Visar Berisha

Sean Kinahan, Julie Liss, Visar Berisha

The DIVA model is a computational model of speech motor control that combines a simulation of the brain regions responsible for speech production with a model of the human vocal tract. The model is currently implemented in Matlab Simulink; however, this is less than ideal as most of the development in speech technology research is done in Python. This means there is a wealth of machine learning tools which are freely available in the Python ecosystem that cannot be easily integrated with DIVA. We present TorchDIVA, a full rebuild of DIVA in Python using PyTorch tensors. DIVA source code was directly translated from Matlab to Python, and built-in Simulink signal blocks were implemented from scratch. After implementation, the accuracy of each module was evaluated via systematic block-by-block validation. The TorchDIVA model is shown to produce outputs that closely match those of the original DIVA model, with a negligible difference between the two. We additionally present an example of the extensibility of TorchDIVA as a research platform. Speech quality enhancement in TorchDIVA is achieved through an integration with an existing PyTorch generative vocoder called DiffWave. A modified DiffWave mel-spectrum upsampler was trained on human speech waveforms and conditioned on the TorchDIVA speech production. The results indicate improved speech quality metrics in the DiffWave-enhanced output as compared to the baseline. This enhancement would have been difficult or impossible to accomplish in the original Matlab implementation. This proof-of-concept demonstrates the value TorchDIVA will bring to the research community. Researchers can download the new implementation at: https://github.com/skinahan/DIVA_PyTorch

Eunjung Yeo, Julie M. Liss, Visar Berisha et al. · cmu

The growing prevalence of neurological disorders associated with dysarthria motivates the need for automated intelligibility assessment methods that are applicalbe across languages. However, most existing approaches are either limited to a single language or fail to capture language-specific factors shaping intelligibility. We present a multilingual phoneme-production assessment framework that integrates universal phone recognition with language-specific phoneme interpretation using contrastive phonological feature distances for phone-to-phoneme mapping and sequence alignment. The framework yields three metrics: phoneme error rate (PER), phonological feature error rate (PFER), and a newly proposed alignment-free measure, phoneme coverage (PhonCov). Analysis on English, Spanish, Italian, and Tamil show that PER benefits from the combination of mapping and alignment, PFER from alignment alone, and PhonCov from mapping. Further analyses demonstrate that the proposed framework captures clinically meaningful patterns of intelligibility degradation consistent with established observations of dysarthric speech.

Jianwei Zhang, Julie Liss, Suren Jayasuriya et al.

Approximately 1.2% of the world's population has impaired voice production. As a result, automatic dysphonic voice detection has attracted considerable academic and clinical interest. However, existing methods for automated voice assessment often fail to generalize outside the training conditions or to other related applications. In this paper, we propose a deep learning framework for generating acoustic feature embeddings sensitive to vocal quality and robust across different corpora. A contrastive loss is combined with a classification loss to train our deep learning model jointly. Data warping methods are used on input voice samples to improve the robustness of our method. Empirical results demonstrate that our method not only achieves high in-corpus and cross-corpus classification accuracy but also generates good embeddings sensitive to voice quality and robust across different corpora. We also compare our results against three baseline methods on clean and three variations of deteriorated in-corpus and cross-corpus datasets and demonstrate that the proposed model consistently outperforms the baseline methods.

Weizhi Li, Prad Kadambi, Pouria Saidi et al.

A two-sample hypothesis test is a statistical procedure used to determine whether the distributions generating two samples are identical. We consider the two-sample testing problem in a new scenario where the sample measurements (or sample features) are inexpensive to access, but their group memberships (or labels) are costly. To address the problem, we devise the first \emph{active sequential two-sample testing framework} that not only sequentially but also \emph{actively queries}. Our test statistic is a likelihood ratio where one likelihood is found by maximization over all class priors, and the other is provided by a probabilistic classification model. The classification model is adaptively updated and used to predict where the (unlabelled) features have a high dependency on labels; labeling the ``high-dependency'' features leads to the increased power of the proposed testing framework. In theory, we provide the proof that our framework produces an \emph{anytime-valid} $p$-value. In addition, we characterize the proposed framework's gain in testing power by analyzing the mutual information between the feature and label variables in asymptotic and finite-sample scenarios. In practice, we introduce an instantiation of our framework and evaluate it using several experiments; the experiments on the synthetic, MNIST, and application-specific datasets demonstrate that the testing power of the instantiated active sequential test significantly increases while the Type I error is under control.

