Decoder-only Architecture for Speech Recognition with CTC Prompts and Text Data Augmentation

Collecting audio-text pairs is expensive; however, it is much easier to access text-only data. Unless using shallow fusion, end-to-end automatic speech recognition (ASR) models require architecture modifications or additional training schemes to use text-only data. Inspired by recent advances in dec...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Tsunoo, Emiru, Futami, Hayato, Kashiwagi, Yosuke, Arora, Siddhant, Watanabe, Shinji
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 Tsunoo, Emiru
Futami, Hayato
Kashiwagi, Yosuke
Arora, Siddhant
Watanabe, Shinji
description Collecting audio-text pairs is expensive; however, it is much easier to access text-only data. Unless using shallow fusion, end-to-end automatic speech recognition (ASR) models require architecture modifications or additional training schemes to use text-only data. Inspired by recent advances in decoder-only language models (LMs), such as GPT-3 and PaLM adopted for speech-processing tasks, we propose using a decoder-only architecture for ASR with simple text augmentation. To provide audio information, encoder features compressed by CTC prediction are used as prompts for the decoder, which can be regarded as refining CTC prediction using the decoder-only model. Because the decoder architecture is the same as an autoregressive LM, it is simple to enhance the model by leveraging external text data with LM training. An experimental comparison using LibriSpeech and Switchboard shows that our proposed models with text augmentation training reduced word error rates from ordinary CTC by 0.3% and 1.4% on LibriSpeech test-clean and testother set, respectively, and 2.9% and 5.0% on Switchboard and CallHome. The proposed model had advantage on computational efficiency compared with conventional encoder-decoder ASR models with a similar parameter setup, and outperformed them on the LibriSpeech 100h and Switchboard training scenarios.
doi_str_mv 10.48550/arxiv.2309.08876
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2309_08876</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2309_08876</sourcerecordid><originalsourceid>FETCH-LOGICAL-a676-5123a8435d694fbb4565fecb418ca1314e89bb063f578dec23a4d1327bc6def73</originalsourceid><addsrcrecordid>eNotz8lOwzAUhWFvWKDCA7DivkBCHA9xllHKJFUCQZZIkYfrxlKTVK4L7dvTFlZn8-lIPyF3tMi5EqJ40PEQvvOSFXVeKFXJa_K1RDs7jNk8bY7QRDuEhDbtI4KfI3xuEe0AHye0nkIK8wQ_IQ3Qdi28x3ncph3oyUGHhwRLnTQ0-_WIU9Jne0OuvN7s8PZ_F6R7euzal2z19vzaNqtMy0pmgpZMK86EkzX3xnAhhUdrOFVWU0Y5qtqYQjIvKuXQnjR3lJWVsdKhr9iC3P_dXvL6bQyjjsf-nNlfMtkvTjlN_g</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>Decoder-only Architecture for Speech Recognition with CTC Prompts and Text Data Augmentation</title><source>arXiv.org</source><creator>Tsunoo, Emiru ; Futami, Hayato ; Kashiwagi, Yosuke ; Arora, Siddhant ; Watanabe, Shinji</creator><creatorcontrib>Tsunoo, Emiru ; Futami, Hayato ; Kashiwagi, Yosuke ; Arora, Siddhant ; Watanabe, Shinji</creatorcontrib><description>Collecting audio-text pairs is expensive; however, it is much easier to access text-only data. Unless using shallow fusion, end-to-end automatic speech recognition (ASR) models require architecture modifications or additional training schemes to use text-only data. Inspired by recent advances in decoder-only language models (LMs), such as GPT-3 and PaLM adopted for speech-processing tasks, we propose using a decoder-only architecture for ASR with simple text augmentation. To provide audio information, encoder features compressed by CTC prediction are used as prompts for the decoder, which can be regarded as refining CTC prediction using the decoder-only model. Because the decoder architecture is the same as an autoregressive LM, it is simple to enhance the model by leveraging external text data with LM training. An experimental comparison using LibriSpeech and Switchboard shows that our proposed models with text augmentation training reduced word error rates from ordinary CTC by 0.3% and 1.4% on LibriSpeech test-clean and testother set, respectively, and 2.9% and 5.0% on Switchboard and CallHome. The proposed model had advantage on computational efficiency compared with conventional encoder-decoder ASR models with a similar parameter setup, and outperformed them on the LibriSpeech 100h and Switchboard training scenarios.</description><identifier>DOI: 10.48550/arxiv.2309.08876</identifier><language>eng</language><subject>Computer Science - Sound</subject><creationdate>2023-09</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/2309.08876$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2309.08876$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Tsunoo, Emiru</creatorcontrib><creatorcontrib>Futami, Hayato</creatorcontrib><creatorcontrib>Kashiwagi, Yosuke</creatorcontrib><creatorcontrib>Arora, Siddhant</creatorcontrib><creatorcontrib>Watanabe, Shinji</creatorcontrib><title>Decoder-only Architecture for Speech Recognition with CTC Prompts and Text Data Augmentation</title><description>Collecting audio-text pairs is expensive; however, it is much easier to access text-only data. Unless