Masked Self-distilled Transducer-based Keyword Spotting with Semi-autoregressive Decoding Article Swipe

PDF

Yu Xi , Xiaoyu Gu , Haoyu Li , Jun Song , B. Zheng , Kai Yu ·

YOU? · · 2025 · Open Access · · DOI: https://doi.org/10.48550/arxiv.2505.24820

RNN-T-based keyword spotting (KWS) with autoregressive decoding~(AR) has gained attention due to its streaming architecture and superior performance. However, the simplicity of the prediction network in RNN-T poses an overfitting issue, especially under challenging scenarios, resulting in degraded performance. In this paper, we propose a masked self-distillation (MSD) training strategy that avoids RNN-Ts overly relying on prediction networks to alleviate overfitting. Such training enables masked non-autoregressive (NAR) decoding, which fully masks the RNN-T predictor output during KWS decoding. In addition, we propose a semi-autoregressive (SAR) decoding approach to integrate the advantages of AR and NAR decoding. Our experiments across multiple KWS datasets demonstrate that MSD training effectively alleviates overfitting. The SAR decoding method preserves the superior performance of AR decoding while benefits from the overfitting suppression of NAR decoding, achieving excellent results.

Related Topics

Truth And Reconciliation Commission Of Canada

2025 Nba Draft

28 Years Later

Mahmood Mamdani

Reich Ministry Of Public Enlightenment And Propaganda

Concepts

No concepts available.

Metadata

Type: preprint
Language: en
Landing Page: http://arxiv.org/abs/2505.24820
PDF: https://arxiv.org/pdf/2505.24820
OA Status: green
OpenAlex ID: https://openalex.org/W4414858722

All OpenAlex metadata

Raw OpenAlex JSON

OpenAlex ID: https://openalex.org/W4414858722

Canonical identifier for this work in OpenAlex
DOI: https://doi.org/10.48550/arxiv.2505.24820

Digital Object Identifier
Title: Masked Self-distilled Transducer-based Keyword Spotting with Semi-autoregressive Decoding

Work title
Type: preprint

OpenAlex work type
Language: en

Primary language
Publication year: 2025

Year of publication
Publication date: 2025-05-30

Full publication date if available
Authors: Yu Xi, Xiaoyu Gu, Haoyu Li, Jun Song, B. Zheng, Kai Yu

