Search

A Multilingual Keyword Spotting (KWS) system detects spokenkeywords over multiple locales. Conventional monolingual KWSapproaches do not scale well to multilingual scenarios because ofhigh development/maintenance costs and lack of resource sharing.To overcome this limit, we propose two locale-conditioned universalmodels with locale feature concatenation and feature-wise linearmodulation (FiLM). We compare these models with two baselinemethods: locale-specific monolingual KWS, and a single universalmodel trained over all data. Experiments over 10 localized languagedatasets show that locale-conditioned models substantially improveaccuracy over baseline methods across all locales in different noiseconditions.FiLMperformed the best, improving on average FRRby 61% (relative) compared to monolingual KWS models of similarsizes., Comment: Accepted for ICASSP 2023

Poster created by students of The Island School semester research class of Fall 2017

We trained a keyword spotting model using federated learning on real user devices and observed significant improvements when the model was deployed for inference on phones. To compensate for data domains that are missing from on-device training caches, we employed joint federated-centralized training. And to learn in the absence of curated labels on-device, we formulated a confidence filtering strategy based on user-feedback signals for federated distillation. These techniques created models that significantly improved quality metrics in offline evaluations and user-experience metrics in live A/B experiments., Comment: Accepted to Interspeech 2022

Acoustic Echo Cancellation (AEC) is essential for accurate recognition of queries spoken to a smart speaker that is playing out audio. Previous work has shown that a neural AEC model operating on log-mel spectral features (denoted "logmel" hereafter) can greatly improve Automatic Speech Recognition (ASR) accuracy when optimized with an auxiliary loss utilizing a pre-trained ASR model encoder. In this paper, we develop a conformer-based waveform-domain neural AEC model inspired by the "TasNet" architecture. The model is trained by jointly optimizing Negative Scale-Invariant SNR (SISNR) and ASR losses on a large speech dataset. On a realistic rerecorded test set, we find that cascading a linear adaptive AEC and a waveform-domain neural AEC is very effective, giving 56-59% word error rate (WER) reduction over the linear AEC alone. On this test set, the 1.6M parameter waveform-domain neural AEC also improves over a larger 6.5M parameter logmel-domain neural AEC model by 20-29% in easy to moderate conditions. By operating on smaller frames, the waveform neural model is able to perform better at smaller sizes and is better suited for applications where memory is limited., Comment: Submitted to Interspeech 2022

We present a frontend for improving robustness of automatic speech recognition (ASR), that jointly implements three modules within a single model: acoustic echo cancellation, speech enhancement, and speech separation. This is achieved by using a contextual enhancement neural network that can optionally make use of different types of side inputs: (1) a reference signal of the playback audio, which is necessary for echo cancellation; (2) a noise context, which is useful for speech enhancement; and (3) an embedding vector representing the voice characteristic of the target speaker of interest, which is not only critical in speech separation, but also helpful for echo cancellation and speech enhancement. We present detailed evaluations to show that the joint model performs almost as well as the task-specific models, and significantly reduces word error rate in noisy conditions even when using a large-scale state-of-the-art ASR model. Compared to the noisy baseline, the joint model reduces the word error rate in low signal-to-noise ratio conditions by at least 71% on our echo cancellation dataset, 10% on our noisy dataset, and 26% on our multi-speaker dataset. Compared to task-specific models, the joint model performs within 10% on our echo cancellation dataset, 2% on the noisy dataset, and 3% on the multi-speaker dataset., Comment: Will appear in IEEE-ASRU 2021

Highlights • Mechanization provided more predictable profit and return on investment of inputs. • Applying inputs by hand disassociated fertilizer from end of season yield. • Chest-mounted spreader can be stop-gap between hand distributed inputs and zero-till. • Labor efficiency was roughly doubled by using chest-mounted spreader. • Chest-mounted spreader increased within and between-field uniformity of yield., Nepal is at a crossroads of diminishing farm-labor and inadequate investment into farming operations that, among other factors, have stagnated domestic wheat yield. Cultural and economic constraints have hindered the widespread adoption of more expensive precision agriculture technologies like zero-till that have the capacity to improve labor and farm input efficiencies. To capture the benefits from added precision of application but with the ability to fit within the current semi-mechanized seed bed preparation and tillage system, we introduced a low-cost, chest mounted seed and fertilizer. We found that simple mechanization caused yield efficiencies to be positive and significant for nitrogen and phosphate. Seed rates using this method were positively associated with seedling density. This led to both yield and profit being more predictable for farmers. Conversely, hand-applied inputs caused a disassociation between inputs and end of season yield and therefore added a large measure of risk to their farming operations.