using shallow fusion, end-to-end automatic speech recognition (ASR) models require architecture modifications or additional training schemes to use text-only data. Inspired by recent advances in decoder-only language models (LMs), such as GPT-3 and PaLM adopted for speech-processing tasks, we propose using a decoder-only architecture for ASR with simple text augmentation. To provide audio information, encoder features compressed by CTC prediction are used as prompts for the decoder, which can be regarded as refining CTC prediction using the decoder-only model. Because the decoder architecture is the same as an autoregressive LM, it is simple to enhance the model by leveraging external text data with LM training. An experimental comparison using LibriSpeech and Switchboard shows that our proposed models with text augmentation training reduced word error rates from ordinary CTC by 0.3% and 1.4% on LibriSpeech test-clean and testother set, respectively, and 2.9% and 5.0% on Switchboard and CallHome. The proposed model had advantage on computational efficiency compared with conventional encoder-decoder ASR models with a similar parameter setup, and outperformed them on the LibriSpeech 100h and Switchboard training scenarios.</description><subject>Computer Science - Sound</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2023</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotz8lOwzAUhWFvWKDCA7DivkBCHA9xllHKJFUCQZZIkYfrxlKTVK4L7dvTFlZn8-lIPyF3tMi5EqJ40PEQvvOSFXVeKFXJa_K1RDs7jNk8bY7QRDuEhDbtI4KfI3xuEe0AHye0nkIK8wQ_IQ3Qdi28x3ncph3oyUGHhwRLnTQ0-_WIU9Jne0OuvN7s8PZ_F6R7euzal2z19vzaNqtMy0pmgpZMK86EkzX3xnAhhUdrOFVWU0Y5qtqYQjIvKuXQnjR3lJWVsdKhr9iC3P_dXvL6bQyjjsf-nNlfMtkvTjlN_g</recordid><startdate>20230916</startdate><enddate>20230916</enddate><creator>Tsunoo, Emiru</creator><creator>Futami, Hayato</creator><creator>Kashiwagi, Yosuke</creator><creator>Arora, Siddhant</creator><creator>Watanabe, Shinji</creator><scope>AKY</scope><scope>GOX</scope></search><sort><creationdate>20230916</creationdate><title>Decoder-only Architecture for Speech Recognition with CTC Prompts and Text Data Augmentation</title><author>Tsunoo, Emiru ; Futami, Hayato ; Kashiwagi, Yosuke ; Arora, Siddhant ; Watanabe, Shinji</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a676-5123a8435d694fbb4565fecb418ca1314e89bb063f578dec23a4d1327bc6def73</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2023</creationdate><topic>Computer Science - Sound</topic><toplevel>online_resources</toplevel><creatorcontrib>Tsunoo, Emiru</creatorcontrib><creatorcontrib>Futami, Hayato</creatorcontrib><creatorcontrib>Kashiwagi, Yosuke</creatorcontrib><creatorcontrib>Arora, Siddhant</creatorcontrib><creatorcontrib>Watanabe, Shinji</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Tsunoo, Emiru</au><au>Futami, Hayato</au><au>Kashiwagi, Yosuke</au><au>Arora, Siddhant</au><au>Watanabe, Shinji</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Decoder-only Architecture for Speech Recognition with CTC Prompts and Text Data Augmentation</atitle><date>2023-09-16</date><risdate>2023</risdate><abstract>Collecting audio-text pairs is expensive; however, it is much easier to access text-only data. Unless using shallow fusion, end-to-end automatic speech recognition (ASR) models require architecture modifications or additional training schemes to use text-only data. Inspired by recent advances in decoder-only language models (LMs), such as GPT-3 and PaLM adopted for speech-processing tasks, we propose using a decoder-only architecture for ASR with simple text augmentation. To provide audio information, encoder features compressed by CTC prediction are used as prompts for the decoder, which can be regarded as refining CTC prediction using the decoder-only model. Because the decoder architecture is the same as an autoregressive LM, it is simple to enhance the model by leveraging external text data with LM training. An experimental comparison using LibriSpeech and Switchboard shows that our proposed models with text augmentation training reduced word error rates from ordinary CTC by 0.3% and 1.4% on LibriSpeech test-clean and testother set, respectively, and 2.9% and 5.0% on Switchboard and CallHome. The proposed model had advantage on computational efficiency compared with conventional encoder-decoder ASR models with a similar parameter setup, and outperformed them on the LibriSpeech 100h and Switchboard training scenarios.</abstract><doi>10.48550/arxiv.2309.08876</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.2309.08876
ispartof
issn
language eng
recordid cdi_arxiv_primary_2309_08876
source arXiv.org
subjects Computer Science - Sound
title Decoder-only Architecture for Speech Recognition with CTC Prompts and Text Data Augmentation
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-02-06T05%3A02%3A13IST&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=Decoder-only%20Architecture%20for%20Speech%20Recognition%20with%20CTC%20Prompts%20and%20Text%20Data%20Augmentation&rft.au=Tsunoo,%20Emiru&rft.date=2023-09-16&rft_id=info:doi/10.48550/arxiv.2309.08876&rft_dat=%3Carxiv_GOX%3E2309_08876%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