One-pass Multiple Conformer and Foundation Speech Systems Compression and Quantization Using An All-in-one Neural Model

We propose a novel one-pass multiple ASR systems joint compression and quantization approach using an all-in-one neural model. A single compression cycle allows multiple nested systems with varying Encoder depths, widths, and quantization precision settings to be simultaneously constructed without t...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Li, Zhaoqing, Xu, Haoning, Wang, Tianzi, Hu, Shoukang, Jin, Zengrui, Hu, Shujie, Deng, Jiajun, Cui, Mingyu, Geng, Mengzhe, Liu, Xunying
Format: Artikel
Sprache:eng
Schlagworte:
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 Li, Zhaoqing
Xu, Haoning
Wang, Tianzi
Hu, Shoukang
Jin, Zengrui
Hu, Shujie
Deng, Jiajun
Cui, Mingyu
Geng, Mengzhe
Liu, Xunying
description We propose a novel one-pass multiple ASR systems joint compression and quantization approach using an all-in-one neural model. A single compression cycle allows multiple nested systems with varying Encoder depths, widths, and quantization precision settings to be simultaneously constructed without the need to train and store individual target systems separately. Experiments consistently demonstrate the multiple ASR systems compressed in a single all-in-one model produced a word error rate (WER) comparable to, or lower by up to 1.01\% absolute (6.98\% relative) than individually trained systems of equal complexity. A 3.4x overall system compression and training time speed-up was achieved. Maximum model size compression ratios of 12.8x and 3.93x were obtained over the baseline Switchboard-300hr Conformer and LibriSpeech-100hr fine-tuned wav2vec2.0 models, respectively, incurring no statistically significant WER increase.
doi_str_mv 10.48550/arxiv.2406.10160
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2406_10160</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2406_10160</sourcerecordid><originalsourceid>FETCH-LOGICAL-a670-d9c2cc0567f8f5bcf30f60a50ce7ee5aa29639f40d41c626c1524a67293ac9ec3</originalsourceid><addsrcrecordid>eNotkMtOwzAURL1hgQofwAr_gIPzsNMso4gCUh9CLevo4lyDJceO7AQoX0_TsprFnBlphpC7lCfFUgj-AOHHfCVZwWWS8lTya_K9c8gGiJFuJjuawSJtvNM-9BgouI6u_OQ6GI13dD8gqk-6P8YR-3ji-iFgjLM1k68TuNH8Xti3aNwHrR2trWXGMe-QbnEKYOnGd2hvyJUGG_H2XxfksHo8NM9svXt6aeo1A1ly1lUqU4oLWeqlFu9K51xLDoIrLBEFQFbJvNIF74pUyUyqVGTFKZlVOagKVb4g95fa8_J2CKaHcGznB9rzA_kfrUtYAA</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>One-pass Multiple Conformer and Foundation Speech Systems Compression and Quantization Using An All-in-one Neural Model</title><source>arXiv.org</source><creator>Li, Zhaoqing ; Xu, Haoning ; Wang, Tianzi ; Hu, Shoukang ; Jin, Zengrui ; Hu, Shujie ; Deng, Jiajun ; Cui, Mingyu ; Geng, Mengzhe ; Liu, Xunying</creator><creatorcontrib>Li, Zhaoqing ; Xu, Haoning ; Wang, Tianzi ; Hu, Shoukang ; Jin, Zengrui ; Hu, Shujie ; Deng, Jiajun ; Cui, Mingyu ; Geng, Mengzhe ; Liu, Xunying</creatorcontrib><description>We propose a novel one-pass multiple ASR systems joint compression and quantization approach using an all-in-one neural model. A single compression cycle allows multiple nested systems with varying Encoder depths, widths, and quantization precision settings to be simultaneously constructed without the need to train and store individual target systems separately. Experiments consistently demonstrate the multiple ASR systems compressed in a single all-in-one model produced a word error rate (WER) comparable to, or lower by up to 1.01\% absolute (6.98\% relative) than individually trained systems of equal complexity. A 3.4x overall system compression and training time speed-up was achieved. Maximum model size compression ratios of 12.8x and 3.93x were obtained over the baseline Switchboard-300hr Conformer and LibriSpeech-100hr fine-tuned wav2vec2.0 models, respectively, incurring no statistically significant WER increase.</description><identifier>DOI: 10.48550/arxiv.2406.10160</identifier><language>eng</language><subject>Computer Science - Artificial Intelligence ; Computer Science - Sound</subject><creationdate>2024-06</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>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,776,881</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2406.10160$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2406.10160$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Li, Zhaoqing</creatorcontrib><creatorcontrib>Xu, Haoning</creatorcontrib><creatorcontrib>Wang, Tianzi</creatorcontrib><creatorcontrib>Hu, Shoukang</creatorcontrib><creatorcontrib>Jin, Zengrui</creatorcontrib><creatorcontrib>Hu, Shujie</creatorcontrib><creatorcontrib>Deng, Jiajun</creatorcontrib><creatorcontrib>Cui, Mingyu</creatorcontrib><creatorcontrib>Geng, Mengzhe</creatorcontrib><creatorcontrib>Liu, Xunying</creatorcontrib><title>One-pass Multiple Conformer and Foundation Speech Systems Compression and Quantization Using An All-in-one Neural Model</title><description>We propose a novel one-pass multiple ASR systems joint compression and quantization approach using an all-in-one neural model. A