Showing papers in "Computer Speech & Language in 2019"

TL;DR: An automatic speech recognition based procedure for the extraction of a special set of acoustic features and a linguistic feature set that is extracted from the transcripts of the same speech signals to tell apart Alzheimer’s patients from those with mild cognitive impairment.

TL;DR: The requirements for effective privacy preservation are established, generic cryptography-based solutions are reviewed, followed by specific techniques that are applicable to speaker characterisation and speech characterisation (biometrics and non-biometric applications), and common, empirical evaluation metrics for the assessment of privacy-preserving technologies for speech data are outlined.

TL;DR: An automatic classification system using an intelligent virtual agent (IVA) is presented and it is shown that using acoustic, lexical and CA-inspired features enable ND/FMD classification rates of 90.0% for the neurologist-patient conversations, and 90.9%" for the IVA- patient conversations.

TL;DR: This paper propose an unsupervised method that models a sentence as a weighted series of word embeddings, which are fitted by using Shannon's Mutual Information (MI) among words, sentences and the corpus.

TL;DR: An algorithm is proposed that estimates the articulatory coordination of speech from audio and video signals, and uses these coordination features to learn a prediction model to track depression severity with treatment, allowing rapid assessment of treatment efficacy as well as improved long term care of individuals at high risk for depression.

TL;DR: This paper proposes to augment the existing LSTM neural tagging model for Arabic NER with a Convolutional Neural Network for the extraction of relevant character- level features and shows that character CNN is able to outperform the previously used character-level Bi-directional Long Short-Term Memory Networks (BiLSTM) in many settings.

TL;DR: This article presents the four Sub-Challenges of ComParE 2013 and provides details of the Challenge databases and a meta-analysis by conducting experiments of logistic regression on single features and evaluating the performances achieved by the participants.

TL;DR: The hypothesis that subtle differences in language can be detected in narrative speech, even at the very early stages of cognitive decline, is supported when scores on screening tools such as the Mini-Mental State Exam are still in the “normal” range.

TL;DR: In this article, the authors explore the incremental update of neural machine translation systems during the post-editing or interactive translation processes and show that online learning effectively achieves the objective of reducing the human effort required for obtaining high-quality translations.

TL;DR: The blending end-to-end trainable models associated to meaningful prior knowledge performs the best for the restaurant retrieval for Track 1 and Hybrid Code Network and Memory Network have been the best models for this task.

TL;DR: This paper presents an automated assessment framework in quantifying atypical prosody and stereotyped idiosyncratic phrases related to ASD and proposes both the hand-crafted feature based method as well as the end-to-end deep learning framework for detecting atypicals prosody from speech.

TL;DR: This work proposes a framework that learns to embed semantic correspondence between text and its extracted semantic knowledge, called semantic frame, and demonstrates three key areas where the embedding model can be effective: visualization, distance based semantic search, similarity-based intent classification and re-ranking.

TL;DR: In this article, a DNN-based autoencoder for speech enhancement, deverberation and denoising is presented, which can be used to build a robust speaker verification system for various target domains.

TL;DR: Experiments on two real-world microblog data sets demonstrate that the proposed novel sentiment unit context propagation framework can generate microblog-specific sentiment lexicons effectively and significantly outperform state-of-the-art baselines.

TL;DR: By examining vowel articulatory parameters, statistically significant differences in articulatory characteristics are found at a paraphonetic level and linguistic stress feature results indicate that specific vowel set analysis provides better discrimination of clinically depressed and non-depressed speakers.

TL;DR: Two compensation methods are proposed to tackle the mismatch in a shouted versus normal speaker recognition task by modifying the spectral envelopes of shouts to be closer to those in normal speech.

TL;DR: This paper addresses Korean NER tasks and proposes an extension of a bidirectional LSTM CRF by investigating character-based representation and deploys a hybrid representation using ConvNet and L STM for the sequential modeling of characters, namely a character- based LSTm-ConvNet hybrid representation.

TL;DR: This work has proposed a new graph-based measure for keyword extraction, by leveraging higher-order structural features of word co-occurrence graph, and shows superior performance of the proposed method, compared to TF-IDF and PageRank based methods.

TL;DR: Residual convolutional neural networks are employed to this end, aiming at exploiting the ability of such architectures to take into account large contextual segments of input data, and learnable attention mechanisms are introduced on top of the convolutionAL stack for data-driven feature pooling across time.

TL;DR: A model to satisfy the requirements of Dialog System Technology Challenge 6 (DSTC6) Track 1: building an end-to-end dialog systems for goal-oriented applications is developed and achieves state-of-the-art performance among the memory networks, and is comparable to hybrid code networks and hierarchical LSTM model.

TL;DR: A two-dimensional evaluation metric that is designed to operate at sentence level, which considers the syntactic and semantic information carried along the answers generated by an end-to-end dialog system with respect to a set of references is evaluated.

TL;DR: In this article, the authors focus on two laws that have been studied less intensively: the meaning-frequency law, i.e. the tendency of more frequent words to be more polysemous, and the law of abbreviation, which refers to the fact that more frequently words tend to be shorter.

TL;DR: This work compares the ASR performance of speaker-independent bottleneck and articulatory features on dysarthric speech used in conjunction with dedicated neural network-based acoustic models that have been shown to be robust against spectrotemporal deviations.

TL;DR: This work designs complementary models for two different tasks such as sentence clustering and neural sentence fusion and applies them to implement a full abstractive multi-document summarization system which simultaneously considers importance, coverage, and diversity under a desired length limit.

TL;DR: This paper describes the attempt at generating natural and informative responses for customer service oriented dialog systems, by incorporating dialog history related information and external knowledge in two improved sequence-to-sequence frameworks.

TL;DR: An effective combination method of a statistical model, C-value method, and a graph-based model to overcome the drawbacks of each model is proposed and its results outperformed the state-of-the-art model among unsupervised models and the existing graph- based ranking models.

TL;DR: This work proposes a method to predict the emotional reactions that Twitter users would have after reading a news article by using a multi-target classification strategy and obtains an emotional reactions similarity of 89%.

TL;DR: Under this unified framework, several integrated models are proposed to incorporate different types of information extracted from TM to guide the SMT decoding and let SMT implicitly and indirectly utilize global context with a local dependency model.

TL;DR: This work proposes to compute signal-derived multimodal behavior descriptors of ASD subjects during dyadic interactions of Autism Diagnostic Observation Schedule (ADOS), and further examines these behavior features’ discriminatory power in differentiating between the three groups in ASD.

TL;DR: A comparative study shows that the best computational architecture is an FFNN along with a combination of word embeddings and acoustic features, suggesting that lexical information can be utilized as a powerful cue for overlap classification.