Search results

We are the Alquist team from CTU, Prague, Czech Republic.

We are Edina, from the University of Edinburgh, a world-leading institution in Artificial Intelligence.

Our international team of 6 PhD students and faculty advisors has a wide range of experience from both academic and industrial research.

We are five graduate and undergraduate students of cognitive science, computer science, and applied physics from Rensselaer Polytechnic Institute.

Our team has been developed from a deep learning study group at SNU.

The DeisBot team is comprised of seven graduate students in the Computational Linguistics department at Brandeis University.

Several consumer speech devices feature voice interfaces that perform on-device keyword spotting to initiate user interactions. Accurate on-device keyword spotting within a tight CPU budget is crucial for such devices. Motivated by this, we investigated two ways to improve deep neural network (DNN) acoustic models for keyword spotting without increasing CPU usage. First, we used low-rank weight matrices

We propose a max-pooling based loss function for training Long Short-Term Memory (LSTM) networks for small-footprint keyword spotting (KWS), with low CPU, memory, and latency requirements. The max-pooling loss training can be further guided by initializing with a cross-entropy loss trained network. A posterior smoothing based evaluation approach is employed to measure keyword spotting performance. Our experimental

We propose improved Deep Neural Network (DNN) training loss functions for more accurate single keyword spotting on resource-constrained embedded devices. The loss function modifications consist of a combination of multi-task training and weighted cross entropy. In the multi-task architecture, the keyword DNN acoustic model is trained with two tasks in parallel - the main task of predicting the keyword-specific

Producing large enough quantities of high-quality transcriptions for accurate and reliable evaluation of an automatic speech recognition (ASR) system can be costly. It is therefore desirable to minimize the manual transcription work for producing metrics with an agreed precision. In this paper we demonstrate how to improve ASR evaluation precision using stratified sampling. We show that by altering the

Voice search applications are typically evaluated by comparing the predicted query to a reference human transcript, regardless of the search results returned by the query. While we find that an exact transcript match is highly indicative of user satisfaction, a transcript which does not match the reference still produces satisfactory search results a significant fraction of the time. This paper therefore

We exhibit a foldable Extended Kalman Filter that internally integrates non-linear equations of motion with a nested fold of generic integrators over lazy streams in constant memory. Functional form allows us to switch integrators easily and to diagnose filter divergence accurately, achieving orders of magnitude better speed than the source example from the literature. As with all Kalman folds, we can move

The goal of this paper is to simulate the benefits of jointly applying active learning (AL) and semi-supervised training (SST) in a new speech recognition application. Our data selection approach relies on confidence filtering, and its impact on both the acoustic and language models (AM and LM) is studied. While AL is known to be beneficial to AM training, we show that it also carries out substantial improvements

We present a new model called LATTICERNN, which generalizes recurrent neural networks (RNNs) to process weighted lattices as input, instead of sequences. A LATTICERNN can encode the complete structure of a lattice into a dense representation, which makes it suitable to a variety of problems, including rescoring, classifying, parsing, or translating lattices using deep neural networks (DNNs). In this paper

We propose two new methods of speech detection in the context of voice-controlled far-field appliances. While conventional detection methods are designed to differentiate between speech and nonspeech, we aim at distinguishing desired speech, which we define as speech originating from the person interacting with the device, from background noise and interfering talkers. Our two proposed methods use the first

Search queries are appropriate when users have explicit intent, but they perform poorly when the intent is difficult to express or if the user is simply looking to be inspired. Visual browsing systems allow e-commerce platforms to address these scenarios while offering the user an engaging shopping experience. Here we explore extensions in the direction of adaptive personalization and item diversification

Amazon is one of the world’s largest e-commerce sites and Amazon Search powers the majority of Amazon’s sales. As a consequence, even small improvements in relevance ranking both positively influence the shopping experience of millions of customers and significantly impact revenue. In the past, Amazon’s product search engine consisted of several handtuned ranking functions using a handful of input features

An adaptive image sampling framework is proposed for identifying text regions in natural scene images. A small fraction of the pixels actually correspond to text regions. It is desirable to eliminate non-text regions at the early stages of text detection. First, the image is sampled row-by-row at a specific rate and each row is tested for containing text using an 1D adaptation of the Maximally Stable Extremal

In structured prediction, a predictor optimizes an objective function over a combinatorial search space, such as the set of all image segmentations, or the set of all part-of-speech taggings. Unfortunately, finding the optimal structured labeling—sometimes referred to as maximum a posteriori (MAP) inference—is, in general, NP-hard [12], due to the combinatorial structure of the problem. Many inference approximations

We present a scalable and robust Bayesian method for demand forecasting in the context of a large e-commerce platform, paying special attention to intermittent and bursty target statistics. Inference is approximated by the Newton-Raphson algorithm, reduced to linear-time Kalman smoothing, which allows us to operate on several orders of magnitude larger problems than previous related work. In a study on