Jianwei Zhang, Suren Jayasuriya, Visar Berisha

A good supervised embedding for a specific machine learning task is only sensitive to changes in the label of interest and is invariant to other confounding factors. We leverage the concept of repeatability from measurement theory to describe this property and propose to use the intra-class correlation coefficient (ICC) to evaluate the repeatability of embeddings. We then propose a novel regularizer, the ICC regularizer, as a complementary component for contrastive losses to guide deep neural networks to produce embeddings with higher repeatability. We use simulated data to explain why the ICC regularizer works better on minimizing the intra-class variance than the contrastive loss alone. We implement the ICC regularizer and apply it to three speech tasks: speaker verification, voice style conversion, and a clinical application for detecting dysphonic voice. The experimental results demonstrate that adding an ICC regularizer can improve the repeatability of learned embeddings compared to only using the contrastive loss; further, these embeddings lead to improved performance in these downstream tasks.

Leo Hsu, Visar Berisha

Researchers have observed that the frequencies of leading digits in many man-made and naturally occurring datasets follow a logarithmic curve, with digits that start with the number 1 accounting for $\sim 30\%$ of all numbers in the dataset and digits that start with the number 9 accounting for $\sim 5\%$ of all numbers in the dataset. This phenomenon, known as Benford's Law, is highly repeatable and appears in lists of numbers from electricity bills, stock prices, tax returns, house prices, death rates, lengths of rivers, and naturally occurring images. In this paper we demonstrate that human speech spectra also follow Benford's Law on average. That is, when averaged over many speakers, the frequencies of leading digits in speech magnitude spectra follow this distribution, although with some variability at the individual sample level. We use this observation to motivate a new set of features that can be efficiently extracted from speech and demonstrate that these features can be used to classify between human speech and synthetic speech.

Tyler Sypherd, Nathan Stromberg, Richard Nock et al.

There is a growing need for models that are interpretable and have reduced energy and computational cost (e.g., in health care analytics and federated learning). Examples of algorithms to train such models include logistic regression and boosting. However, one challenge facing these algorithms is that they provably suffer from label noise; this has been attributed to the joint interaction between oft-used convex loss functions and simpler hypothesis classes, resulting in too much emphasis being placed on outliers. In this work, we use the margin-based $α$-loss, which continuously tunes between canonical convex and quasi-convex losses, to robustly train simple models. We show that the $α$ hyperparameter smoothly introduces non-convexity and offers the benefit of "giving up" on noisy training examples. We also provide results on the Long-Servedio dataset for boosting and a COVID-19 survey dataset for logistic regression, highlighting the efficacy of our approach across multiple relevant domains.

Tanusree Sharma, Anish Krishnagiri, Lili Dudas et al.

As generative voice models are rapidly advancing in both capabilities and public utilization, the unconsented collection, reuse, and synthesis of voice data are introducing new classes of privacy, security and governance risk that are poorly captured by existing, largely uniform threat models. To fill the gap, we present V.O.I.C.E, a taxonomy of voice generation risk grounded in a multi-source threat modeling effort with 569 incidents from major AI incident database, FTC and Internet Crime Complaint Center (IC3); 1067 direct incident reports from U.S. based participants across diverse groups (including voice actors, internet personalities, political personnel, and general public); and 2,221 Reddit discussions. Grounded in real-world data, our taxonomy explicitly models how risk emerges, interact with contextual factors such as degree of exposure, social visibility, and the availability of legal protections for various affected groups.

Si-Ioi Ng, Pranav S. Ambadi, Kimberly D. Mueller et al.