List of authors in order
Landing page: https://arxiv.org/abs/2505.24820

Publisher landing page
PDF URL: https://arxiv.org/pdf/2505.24820

Direct link to full text PDF
Open access: Yes

Whether a free full text is available
OA status: green

Open access status per OpenAlex
OA URL: https://arxiv.org/pdf/2505.24820

Direct OA link when available
Cited by: 0

Total citation count in OpenAlex

Full payload

id	https://openalex.org/W4414858722
doi	https://doi.org/10.48550/arxiv.2505.24820
ids.doi	https://doi.org/10.48550/arxiv.2505.24820
ids.openalex	https://openalex.org/W4414858722
fwci	0.0
type	preprint
title	Masked Self-distilled Transducer-based Keyword Spotting with Semi-autoregressive Decoding
biblio.issue
biblio.volume
biblio.last_page
biblio.first_page
topics[0].id	https://openalex.org/T13083
topics[0].field.id	https://openalex.org/fields/17
topics[0].field.display_name	Computer Science
topics[0].score	0.9764000177383423
topics[0].domain.id	https://openalex.org/domains/3
topics[0].domain.display_name	Physical Sciences
topics[0].subfield.id	https://openalex.org/subfields/1702
topics[0].subfield.display_name	Artificial Intelligence
topics[0].display_name	Advanced Text Analysis Techniques
is_xpac	False
apc_list
apc_paid
language	en
locations[0].id	pmh:oai:arXiv.org:2505.24820
locations[0].is_oa	True
locations[0].source.id	https://openalex.org/S4306400194
locations[0].source.issn
locations[0].source.type	repository
locations[0].source.is_oa	True
locations[0].source.issn_l
locations[0].source.is_core	False
locations[0].source.is_in_doaj	False
locations[0].source.display_name	arXiv (Cornell University)
locations[0].source.host_organization	https://openalex.org/I205783295
locations[0].source.host_organization_name	Cornell University
locations[0].source.host_organization_lineage	https://openalex.org/I205783295
locations[0].license
locations[0].pdf_url	https://arxiv.org/pdf/2505.24820
locations[0].version	submittedVersion
locations[0].raw_type	text
locations[0].license_id
locations[0].is_accepted	False
locations[0].is_published	False
locations[0].raw_source_name
locations[0].landing_page_url	http://arxiv.org/abs/2505.24820
locations[1].id	doi:10.48550/arxiv.2505.24820
locations[1].is_oa	True
locations[1].source.id	https://openalex.org/S4306400194
locations[1].source.issn
locations[1].source.type	repository
locations[1].source.is_oa	True
locations[1].source.issn_l
locations[1].source.is_core	False
locations[1].source.is_in_doaj	False
locations[1].source.display_name	arXiv (Cornell University)
locations[1].source.host_organization	https://openalex.org/I205783295
locations[1].source.host_organization_name	Cornell University
locations[1].source.host_organization_lineage	https://openalex.org/I205783295
locations[1].license	cc-by
locations[1].pdf_url
locations[1].version
locations[1].raw_type	article
locations[1].license_id	https://openalex.org/licenses/cc-by
locations[1].is_accepted	False
locations[1].is_published
locations[1].raw_source_name
locations[1].landing_page_url	https://doi.org/10.48550/arxiv.2505.24820
indexed_in	arxiv, datacite
authorships[0].author.id	https://openalex.org/A5102383536
authorships[0].author.orcid
authorships[0].author.display_name	Yu Xi
authorships[0].author_position	first
authorships[0].raw_author_name	Xi, Yu
authorships[0].is_corresponding	False
authorships[1].author.id	https://openalex.org/A5015718206
authorships[1].author.orcid	https://orcid.org/0000-0003-2929-2441
authorships[1].author.display_name	Xiaoyu Gu
authorships[1].author_position	middle
authorships[1].raw_author_name	Gu, Xiaoyu
authorships[1].is_corresponding	False
authorships[2].author.id	https://openalex.org/A5091165662
authorships[2].author.orcid	https://orcid.org/0000-0001-8015-5165
authorships[2].author.display_name	Haoyu Li
authorships[2].author_position	middle
authorships[2].raw_author_name	Li, Haoyu
authorships[2].is_corresponding	False
authorships[3].author.id	https://openalex.org/A5101794446
authorships[3].author.orcid	https://orcid.org/0009-0000-5875-484X
authorships[3].author.display_name	Jun Song
authorships[3].author_position	middle
authorships[3].raw_author_name	Song, Jun
authorships[3].is_corresponding	False
authorships[4].author.id	https://openalex.org/A5050479679
authorships[4].author.orcid	https://orcid.org/0000-0002-6544-429X
authorships[4].author.display_name	B. Zheng
authorships[4].author_position	middle
authorships[4].raw_author_name	Zheng, Bo
authorships[4].is_corresponding	False
authorships[5].author.id	https://openalex.org/A5001247903
authorships[5].author.orcid	https://orcid.org/0000-0002-5453-8725
authorships[5].author.display_name	Kai Yu
authorships[5].author_position	last
authorships[5].raw_author_name	Yu, Kai
authorships[5].is_corresponding	False
has_content.pdf	False
has_content.grobid_xml	False
is_paratext	False
open_access.is_oa	True
open_access.oa_url	https://arxiv.org/pdf/2505.24820
open_access.oa_status	green
open_access.any_repository_has_fulltext	False
created_date	2025-10-10T00:00:00
display_name	Masked Self-distilled Transducer-based Keyword Spotting with Semi-autoregressive Decoding
has_fulltext	False
is_retracted	False
updated_date	2025-11-06T06:51:31.235846
primary_topic.id	https://openalex.org/T13083
primary_topic.field.id	https://openalex.org/fields/17
primary_topic.field.display_name	Computer Science
primary_topic.score	0.9764000177383423
primary_topic.domain.id	https://openalex.org/domains/3
primary_topic.domain.display_name	Physical Sciences
primary_topic.subfield.id	https://openalex.org/subfields/1702
primary_topic.subfield.display_name	Artificial Intelligence
primary_topic.display_name	Advanced Text Analysis Techniques
cited_by_count	0
locations_count	2
best_oa_location.id	pmh:oai:arXiv.org:2505.24820