Le virus de l’hépatite C (VHC) est un agent causateur de maladies du foie important responsable d’une pandémie affectant près de 180 millions d’individus mondialement. L’absence de symptômes dans les premières années d’infection entraîne des diagnostics tardifs qui empêchent la prise en charge rapide des patients avant l’apparition d’une fibrose et, dans près de 16 % des cas d’infection, d’une cirrhose. En exploitant les interactions protéine-protéine membranaires, des essais utilisant la technologie BRET, dans les cellules vivantes, ont été précédemment optimisés afin d’établir le réseau complet des interactions du VHC. En utilisant les fondements de cette étude, un essai à haut débit dans les cellules vivantes a été réalisé pour identifier de nouveaux composés anti-VHC ciblant une nouvelle interaction NS3/4A-NS3/4A. Approximativement 110,000 petites molécules ont été criblées pour leurs effets sur l’homodimérization de NS3/4A et ont été classées par rapport à leur spécificité et à leur puissance contre le VHC. Au terme de cette étude, UM42811 a été identifié comme un activateur potentiel de l’interaction NS3/4A-NS3/4A offrant une activité antivirale prometteuse dotant une excellente fenêtre thérapeutique. Par la suite, un séquençage exhaustif des virus, soumis à un traitement de UM42811, a permis d’établir le profil de résistance du VHC contre ce composé. Grâce à cette fine cartographie, il a été possible d’identifier un nouveau mécanisme d’inhibition de NS3/4A qui est indépendant de son activité protéase. En utilisant les données de notre groupe sur les interactions VHC-hôte, il a été possible de continuer la caractérisation fonctionnelle du composé UM42811 en étudiant son effet sur les interactions potentiellement bénéfiques à la persistance virale. Pour ce faire, les protéines associées au transport nucléaire et mitochondriale qui sont des interactants de choix de NS3/4A ont été priorisées. Parmi ces facteurs de l’hôte, l’étude de karyopherin subunit beta 1 (KPNB1) et de heat shock protein 60 (HSP60) a été priorisée. De façon intéressante, les expériences de co-immunoprécipitation ont démontré que UM42811 était capable de prévenir l’interaction KPNB1-NS3/4A ainsi que l’interaction HSP60-NS3/4A. De plus, les études ii fonctionnelles et les analyses d’immunobuvardage de type western ont démontré que l’interaction KPNB1-NS3/4A avait des effets délétères sur l’induction des gènes stimulés par l’interféron (ISG). Finalement, il a été démontré que KPNB1 est possiblement clivé par NS3/4A suggérant la présence potentielle d’un mécanisme de subversion ou d’échappement. En bref, cette étude démontre la puissance des stratégies impliquant les interactions protéine-protéine dans les cellules vivantes pour l’identification de nouveaux composés inhibiteurs, caractérise un nouveau mécanisme d’inhibition anti-VHC et révèle la possibilité d’un nouveau mécanisme d’évasion du système immunitaire., Hepatitis C virus (HCV) is an important causative agent for liver diseases and is responsible for a worldwide pandemic affecting roughly 180 million individuals worldwide. Late diagnosis following the progression to fibrosis and to cirrhosis, in nearly 16% of chronic infections, is attributed to the absence of symptoms in the first years of infection. By exploiting membrane protein-protein interactions (PPI), live cell assays using bioluminescence resonance energy transfer (BRET) technology have previously been optimized to complete a comprehensive hepatitis C virus (HCV) protein interaction network. Using the groundwork laid by this network study, a high-throughput assay (HTS) cell-based assay was implemented to identify novel inhibitory compounds targeting an unreported NS3/4A-NS3/4A interaction. Approximately 110,000 compounds from a small-molecule collection were screened to monitor modulation of NS3/4A homodimerization and were discriminated based on specificity and potency. UM42811 was identified as a potential NS3/4A-NS3/4A interaction activator and found to have a promising antiviral activity boasting an excellent therapeutic window. Combined deep sequencing and mutation mapping have yielded a resistance profile based on statistical and functional probability pointing towards a novel inhibitory mechanism targeting the HCV NS3/4A independent from protease activity inhibition. Data from an HCV to host protein interaction network generated by our group was used to analyze alternative effects of UM42811 on interactions which potentially benefit viral persistence. NS3/4A-specific host interactors were heavily associated with nuclear and mitochondrial transport based on Gene Ontology (GO). Among these specific interactors, karyopherin subunit beta 1 (KPNB1) and heat shock protein 60 (HSP60) were selected for further study. Interestingly, co-immunoprecipitation experiments revealed that UM42811 was able to prevent both KPNB1-NS3/4A and HSP60-NS3/4A interactions. Moreover, functional and western analysis revealed the KPNB1-NS3/4A interaction to have deleterious effects on iv interferon stimulated gene (ISG) induction. Unexpectedly, analysis revealed a putative NS3/4A mediated cleavage of KPNB1. Overall, this study demonstrates the strength of cell-based PPI strategies in the identification of novel HCV antiviral compounds, characterizes a novel inhibitory mechanism for HCV and reveals a potentially novel viral immune evasion mechanism.