Current methods for automated assessment of cognitive-linguistic impairment via picture description often neglect the visual narrative path - the sequence and locations of elements a speaker described in the picture. Analyses of spatio-semantic features capture this path using content information units (CIUs), but manual tagging or dictionary-based mapping is labor-intensive. This study proposes a BERT-based pipeline, fine tuned with binary cross-entropy and pairwise ranking loss, for automated CIU extraction and ordering from the Cookie Theft picture description. Evaluated by 5-fold cross-validation, it achieves 93% median precision, 96% median recall in CIU detection, and 24% sequence error rates. The proposed method extracts features that exhibit strong Pearson correlations with ground truth, surpassing the dictionary-based baseline in external validation. These features also perform comparably to those derived from manual annotations in evaluating group differences via ANCOVA. The pipeline is shown to effectively characterize visual narrative paths for cognitive impairment assessment, with the implementation and models open-sourced to public.

Weizhi Li, Gautam Dasarathy, Karthikeyan Natesan Ramamurthy et al.

Two-sample tests evaluate whether two samples are realizations of the same distribution (the null hypothesis) or two different distributions (the alternative hypothesis). We consider a new setting for this problem where sample features are easily measured whereas sample labels are unknown and costly to obtain. Accordingly, we devise a three-stage framework in service of performing an effective two-sample test with only a small number of sample label queries: first, a classifier is trained with samples uniformly labeled to model the posterior probabilities of the labels; second, a novel query scheme dubbed \emph{bimodal query} is used to query labels of samples from both classes, and last, the classical Friedman-Rafsky (FR) two-sample test is performed on the queried samples. Theoretical analysis and extensive experiments performed on several datasets demonstrate that the proposed test controls the Type I error and has decreased Type II error relative to uniform querying and certainty-based querying. Source code for our algorithms and experimental results is available at \url{https://github.com/wayne0908/Label-Efficient-Two-Sample}.

Weizhi Li, Gautam Dasarathy, Karthikeyan Natesan Ramamurthy et al.

Accurately and efficiently characterizing the decision boundary of classifiers is important for problems related to model selection and meta-learning. Inspired by topological data analysis, the characterization of decision boundaries using their homology has recently emerged as a general and powerful tool. In this paper, we propose an active learning algorithm to recover the homology of decision boundaries. Our algorithm sequentially and adaptively selects which samples it requires the labels of. We theoretically analyze the proposed framework and show that the query complexity of our active learning algorithm depends naturally on the intrinsic complexity of the underlying manifold. We demonstrate the effectiveness of our framework in selecting best-performing machine learning models for datasets just using their respective homological summaries. Experiments on several standard datasets show the sample complexity improvement in recovering the homology and demonstrate the practical utility of the framework for model selection. Source code for our algorithms and experimental results is available at https://github.com/wayne0908/Active-Learning-Homology.

Pouria Saidi, Gautam Dasarathy, Visar Berisha

Machine Learning (ML) is increasingly used across many disciplines with impressive reported results. However, recent studies suggest published performance of ML models are often overoptimistic. Validity concerns are underscored by findings of an inverse relationship between sample size and reported accuracy in published ML models, contrasting with the theory of learning curves where accuracy should improve or remain stable with increasing sample size. This paper investigates factors contributing to overoptimism in ML-driven science, focusing on overfitting and publication bias. We introduce a novel stochastic model for observed accuracy, integrating parametric learning curves and the aforementioned biases. We construct an estimator that corrects for these biases in observed data. Theoretical and empirical results show that our framework can estimate the underlying learning curve, providing realistic performance assessments from published results. Applying the model to meta-analyses of classifications of neurological conditions, we estimate the inherent limits of ML-based prediction in each domain.

Si-Ioi Ng, Pranav S. Ambadi, Kimberly D. Mueller et al.

Existing methods for analyzing linguistic content from picture descriptions for assessment of cognitive-linguistic impairment often overlook the participant's visual narrative path, which typically requires eye tracking to assess. Spatio-semantic graphs are a useful tool for analyzing this narrative path from transcripts alone, however they are limited by the need for manual tagging of content information units (CIUs). In this paper, we propose an automated approach for estimation of spatio-semantic graphs (via automated extraction of CIUs) from the Cookie Theft picture commonly used in cognitive-linguistic analyses. The method enables the automatic characterization of the visual semantic path during picture description. Experiments demonstrate that the automatic spatio-semantic graphs effectively differentiate between cognitively impaired and unimpaired speakers. Statistical analyses reveal that the features derived by the automated method produce comparable results to the manual method, with even greater group differences between clinical groups of interest. These results highlight the potential of the automated approach for extracting spatio-semantic features in developing clinical speech models for cognitive impairment assessment.