best_oa_location.is_oa	True
best_oa_location.source.id	https://openalex.org/S4306400194
best_oa_location.source.issn
best_oa_location.source.type	repository
best_oa_location.source.is_oa	True
best_oa_location.source.issn_l
best_oa_location.source.is_core	False
best_oa_location.source.is_in_doaj	False
best_oa_location.source.display_name	arXiv (Cornell University)
best_oa_location.source.host_organization	https://openalex.org/I205783295
best_oa_location.source.host_organization_name	Cornell University
best_oa_location.source.host_organization_lineage	https://openalex.org/I205783295
best_oa_location.license
best_oa_location.pdf_url	https://arxiv.org/pdf/2505.24820
best_oa_location.version	submittedVersion
best_oa_location.raw_type	text
best_oa_location.license_id
best_oa_location.is_accepted	False
best_oa_location.is_published	False
best_oa_location.raw_source_name
best_oa_location.landing_page_url	http://arxiv.org/abs/2505.24820
primary_location.id	pmh:oai:arXiv.org:2505.24820
primary_location.is_oa	True
primary_location.source.id	https://openalex.org/S4306400194
primary_location.source.issn
primary_location.source.type	repository
primary_location.source.is_oa	True
primary_location.source.issn_l
primary_location.source.is_core	False
primary_location.source.is_in_doaj	False
primary_location.source.display_name	arXiv (Cornell University)
primary_location.source.host_organization	https://openalex.org/I205783295
primary_location.source.host_organization_name	Cornell University
primary_location.source.host_organization_lineage	https://openalex.org/I205783295
primary_location.license
primary_location.pdf_url	https://arxiv.org/pdf/2505.24820
primary_location.version	submittedVersion
primary_location.raw_type	text
primary_location.license_id
primary_location.is_accepted	False
primary_location.is_published	False
primary_location.raw_source_name
primary_location.landing_page_url	http://arxiv.org/abs/2505.24820
publication_date	2025-05-30
publication_year	2025
referenced_works_count	0
abstract_inverted_index.a	44, 82
abstract_inverted_index.AR	92, 118
abstract_inverted_index.In	39, 78
abstract_inverted_index.an	28
abstract_inverted_index.in	25, 36
abstract_inverted_index.of	21, 91, 117, 126
abstract_inverted_index.on	55
abstract_inverted_index.to	11, 58, 87
abstract_inverted_index.we	42, 80
abstract_inverted_index.KWS	76, 100
abstract_inverted_index.MSD	104
abstract_inverted_index.NAR	94, 127
abstract_inverted_index.Our	96
abstract_inverted_index.SAR	110
abstract_inverted_index.The	109
abstract_inverted_index.and	15, 93
abstract_inverted_index.due	10
abstract_inverted_index.has	7
abstract_inverted_index.its	12
abstract_inverted_index.the	19, 22, 71, 89, 114, 123
abstract_inverted_index.Such	61
abstract_inverted_index.from	122
abstract_inverted_index.that	50, 103
abstract_inverted_index.this	40
abstract_inverted_index.with	4
abstract_inverted_index.(KWS)	3
abstract_inverted_index.(MSD)	47
abstract_inverted_index.(NAR)	66
abstract_inverted_index.(SAR)	84
abstract_inverted_index.RNN-T	26, 72
abstract_inverted_index.fully	69
abstract_inverted_index.masks	70
abstract_inverted_index.poses	27
abstract_inverted_index.under	32
abstract_inverted_index.which	68
abstract_inverted_index.while	120
abstract_inverted_index.RNN-Ts	52
abstract_inverted_index.across	98
abstract_inverted_index.avoids	51
abstract_inverted_index.during	75
abstract_inverted_index.gained	8
abstract_inverted_index.issue,	30
abstract_inverted_index.masked	45, 64
abstract_inverted_index.method	112
abstract_inverted_index.output	74
abstract_inverted_index.overly	53
abstract_inverted_index.paper,	41
abstract_inverted_index.enables	63
abstract_inverted_index.keyword	1
abstract_inverted_index.network	24
abstract_inverted_index.propose	43, 81
abstract_inverted_index.relying	54
abstract_inverted_index.However,	18
abstract_inverted_index.approach	86
abstract_inverted_index.benefits	121
abstract_inverted_index.datasets	101
abstract_inverted_index.decoding	85, 111, 119
abstract_inverted_index.degraded	37
abstract_inverted_index.multiple	99
abstract_inverted_index.networks	57
abstract_inverted_index.results.	131
abstract_inverted_index.spotting	2
abstract_inverted_index.strategy	49
abstract_inverted_index.superior	16, 115
abstract_inverted_index.training	48, 62, 105
abstract_inverted_index.achieving	129
abstract_inverted_index.addition,	79
abstract_inverted_index.alleviate	59
abstract_inverted_index.attention	9
abstract_inverted_index.decoding,	67, 128
abstract_inverted_index.decoding.	77, 95
abstract_inverted_index.excellent	130
abstract_inverted_index.integrate	88
abstract_inverted_index.predictor	73
abstract_inverted_index.preserves	113
abstract_inverted_index.resulting	35
abstract_inverted_index.streaming	13
abstract_inverted_index.advantages	90
abstract_inverted_index.alleviates	107
abstract_inverted_index.especially	31
abstract_inverted_index.prediction	23, 56
abstract_inverted_index.scenarios,	34
abstract_inverted_index.simplicity	20
abstract_inverted_index.RNN-T-based	0
abstract_inverted_index.challenging	33
abstract_inverted_index.demonstrate	102
abstract_inverted_index.effectively	106
abstract_inverted_index.experiments	97
abstract_inverted_index.overfitting	29, 124
abstract_inverted_index.performance	116
abstract_inverted_index.suppression	125
abstract_inverted_index.architecture	14
abstract_inverted_index.overfitting.	60, 108
abstract_inverted_index.performance.	17, 38
abstract_inverted_index.decoding~(AR)	6
abstract_inverted_index.autoregressive	5
abstract_inverted_index.self-distillation	46
abstract_inverted_index.non-autoregressive	65
abstract_inverted_index.semi-autoregressive	83
cited_by_percentile_year
countries_distinct_count	0
institutions_distinct_count	6
citation_normalized_percentile