Show simple item record

dc.contributor.advisorAnantha P. Chandrakasan and James R. Glass.en_US
dc.contributor.authorKoppula, Skandaen_US
dc.contributor.otherMassachusetts Institute of Technology. Department of Electrical Engineering and Computer Science.en_US
dc.date.accessioned2018-12-18T20:04:13Z
dc.date.available2018-12-18T20:04:13Z
dc.date.copyright2018en_US
dc.date.issued2018en_US
dc.identifier.urihttp://hdl.handle.net/1721.1/119777
dc.descriptionThesis: M. Eng., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 2018.en_US
dc.descriptionCataloged from PDF version of thesis.en_US
dc.descriptionIncludes bibliographical references (pages 91-96).en_US
dc.description.abstractIn this thesis, I demonstrate an approach for text-independent speaker identification, targeting evaluation on low-cost, low-resource FPGAs. In the first half of this work, we contribute a set of of speaker ID models that build on prior existing small-model state-of-art, and reduce bytesize by >85%, with a 3% accuracy change tolerance. We employ model quantization and pruning to achieve this size reduction. To the best of our knowledge, this is the first speaker identification model sized to fit in the on-chip memory of commodity FPGAs, allowing us to reduce power consumption. Our experiments allow us to illustrate the accuracy/memory-footprint trade-off for baseline and compressed speaker identification models. Second, I build an RTL design for efficient evaluation of a subset of our speaker ID models. In particular, I design, implement, and benchmark architectures for low-precision fixed point neural network evaluation and ternary network evaluation. Compared to a baseline full-precision network accelerator with the same timing constraints based on designs from prior work, our low-precision, sparsity-cognizant design decreases LUT/FF resource utilization by 27% and power consumption by 12% in simulation [Chen et al., 2017]. This work has applications to the growing number of speech systems run on consumer devices and data centers. A demonstration video and testing logs illustrating results are available at https: //skoppula. github. io/thesis.html.en_US
dc.description.statementofresponsibilityby Skanda Koppula.en_US
dc.format.extent96 pagesen_US
dc.language.isoengen_US
dc.publisherMassachusetts Institute of Technologyen_US
dc.rightsMIT theses are protected by copyright. They may be viewed, downloaded, or printed from this source but further reproduction or distribution in any format is prohibited without written permission.en_US
dc.rights.urihttp://dspace.mit.edu/handle/1721.1/7582en_US
dc.subjectElectrical Engineering and Computer Science.en_US
dc.titleEnergy-efficient speaker identification with low-precision networksen_US
dc.typeThesisen_US
dc.description.degreeM. Eng.en_US
dc.contributor.departmentMassachusetts Institute of Technology. Department of Electrical Engineering and Computer Science
dc.identifier.oclc1078689815en_US


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record