Weizhi Li, Visar Berisha, Gautam Dasarathy

Hypothesis testing is a statistical inference approach used to determine whether data supports a specific hypothesis. An important type is the two-sample test, which evaluates whether two sets of data points are from identical distributions. This test is widely used, such as by clinical researchers comparing treatment effectiveness. This tutorial explores two-sample testing in a context where an analyst has many features from two samples, but determining the sample membership (or labels) of these features is costly. In machine learning, a similar scenario is studied in active learning. This tutorial extends active learning concepts to two-sample testing within this \textit{label-costly} setting while maintaining statistical validity and high testing power. Additionally, the tutorial discusses practical applications of these label-efficient two-sample tests.

Dohyun Ku, Catherine D. Chong, Visar Berisha et al.

Time series analysis has emerged as an important tool for improving patient diagnosis and management in healthcare applications. However, these applications commonly face two critical challenges: time misalignment and data sparsity. Traditional approaches address these issues through a two-step process of imputation followed by prediction. We propose MAGIC (Multi-tAsk Gaussian Process for Imputation and Classification), a novel unified framework that simultaneously performs class-informed missing value imputation and label prediction within a hierarchical multi-task Gaussian process coupled with functional logistic regression. To handle intractable likelihood components, MAGIC employs Taylor expansion approximations with bounded error analysis, and parameter estimation is performed using EM algorithm with block coordinate optimization supported by convergence analysis. We validate MAGIC through two healthcare applications: prediction of post-traumatic headache improvement following mild traumatic brain injury and prediction of in-hospital mortality within 48 hours after ICU admission. In both applications, MAGIC achieves superior predictive accuracy compared to existing methods. The ability to generate real-time and accurate predictions with limited samples facilitates early clinical assessment and treatment planning, enabling healthcare providers to make more informed treatment decisions.

Weizhi Li, Gautam Dasarathy, Visar Berisha

Matched-pair experimental designs aim to detect treatment effects by pairing participants and comparing within-pair outcome differences. In many situations, the overall effect size across the entire population is small. Then, the focus naturally shifts to identifying and targeting high treatment-effect regions where the intervention is most effective. This paper proposes a matched-pair experimental design that sequentially and actively enrolls patients in high treatment-effect regions. Importantly, we frame the identification of the target region as a classification problem and propose an active learning framework tailored to matched-pair designs. Our design not only reduces the experimental cost of detecting treatment efficacy, but also ensures that the identified regions enclose the entire high-treatment-effect regions. Our theoretical analysis of the framework's label complexity and experiments in practical scenarios demonstrate the efficiency and advantages of the approach.

Tanusree Sharma, Yihao Zhou, Visar Berisha

Early large-scale audio datasets, such as LibriSpeech, were built with hundreds of individual contributors whose voices were instrumental in the development of speech technologies, including audiobooks and voice assistants. Yet, a decade later, these same contributions have exposed voice actors to a range of risks. While existing ethical frameworks emphasize Consent, Credit, and Compensation (C3), they do not adequately address the emergent risks involving vocal identities that are increasingly decoupled from context, authorship, and control. Drawing on qualitative interviews with 20 professional voice actors, this paper reveals how the synthetic replication of voice without enforceable constraints exposes individuals to a range of threats. Beyond reputational harm, such as re-purposing voice data in erotic content, offensive political messaging, and meme culture, we document concerns about accountability breakdowns when their voice is leveraged to clone voices that are deployed in high-stakes scenarios such as financial fraud, misinformation campaigns, or impersonation scams. In such cases, actors face social and legal fallout without recourse, while very few of them have a legal representative or union protection. To make sense of these shifting dynamics, we introduce the PRAC3 framework, an expansion of C3 that foregrounds Privacy, Reputation, Accountability, Consent, Credit, and Compensation as interdependent pillars of data used in the synthetic voice economy. This framework captures how privacy risks are amplified through non-consensual training, how reputational harm arises from decontextualized deployment, and how accountability can be reimagined AI Data ecosystems. We argue that voice, as both a biometric identifier and creative labor, demands governance models that restore creator agency, ensure traceability, and establish enforceable boundaries for ethical reuse.

