False triggers in voice assistants are unintended invocations of the assistant, which not only degrade the user experience but may also compromise privacy. False trigger mitigation (FTM) is a process to detect the false trigger events and respond appropriately to the user. In this paper, we propose a novel solution to the FTM problem by introducing a parallel ASR decoding process with a special language model trained from "out-of-domain" data sources. Such language model is complementary to the existing language model optimized for the assistant task. A bidirectional lattice RNN (Bi-LRNN) classifier trained from the lattices generated by the complementary language model shows a $38.34\%$ relative reduction of the false trigger (FT) rate at the fixed rate of $0.4\%$ false suppression (FS) of correct invocations, compared to the current Bi-LRNN model. In addition, we propose to train a parallel Bi-LRNN model based on the decoding lattices from both language models, and examine various ways of implementation. The resulting model leads to further reduction in the false trigger rate by $10.8\%$.

In this work, we uncover a theoretical connection between two language model interpolation techniques, count merging and Bayesian interpolation. We compare these techniques as well as linear interpolation in three scenarios with abundant training data per component model. Consistent with prior work, we show that both count merging and Bayesian interpolation outperform linear interpolation. We include the first (to our knowledge) published comparison of count merging and Bayesian interpolation, showing that the two techniques perform similarly. Finally, we argue that other considerations will make Bayesian interpolation the preferred approach in most circumstances.

This paper studies two ways for helping non-specialist users develop speech systems from limited data for new languages. Focused web re-crawling finds additional examples of text matching the domain as specified by the user. This improves the language model and cuts word error rate nearly in half. Iterative voice building with interleaved lexicon construction uses the voice from a previous iteration to help construct an improved voice. 4.5 hours of the user’s time reduces transcription error rate from 32% to 4%.

The problem of identifying good features for improving conventional language models like trigrams is presented as a classification task in this paper. The idea is to use various syntactic and semantic features extracted from a language for classifying between real-world articles and articles generated by sampling a trigram language model. In doing so, a good accuracy obtained on the classification task implies that the extracted features capture those aspects of the language that a trigram model may not. Such features can be used to improve the existing trigram language models. We describe the results of our experiments on the classification task performed on a Broadcast News Corpus and discuss their effects on language modeling in general.

In this paper we describe the design and implementation of a user interface for SPICE, a web-based toolkit for rapid prototyping of speech and language processing components. We report on the challenges and experiences gathered from testing these tools in an advanced graduate hands-on course, in which we created speech recognition, speech synthesis, and small-domain translation components for 10 different languages within only 6 weeks.

Many applications in information extraction, natural language understanding, information retrieval require an understanding of the semantic relations between entities. We present a comprehensive review of various aspects of the entity relation extraction task. Some of the most important supervised and semi-supervised classification approaches to the relation extraction task are covered in sufficient detail along with critical analyses. We also discuss extensions to higher-order relations. Evaluation methodologies for both supervised and semi-supervised meth-ods are described along with pointers to the commonly used performance evaluation datasets. Finally, we also give short descriptions of two important applications of relation extraction, namely question answering and biotext mining.

In this paper we aim to improve the performance of Gaussian Mixture Model (GMM) classifier using Impostor model parameters for a closed set Speaker Identification task. We propose a novel method of speaker model training which uses the parameters of an Impostor Model to discriminatively train, in order to improve the performance of the GMM based classifier. This is unlike conventional techniques which use Minimum Classification Error (MCE) criterion which uses the ensemble of speaker data, instead. An experimental comparison between the conventional GMM and the proposed technique is presented.