level, pitch range, speaking and articulation rate, and time duration of pauses. It is also noted that the effect of PA is correlated to some of the PA and PL measures.