Eunjung Yeo, Julie Liss, Visar Berisha et al.

Purpose: Speech intelligibility is a critical outcome in the assessment and management of dysarthria, yet most research and clinical practices have focused on English, limiting their applicability across languages. This commentary introduces a conceptual framework--and a demonstration of how it can be implemented--leveraging artificial intelligence (AI) to advance cross-language intelligibility assessment of dysarthric speech. Method: We propose a two-tiered conceptual framework consisting of a universal speech model that encodes dysarthric speech into acoustic-phonetic representations, followed by a language-specific intelligibility assessment model that interprets these representations within the phonological or prosodic structures of the target language. We further identify barriers to cross-language intelligibility assessment of dysarthric speech, including data scarcity, annotation complexity, and limited linguistic insights into dysarthric speech, and outline potential AI-driven solutions to overcome these challenges. Conclusion: Advancing cross-language intelligibility assessment of dysarthric speech necessitates models that are both efficient and scalable, yet constrained by linguistic rules to ensure accurate and language-sensitive assessment. Recent advances in AI provide the foundational tools to support this integration, shaping future directions toward generalizable and linguistically informed assessment frameworks.

Si-Ioi Ng, Lingfeng Xu, Ingo Siegert et al.

There has been a surge of interest in leveraging speech as a marker of health for a wide spectrum of conditions. The underlying premise is that any neurological, mental, or physical deficits that impact speech production can be objectively assessed via automated analysis of speech. Recent advances in speech-based Artificial Intelligence (AI) models for diagnosing and tracking mental health, cognitive, and motor disorders often use supervised learning, similar to mainstream speech technologies like recognition and verification. However, clinical speech AI has distinct challenges, including the need for specific elicitation tasks, small available datasets, diverse speech representations, and uncertain diagnostic labels. As a result, application of the standard supervised learning paradigm may lead to models that perform well in controlled settings but fail to generalize in real-world clinical deployments. With translation into real-world clinical scenarios in mind, this tutorial paper provides an overview of the key components required for robust development of clinical speech AI. Specifically, this paper will cover the design of speech elicitation tasks and protocols most appropriate for different clinical conditions, collection of data and verification of hardware, development and validation of speech representations designed to measure clinical constructs of interest, development of reliable and robust clinical prediction models, and ethical and participant considerations for clinical speech AI. The goal is to provide comprehensive guidance on building models whose inputs and outputs link to the more interpretable and clinically meaningful aspects of speech, that can be interrogated and clinically validated on clinical datasets, and that adhere to ethical, privacy, and security considerations by design.

Jianwei Zhang, Suren Jayasuriya, Visar Berisha

There are many deterministic mathematical operations (e.g. compression, clipping, downsampling) that degrade speech quality considerably. In this paper we introduce a neural network architecture, based on a modification of the DiffWave model, that aims to restore the original speech signal. DiffWave, a recently published diffusion-based vocoder, has shown state-of-the-art synthesized speech quality and relatively shorter waveform generation times, with only a small set of parameters. We replace the mel-spectrum upsampler in DiffWave with a deep CNN upsampler, which is trained to alter the degraded speech mel-spectrum to match that of the original speech. The model is trained using the original speech waveform, but conditioned on the degraded speech mel-spectrum. Post-training, only the degraded mel-spectrum is used as input and the model generates an estimate of the original speech. Our model results in improved speech quality (original DiffWave model as baseline) on several different experiments. These include improving the quality of speech degraded by LPC-10 compression, AMR-NB compression, and signal clipping. Compared to the original DiffWave architecture, our scheme achieves better performance on several objective perceptual metrics and in subjective comparisons. Improvements over baseline are further amplified in a out-of-corpus evaluation setting.

