Matches in SemOpenAlex for { <https://semopenalex.org/work/W4319862415> ?p ?o ?g. }
Showing items 1 to 82 of
82
with 100 items per page.
- W4319862415 abstract "For on-device automatic speech recognition (ASR), quantization aware training (QAT) is ubiquitous to achieve the trade-off between model predictive performance and efficiency. Among existing QAT methods, one major drawback is that the quantization centroids have to be predetermined and fixed. To overcome this limitation, we introduce a regularization-free, “soft-to-hard” compression mechanism with self-adjustable centroids in a <tex xmlns:mml=http://www.w3.org/1998/Math/MathML xmlns:xlink=http://www.w3.org/1999/xlink>$mu$</tex> -Law constrained space, resulting in a simpler yet more versatile quantization scheme, called General Quantizer (GQ). We apply GQ to ASR tasks using Recurrent Neural Network Transducer (RNN-T) and Conformer architectures on both LibriSpeech and de-identified far-field datasets. Without accuracy degradation, GQ can compress both RNN-T and Conformer into sub-8-bit, and for some RNN-T layers, to 1-bit for fast and accurate inference. We observe a 30.73% memory footprint saving and 31.75% user-perceived latency reduction compared to 8-bit QAT via physical device benchmarking." @default.
- W4319862415 created "2023-02-11" @default.
- W4319862415 creator A5023415990 @default.
- W4319862415 creator A5023708721 @default.
- W4319862415 creator A5029573067 @default.
- W4319862415 creator A5038985606 @default.
- W4319862415 creator A5047870570 @default.
- W4319862415 creator A5062762746 @default.
- W4319862415 date "2023-01-09" @default.
- W4319862415 modified "2023-09-27" @default.
- W4319862415 title "Sub-8-Bit Quantization for On-Device Speech Recognition: A Regularization-Free Approach" @default.
- W4319862415 cites W2061715260 @default.
- W4319862415 cites W2936774411 @default.
- W4319862415 cites W2963414781 @default.
- W4319862415 cites W2964110616 @default.
- W4319862415 cites W2972354707 @default.
- W4319862415 cites W3097522836 @default.
- W4319862415 cites W3097777922 @default.
- W4319862415 cites W3151287998 @default.
- W4319862415 cites W3160766462 @default.
- W4319862415 cites W3163368926 @default.
- W4319862415 cites W3177113534 @default.
- W4319862415 cites W3177265267 @default.
- W4319862415 cites W3186609711 @default.
- W4319862415 cites W3196364802 @default.
- W4319862415 cites W3197125019 @default.
- W4319862415 cites W3202442802 @default.
- W4319862415 cites W3215615641 @default.
- W4319862415 cites W4221138270 @default.
- W4319862415 cites W4224215744 @default.
- W4319862415 cites W4224918069 @default.
- W4319862415 cites W4225307083 @default.
- W4319862415 cites W4284957875 @default.
- W4319862415 cites W4297841491 @default.
- W4319862415 cites W4297841771 @default.
- W4319862415 doi "https://doi.org/10.1109/slt54892.2023.10022821" @default.
- W4319862415 hasPublicationYear "2023" @default.
- W4319862415 type Work @default.
- W4319862415 citedByCount "0" @default.
- W4319862415 crossrefType "proceedings-article" @default.
- W4319862415 hasAuthorship W4319862415A5023415990 @default.
- W4319862415 hasAuthorship W4319862415A5023708721 @default.
- W4319862415 hasAuthorship W4319862415A5029573067 @default.
- W4319862415 hasAuthorship W4319862415A5038985606 @default.
- W4319862415 hasAuthorship W4319862415A5047870570 @default.
- W4319862415 hasAuthorship W4319862415A5062762746 @default.
- W4319862415 hasConcept C111919701 @default.
- W4319862415 hasConcept C11413529 @default.
- W4319862415 hasConcept C118505674 @default.
- W4319862415 hasConcept C146599234 @default.
- W4319862415 hasConcept C154945302 @default.
- W4319862415 hasConcept C2776135515 @default.
- W4319862415 hasConcept C2776214188 @default.
- W4319862415 hasConcept C28490314 @default.
- W4319862415 hasConcept C28855332 @default.
- W4319862415 hasConcept C41008148 @default.
- W4319862415 hasConceptScore W4319862415C111919701 @default.
- W4319862415 hasConceptScore W4319862415C11413529 @default.
- W4319862415 hasConceptScore W4319862415C118505674 @default.
- W4319862415 hasConceptScore W4319862415C146599234 @default.
- W4319862415 hasConceptScore W4319862415C154945302 @default.
- W4319862415 hasConceptScore W4319862415C2776135515 @default.
- W4319862415 hasConceptScore W4319862415C2776214188 @default.
- W4319862415 hasConceptScore W4319862415C28490314 @default.
- W4319862415 hasConceptScore W4319862415C28855332 @default.
- W4319862415 hasConceptScore W4319862415C41008148 @default.
- W4319862415 hasLocation W43198624151 @default.
- W4319862415 hasOpenAccess W4319862415 @default.
- W4319862415 hasPrimaryLocation W43198624151 @default.
- W4319862415 hasRelatedWork W2012337073 @default.
- W4319862415 hasRelatedWork W2030140005 @default.
- W4319862415 hasRelatedWork W2161613934 @default.
- W4319862415 hasRelatedWork W2358489738 @default.
- W4319862415 hasRelatedWork W2376800298 @default.
- W4319862415 hasRelatedWork W2383026893 @default.
- W4319862415 hasRelatedWork W2385621972 @default.
- W4319862415 hasRelatedWork W2778151824 @default.
- W4319862415 hasRelatedWork W3126277921 @default.
- W4319862415 hasRelatedWork W3127732896 @default.
- W4319862415 isParatext "false" @default.
- W4319862415 isRetracted "false" @default.
- W4319862415 workType "article" @default.