Small Footprint Convolutional Recurrent Networks for Streaming Wakeword Detection

Small Footprint Convolutional Recurrent Networks for Streaming Wakeword Detection

In this work, we propose small footprint Convolutional Recurrent Neural Network models applied to the problem of wakeword detection and augment them with scaled dot product attention. We find that false accepts compared to Convolutional Neural Network models in a 250k parameter budget can be reduced...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Khursheed, Mohammad Omar, Jose, Christin, Kumar, Rajath, Fu, Gengshen, Kulis, Brian, Cheekatmalla, Santosh Kumar
Format: Artikel
Sprache:eng
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page
container_issue
container_start_page
container_title
container_volume
creator Khursheed, Mohammad Omar
Jose, Christin
Kumar, Rajath
Fu, Gengshen
Kulis, Brian
Cheekatmalla, Santosh Kumar
description In this work, we propose small footprint Convolutional Recurrent Neural Network models applied to the problem of wakeword detection and augment them with scaled dot product attention. We find that false accepts compared to Convolutional Neural Network models in a 250k parameter budget can be reduced by 25% with a 10% reduction in parameter size by using CRNNs, and we can get up to 32% improvement at a 50k parameter budget with 75% reduction in parameter size compared to word-level Dense Neural Network models. We discuss solutions to the challenging problem of performing inference on streaming audio with CRNNs, as well as differences in start-end index errors and latency in comparison to CNN, DNN, and DNN-HMM models.
doi_str_mv 10.48550/arxiv.2011.12941
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2011_12941</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2011_12941</sourcerecordid><originalsourceid>FETCH-LOGICAL-a671-c9b44b03edaed9186ab93e1f02f66d55e51e49e1a3897dcc890c56c075c24f523</originalsourceid><addsrcrecordid>eNotj8tOwzAURL1hgQofwAr_QILfiZcoUECqQNBKLKMb-xpFTWLkugX-nrSwGulIM5pDyBVnpaq1ZjeQvvtDKRjnJRdW8XPyuh5hGOgyxvyZ-inTJk6HOOxzHycY6Bu6fUo482fMXzFtdzTERNc5IYz99EHfYYsz9_QOM7pj64KcBRh2ePmfC7JZ3m-ax2L18vDU3K4KMBUvnO2U6phED-gtrw10ViIPTARjvNaoOSqLHGRtK-9cbZnTxrFKO6GCFnJBrv9mT07tfH6E9NMe3dqTm_wFZ_pLEw</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>Small Footprint Convolutional Recurrent Networks for Streaming Wakeword Detection</title><source>arXiv.org</source><creator>Khursheed, Mohammad Omar ; Jose, Christin ; Kumar, Rajath ; Fu, Gengshen ; Kulis, Brian ; Cheekatmalla, Santosh Kumar</creator><creatorcontrib>Khursheed, Mohammad Omar ; Jose, Christin ; Kumar, Rajath ; Fu, Gengshen ; Kulis, Brian ; Cheekatmalla, Santosh Kumar</creatorcontrib><description>In this work, we propose small footprint Convolutional Recurrent Neural Network models applied to the problem of wakeword detection and augment them with scaled dot product attention. We find that false accepts compared to Convolutional Neural Network models in a 250k parameter budget can be reduced by 25% with a 10% reduction in parameter size by using CRNNs, and we can get up to 32% improvement at a 50k parameter budget with 75% reduction in parameter size compared to word-level Dense Neural Network models. We discuss solutions to the challenging problem of performing inference on streaming audio with CRNNs, as well as differences in start-end index errors and latency in comparison to CNN, DNN, and DNN-HMM models.</description><identifier>DOI: 10.48550/arxiv.2011.12941</identifier><language>eng</language><creationdate>2020-11</creationdate><rights>http://arxiv.org/licenses/nonexclusive-distrib/1.0</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,780,885</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2011.12941$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2011.12941$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Khursheed, Mohammad Omar</creatorcontrib><creatorcontrib>Jose, Christin</creatorcontrib><creatorcontrib>Kumar, Rajath</creatorcontrib><creatorcontrib>Fu, Gengshen</creatorcontrib><creatorcontrib>Kulis, Brian</creatorcontrib><creatorcontrib>Cheekatmalla, Santosh Kumar</creatorcontrib><title>Small Footprint Convolutional Recurrent Networks for