Weizhi Li, Gautam Dasarathy, Visar Berisha

Regularization is an effective way to promote the generalization performance of machine learning models. In this paper, we focus on label smoothing, a form of output distribution regularization that prevents overfitting of a neural network by softening the ground-truth labels in the training data in an attempt to penalize overconfident outputs. Existing approaches typically use cross-validation to impose this smoothing, which is uniform across all training data. In this paper, we show that such label smoothing imposes a quantifiable bias in the Bayes error rate of the training data, with regions of the feature space with high overlap and low marginal likelihood having a lower bias and regions of low overlap and high marginal likelihood having a higher bias. These theoretical results motivate a simple objective function for data-dependent smoothing to mitigate the potential negative consequences of the operation while maintaining its desirable properties as a regularizer. We call this approach Structural Label Smoothing (SLS). We implement SLS and empirically validate on synthetic, Higgs, SVHN, CIFAR-10, and CIFAR-100 datasets. The results confirm our theoretical insights and demonstrate the effectiveness of the proposed method in comparison to traditional label smoothing.

Michael Saxon, Ayush Tripathi, Yishan Jiao et al.

Hypernasality is a common characteristic symptom across many motor-speech disorders. For voiced sounds, hypernasality introduces an additional resonance in the lower frequencies and, for unvoiced sounds, there is reduced articulatory precision due to air escaping through the nasal cavity. However, the acoustic manifestation of these symptoms is highly variable, making hypernasality estimation very challenging, both for human specialists and automated systems. Previous work in this area relies on either engineered features based on statistical signal processing or machine learning models trained on clinical ratings. Engineered features often fail to capture the complex acoustic patterns associated with hypernasality, whereas metrics based on machine learning are prone to overfitting to the small disease-specific speech datasets on which they are trained. Here we propose a new set of acoustic features that capture these complementary dimensions. The features are based on two acoustic models trained on a large corpus of healthy speech. The first acoustic model aims to measure nasal resonance from voiced sounds, whereas the second acoustic model aims to measure articulatory imprecision from unvoiced sounds. To demonstrate that the features derived from these acoustic models are specific to hypernasal speech, we evaluate them across different dysarthria corpora. Our results show that the features generalize even when training on hypernasal speech from one disease and evaluating on hypernasal speech from another disease (e.g. training on Parkinson's disease, evaluation on Huntington's disease), and when training on neurologically disordered speech but evaluating on cleft palate speech.

Rohit Voleti, Julie M. Liss, Visar Berisha

It is widely accepted that information derived from analyzing speech (the acoustic signal) and language production (words and sentences) serves as a useful window into the health of an individual's cognitive ability. In fact, most neuropsychological testing batteries have a component related to speech and language where clinicians elicit speech from patients for subjective evaluation across a broad set of dimensions. With advances in speech signal processing and natural language processing, there has been recent interest in developing tools to detect more subtle changes in cognitive-linguistic function. This work relies on extracting a set of features from recorded and transcribed speech for objective assessments of speech and language, early diagnosis of neurological disease, and tracking of disease after diagnosis. With an emphasis on cognitive and thought disorders, in this paper we provide a review of existing speech and language features used in this domain, discuss their clinical application, and highlight their advantages and disadvantages. Broadly speaking, the review is split into two categories: language features based on natural language processing and speech features based on speech signal processing. Within each category, we consider features that aim to measure complementary dimensions of cognitive-linguistics, including language diversity, syntactic complexity, semantic coherence, and timing. We conclude the review with a proposal of new research directions to further advance the field.

Rohit Voleti, Stephanie Woolridge, Julie M. Liss et al.

Several studies have shown that speech and language features, automatically extracted from clinical interviews or spontaneous discourse, have diagnostic value for mental disorders such as schizophrenia and bipolar disorder. They typically make use of a large feature set to train a classifier for distinguishing between two groups of interest, i.e. a clinical and control group. However, a purely data-driven approach runs the risk of overfitting to a particular data set, especially when sample sizes are limited. Here, we first down-select the set of language features to a small subset that is related to a well-validated test of functional ability, the Social Skills Performance Assessment (SSPA). This helps establish the concurrent validity of the selected features. We use only these features to train a simple classifier to distinguish between groups of interest. Linear regression reveals that a subset of language features can effectively model the SSPA, with a correlation coefficient of 0.75. Furthermore, the same feature set can be used to build a strong binary classifier to distinguish between healthy controls and a clinical group (AUC = 0.96) and also between patients within the clinical group with schizophrenia and bipolar I disorder (AUC = 0.83).