single compression cycle allows multiple nested systems with varying Encoder depths, widths, and quantization precision settings to be simultaneously constructed without the need to train and store individual target systems separately. Experiments consistently demonstrate the multiple ASR systems compressed in a single all-in-one model produced a word error rate (WER) comparable to, or lower by up to 1.01\% absolute (6.98\% relative) than individually trained systems of equal complexity. A 3.4x overall system compression and training time speed-up was achieved. Maximum model size compression ratios of 12.8x and 3.93x were obtained over the baseline Switchboard-300hr Conformer and LibriSpeech-100hr fine-tuned wav2vec2.0 models, respectively, incurring no statistically significant WER increase.</description><subject>Computer Science - Artificial Intelligence</subject><subject>Computer Science - Sound</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2024</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotkMtOwzAURL1hgQofwAr_gIPzsNMso4gCUh9CLevo4lyDJceO7AQoX0_TsprFnBlphpC7lCfFUgj-AOHHfCVZwWWS8lTya_K9c8gGiJFuJjuawSJtvNM-9BgouI6u_OQ6GI13dD8gqk-6P8YR-3ji-iFgjLM1k68TuNH8Xti3aNwHrR2trWXGMe-QbnEKYOnGd2hvyJUGG_H2XxfksHo8NM9svXt6aeo1A1ly1lUqU4oLWeqlFu9K51xLDoIrLBEFQFbJvNIF74pUyUyqVGTFKZlVOagKVb4g95fa8_J2CKaHcGznB9rzA_kfrUtYAA</recordid><startdate>20240614</startdate><enddate>20240614</enddate><creator>Li, Zhaoqing</creator><creator>Xu, Haoning</creator><creator>Wang, Tianzi</creator><creator>Hu, Shoukang</creator><creator>Jin, Zengrui</creator><creator>Hu, Shujie</creator><creator>Deng, Jiajun</creator><creator>Cui, Mingyu</creator><creator>Geng, Mengzhe</creator><creator>Liu, Xunying</creator><scope>AKY</scope><scope>GOX</scope></search><sort><creationdate>20240614</creationdate><title>One-pass Multiple Conformer and Foundation Speech Systems Compression and Quantization Using An All-in-one Neural Model</title><author>Li, Zhaoqing ; Xu, Haoning ; Wang, Tianzi ; Hu, Shoukang ; Jin, Zengrui ; Hu, Shujie ; Deng, Jiajun ; Cui, Mingyu ; Geng, Mengzhe ; Liu, Xunying</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a670-d9c2cc0567f8f5bcf30f60a50ce7ee5aa29639f40d41c626c1524a67293ac9ec3</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2024</creationdate><topic>Computer Science - Artificial Intelligence</topic><topic>Computer Science - Sound</topic><toplevel>online_resources</toplevel><creatorcontrib>Li, Zhaoqing</creatorcontrib><creatorcontrib>Xu, Haoning</creatorcontrib><creatorcontrib>Wang, Tianzi</creatorcontrib><creatorcontrib>Hu, Shoukang</creatorcontrib><creatorcontrib>Jin, Zengrui</creatorcontrib><creatorcontrib>Hu, Shujie</creatorcontrib><creatorcontrib>Deng, Jiajun</creatorcontrib><creatorcontrib>Cui, Mingyu</creatorcontrib><creatorcontrib>Geng, Mengzhe</creatorcontrib><creatorcontrib>Liu, Xunying</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Li, Zhaoqing</au><au>Xu, Haoning</au><au>Wang, Tianzi</au><au>Hu, Shoukang</au><au>Jin, Zengrui</au><au>Hu, Shujie</au><au>Deng, Jiajun</au><au>Cui, Mingyu</au><au>Geng, Mengzhe</au><au>Liu, Xunying</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>One-pass Multiple Conformer and Foundation Speech Systems Compression and Quantization Using An All-in-one Neural Model</atitle><date>2024-06-14</date><risdate>2024</risdate><abstract>We propose a novel one-pass multiple ASR systems joint compression and quantization approach using an all-in-one neural model. A single compression cycle allows multiple nested systems with varying Encoder depths, widths, and quantization precision settings to be simultaneously constructed without the need to train and store individual target systems separately. Experiments consistently demonstrate the multiple ASR systems compressed in a single all-in-one model produced a word error rate (WER) comparable to, or lower by up to 1.01\% absolute (6.98\% relative) than individually trained systems of equal complexity. A 3.4x overall system compression and training time speed-up was achieved. Maximum model size compression ratios of 12.8x and 3.93x were obtained over the baseline Switchboard-300hr Conformer and LibriSpeech-100hr fine-tuned wav2vec2.0 models, respectively, incurring no statistically significant WER increase.</abstract><doi>10.48550/arxiv.2406.10160</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.2406.10160
ispartof
issn
language eng
recordid cdi_arxiv_primary_2406_10160
source arXiv.org
subjects Computer Science - Artificial Intelligence
Computer Science - Sound
title One-pass Multiple Conformer and Foundation Speech Systems Compression and Quantization Using An All-in-one Neural Model
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-02-02T06%3A51%3A40IST&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=One-pass%20Multiple%20Conformer%20and%20Foundation%20Speech%20Systems%20Compression%20and%20Quantization%20Using%20An%20All-in-one%20Neural%20Model&rft.au=Li,%20Zhaoqing&rft.date=2024-06-14&rft_id=info:doi/10.48550/arxiv.2406.10160&rft_dat=%3Carxiv_GOX%3E2406_10160%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