More Like this
-
In the context of keyword spotting (KWS), the replacement of handcrafted speech features by learnable features has not yielded superior KWS performance. In this study, we demonstrate that filterbank learning outperforms handcrafted speech features for KWS whenever the number of filterbank channels is severely decreased. Reducing the number of channels might yield certain KWS performance drop, but also a substantial energy consumption reduction, which is key when deploying common always-on KWS on low-resource devices. Experimental results on a noisy version of the Google Speech Commands Dataset show that filterbank learning adapts to noise characteristics to provide a higher degree of robustness to noise, especially when dropout is integrated. Thus, switching from typically used 40-channel log-Mel features to 8-channel learned features leads to a relative KWS accuracy loss of only 3.5% while simultaneously achieving a 6.3× energy consumption reduction.more » « less
-
null (Ed.)The increasingly central role of speech based human computer interaction necessitates on-device, low-latency, low-power, high-accuracy key word spotting (KWS). State-of-the-art accuracies on speech-related tasks have been achieved by long short-term memory (LSTM) neural network (NN) models. Such models are typically computationally intensive because of their heavy use of Matrix vector multiplication (MVM) operations. Compute-in-Memory (CIM) architectures, while well suited to MVM operations, have not seen widespread adoption for LSTMs. In this paper we adapt resistive random access memory based CIM architectures for KWS using LSTMs. We find that a hybrid system composed of CIM cores and digital cores achieves 90% test accuracy on the google speech data set at the cost of 25 uJ/decision. Our optimized architecture uses 5-bit inputs, and analog weights to produce 6-bit outputs. All digital computation are performed with 8-bit precision leading to a 3.7× improvement in computational efficiency compared to equivalent digital systems at that accuracy.more » « less
-
Resumo Independentemente da perspetiva ou do critério de análise, os efeitos do paludismo no continente africano têm sido persistentes e profundos. Centrando-se na doença da malária e nas intervenções biomédicas do último século, Graboyes e Alidina levantam questões históricas, éticas e científicas essenciais relativamente à transmissão da verdade, à autonomia africana e às obrigações dos investigadores estrangeiros. Neste artigo, apresentam um resumo da história das iniciativas contra o paludismo que tiveram lugar em África nos últimos 120 anos, com destaque para a história global das tentativas frustradas de eliminar ou controlar a doença. Através de um estudo de caso sobre os riscos da malária ressurgente, exemplificam os problemas práticos e morais que emergem sempre que os conhecimentos históricos são ignorados. Perante os atuais apelos para que se intensifiquem os esforços em prol da erradicação da doença, Graboyes e Alidina demonstram, através de dados factuais, as razões pelas quais o conhecimento histórico tem de ser mais bem integrado nos domínios epistémicos da saúde global.more » « less
-
more » « less
The unprecedented rise of social media platforms, combined with location-aware technologies, has led to continuously producing a significant amount of geo-social data that flows as a user-generated data stream. This data has been exploited in several important use cases in various application domains. This article supports geo-social personalized queries in streaming data environments. We define temporal geo-social queries that provide users with real-time personalized answers based on their social graph. The new queries allow incorporating keyword search to get personalized results that are relevant to certain topics. To efficiently support these queries, we propose an indexing framework that provides lightweight and effective real-time indexing to digest geo-social data in real time. The framework distinguishes highly dynamic data from relatively stable data and uses appropriate data structures and a storage tier for each. Based on this framework, we propose a novel geo-social index and adopt two baseline indexes to support the addressed queries. The query processor then employs different types of pruning to efficiently access the index content and provide a real-time query response. The extensive experimental evaluation based on real datasets has shown the superiority of our proposed techniques to index real-time data and provide low-latency queries compared to existing competitors.
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1145/3557915.3560975