Rohit Voleti, Julie M. Liss, Visar Berisha

A key initial step in several natural language processing (NLP) tasks involves embedding phrases of text to vectors of real numbers that preserve semantic meaning. To that end, several methods have been recently proposed with impressive results on semantic similarity tasks. However, all of these approaches assume that perfect transcripts are available when generating the embeddings. While this is a reasonable assumption for analysis of written text, it is limiting for analysis of transcribed text. In this paper we investigate the effects of word substitution errors, such as those coming from automatic speech recognition errors (ASR), on several state-of-the-art sentence embedding methods. To do this, we propose a new simulator that allows the experimenter to induce ASR-plausible word substitution errors in a corpus at a desired word error rate. We use this simulator to evaluate the robustness of several sentence embedding methods. Our results show that pre-trained neural sentence encoders are both robust to ASR errors and perform well on textual similarity tasks after errors are introduced. Meanwhile, unweighted averages of word vectors perform well with perfect transcriptions, but their performance degrades rapidly on textual similarity tasks for text with word substitution errors.

Huan Song, Megan Willi, Jayaraman J. Thiagarajan et al.

In automatic speech processing systems, speaker diarization is a crucial front-end component to separate segments from different speakers. Inspired by the recent success of deep neural networks (DNNs) in semantic inferencing, triplet loss-based architectures have been successfully used for this problem. However, existing work utilizes conventional i-vectors as the input representation and builds simple fully connected networks for metric learning, thus not fully leveraging the modeling power of DNN architectures. This paper investigates the importance of learning effective representations from the sequences directly in metric learning pipelines for speaker diarization. More specifically, we propose to employ attention models to learn embeddings and the metric jointly in an end-to-end fashion. Experiments are conducted on the CALLHOME conversational speech corpus. The diarization results demonstrate that, besides providing a unified model, the proposed approach achieves improved performance when compared against existing approaches.

Ming Tu, Anna Grabek, Julie Liss et al.

Automatic pronunciation evaluation plays an important role in pronunciation training and second language education. This field draws heavily on concepts from automatic speech recognition (ASR) to quantify how close the pronunciation of non-native speech is to native-like pronunciation. However, it is known that the formation of accent is related to pronunciation patterns of both the target language (L2) and the speaker's first language (L1). In this paper, we propose to use two native speech acoustic models, one trained on L2 speech and the other trained on L1 speech. We develop two sets of measurements that can be extracted from two acoustic models given accented speech. A new utterance-level feature extraction scheme is used to convert these measurements into a fixed-dimension vector which is used as an input to a statistical model to predict the accentedness of a speaker. On a data set consisting of speakers from 4 different L1 backgrounds, we show that the proposed system yields improved correlation with human evaluators compared to systems only using the L2 acoustic model.

Megan M. Willi, Stephanie A. Borrie, Tyson S. Barrett et al.

Acoustic-prosodic entrainment describes the tendency of humans to align or adapt their speech acoustics to each other in conversation. This alignment of spoken behavior has important implications for conversational success. However, modeling the subtle nature of entrainment in spoken dialogue continues to pose a challenge. In this paper, we propose a straightforward definition for local entrainment in the speech domain and operationalize an algorithm based on this: acoustic-prosodic features that capture entrainment should be maximally different between real conversations involving two partners and sham conversations generated by randomly mixing the speaking turns from the original two conversational partners. We propose an approach for measuring local entrainment that quantifies alignment of behavior on a turn-by-turn basis, projecting the differences between interlocutors' acoustic-prosodic features for a given turn onto a discriminative feature subspace that maximizes the difference between real and sham conversations. We evaluate the method using the derived features to drive a classifier aiming to predict an objective measure of conversational success (i.e., low versus high), on a corpus of task-oriented conversations. The proposed entrainment approach achieves 72% classification accuracy using a Naive Bayes classifier, outperforming three previously established approaches evaluated on the same conversational corpus.

Shihui Yin, Gaurav Srivastava, Shreyas K. Venkataramanaiah et al.