Streaming Wakeword Detection</title><description>In this work, we propose small footprint Convolutional Recurrent Neural Network models applied to the problem of wakeword detection and augment them with scaled dot product attention. We find that false accepts compared to Convolutional Neural Network models in a 250k parameter budget can be reduced by 25% with a 10% reduction in parameter size by using CRNNs, and we can get up to 32% improvement at a 50k parameter budget with 75% reduction in parameter size compared to word-level Dense Neural Network models. We discuss solutions to the challenging problem of performing inference on streaming audio with CRNNs, as well as differences in start-end index errors and latency in comparison to CNN, DNN, and DNN-HMM models.</description><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2020</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotj8tOwzAURL1hgQofwAr_QILfiZcoUECqQNBKLKMb-xpFTWLkugX-nrSwGulIM5pDyBVnpaq1ZjeQvvtDKRjnJRdW8XPyuh5hGOgyxvyZ-inTJk6HOOxzHycY6Bu6fUo482fMXzFtdzTERNc5IYz99EHfYYsz9_QOM7pj64KcBRh2ePmfC7JZ3m-ax2L18vDU3K4KMBUvnO2U6phED-gtrw10ViIPTARjvNaoOSqLHGRtK-9cbZnTxrFKO6GCFnJBrv9mT07tfH6E9NMe3dqTm_wFZ_pLEw</recordid><startdate>20201125</startdate><enddate>20201125</enddate><creator>Khursheed, Mohammad Omar</creator><creator>Jose, Christin</creator><creator>Kumar, Rajath</creator><creator>Fu, Gengshen</creator><creator>Kulis, Brian</creator><creator>Cheekatmalla, Santosh Kumar</creator><scope>GOX</scope></search><sort><creationdate>20201125</creationdate><title>Small Footprint Convolutional Recurrent Networks for Streaming Wakeword Detection</title><author>Khursheed, Mohammad Omar ; Jose, Christin ; Kumar, Rajath ; Fu, Gengshen ; Kulis, Brian ; Cheekatmalla, Santosh Kumar</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a671-c9b44b03edaed9186ab93e1f02f66d55e51e49e1a3897dcc890c56c075c24f523</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2020</creationdate><toplevel>online_resources</toplevel><creatorcontrib>Khursheed, Mohammad Omar</creatorcontrib><creatorcontrib>Jose, Christin</creatorcontrib><creatorcontrib>Kumar, Rajath</creatorcontrib><creatorcontrib>Fu, Gengshen</creatorcontrib><creatorcontrib>Kulis, Brian</creatorcontrib><creatorcontrib>Cheekatmalla, Santosh Kumar</creatorcontrib><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Khursheed, Mohammad Omar</au><au>Jose, Christin</au><au>Kumar, Rajath</au><au>Fu, Gengshen</au><au>Kulis, Brian</au><au>Cheekatmalla, Santosh Kumar</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Small Footprint Convolutional Recurrent Networks for Streaming Wakeword Detection</atitle><date>2020-11-25</date><risdate>2020</risdate><abstract>In this work, we propose small footprint Convolutional Recurrent Neural Network models applied to the problem of wakeword detection and augment them with scaled dot product attention. We find that false accepts compared to Convolutional Neural Network models in a 250k parameter budget can be reduced by 25% with a 10% reduction in parameter size by using CRNNs, and we can get up to 32% improvement at a 50k parameter budget with 75% reduction in parameter size compared to word-level Dense Neural Network models. We discuss solutions to the challenging problem of performing inference on streaming audio with CRNNs, as well as differences in start-end index errors and latency in comparison to CNN, DNN, and DNN-HMM models.</abstract><doi>10.48550/arxiv.2011.12941</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.2011.12941
ispartof
issn
language eng
recordid cdi_arxiv_primary_2011_12941
source arXiv.org
title Small Footprint Convolutional Recurrent Networks for Streaming Wakeword Detection
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2024-12-25T01%3A30%3A43IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-arxiv_GOX&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=Small%20Footprint%20Convolutional%20Recurrent%20Networks%20for%20Streaming%20Wakeword%20Detection&rft.au=Khursheed,%20Mohammad%20Omar&rft.date=2020-11-25&rft_id=info:doi/10.48550/arxiv.2011.12941&rft_dat=%3Carxiv_GOX%3E2011_12941%3C/arxiv_GOX%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_id=info:pmid/&rfr_iscdi=true