Deep learning algorithms have shown tremendous success in many recognition tasks; however, these algorithms typically include a deep neural network (DNN) structure and a large number of parameters, which makes it challenging to implement them on power/area-constrained embedded platforms. To reduce the network size, several studies investigated compression by introducing element-wise or row-/column-/block-wise sparsity via pruning and regularization. In addition, many recent works have focused on reducing precision of activations and weights with some reducing down to a single bit. However, combining various sparsity structures with binarized or very-low-precision (2-3 bit) neural networks have not been comprehensively explored. In this work, we present design techniques for minimum-area/-energy DNN hardware with minimal degradation in accuracy. During training, both binarization/low-precision and structured sparsity are applied as constraints to find the smallest memory footprint for a given deep learning algorithm. The DNN model for CIFAR-10 dataset with weight memory reduction of 50X exhibits accuracy comparable to that of the floating-point counterpart. Area, performance and energy results of DNN hardware in 40nm CMOS are reported for the MNIST dataset. The optimized DNN that combines 8X structured compression and 3-bit weight precision showed 98.4% accuracy at 20nJ per classification.

Alan Wisler, Visar Berisha, Andreas Spanias et al.

A number of fundamental quantities in statistical signal processing and information theory can be expressed as integral functions of two probability density functions. Such quantities are called density functionals as they map density functions onto the real line. For example, information divergence functions measure the dissimilarity between two probability density functions and are useful in a number of applications. Typically, estimating these quantities requires complete knowledge of the underlying distribution followed by multi-dimensional integration. Existing methods make parametric assumptions about the data distribution or use non-parametric density estimation followed by high-dimensional integration. In this paper, we propose a new alternative. We introduce the concept of "data-driven basis functions" - functions of distributions whose value we can estimate given only samples from the underlying distributions without requiring distribution fitting or direct integration. We derive a new data-driven complete basis that is similar to the deterministic Bernstein polynomial basis and develop two methods for performing basis expansions of functionals of two distributions. We also show that the new basis set allows us to approximate functions of distributions as closely as desired. Finally, we evaluate the methodology by developing data driven estimators for the Kullback-Leibler divergences and the Hellinger distance and by constructing empirical estimates of tight bounds on the Bayes error rate.

Ming Tu, Visar Berisha, Yu Cao et al.

Deep neural networks are typically represented by a much larger number of parameters than shallow models, making them prohibitive for small footprint devices. Recent research shows that there is considerable redundancy in the parameter space of deep neural networks. In this paper, we propose a method to compress deep neural networks by using the Fisher Information metric, which we estimate through a stochastic optimization method that keeps track of second-order information in the network. We first remove unimportant parameters and then use non-uniform fixed point quantization to assign more bits to parameters with higher Fisher Information estimates. We evaluate our method on a classification task with a convolutional neural network trained on the MNIST data set. Experimental results show that our method outperforms existing methods for both network pruning and quantization.

Visar Berisha, Alan Wisler, Alfred O. Hero et al.

Information divergence functions play a critical role in statistics and information theory. In this paper we show that a non-parametric f-divergence measure can be used to provide improved bounds on the minimum binary classification probability of error for the case when the training and test data are drawn from the same distribution and for the case where there exists some mismatch between training and test distributions. We confirm the theoretical results by designing feature selection algorithms using the criteria from these bounds and by evaluating the algorithms on a series of pathological speech classification tasks.

Visar Berisha, Alfred O. Hero

The Fisher information matrix (FIM) is a foundational concept in statistical signal processing. The FIM depends on the probability distribution, assumed to belong to a smooth parametric family. Traditional approaches to estimating the FIM require estimating the probability distribution function (PDF), or its parameters, along with its gradient or Hessian. However, in many practical situations the PDF of the data is not known but the statistician has access to an observation sample for any parameter value. Here we propose a method of estimating the FIM directly from sampled data that does not require knowledge of the underlying PDF. The method is based on non-parametric estimation of an $f$-divergence over a local neighborhood of the parameter space and a relation between curvature of the $f$-divergence and the FIM. Thus we obtain an empirical estimator of the FIM that does not require density estimation and is asymptotically consistent. We empirically evaluate the validity of our approach using two experiments.