본문

서브메뉴

Improving the Accuracy and Inference Efficiency for Low-Resource Automatic Speech Recognition
Improving the Accuracy and Inference Efficiency for Low-Resource Automatic Speech Recognit...
Improving the Accuracy and Inference Efficiency for Low-Resource Automatic Speech Recognition

상세정보

자료유형  
 학위논문 서양
최종처리일시  
20250211151045
ISBN  
9798381968996
DDC  
621.3
저자명  
Fan, Ruchao.
서명/저자  
Improving the Accuracy and Inference Efficiency for Low-Resource Automatic Speech Recognition
발행사항  
[Sl] : University of California, Los Angeles, 2024
발행사항  
Ann Arbor : ProQuest Dissertations & Theses, 2024
형태사항  
130 p
주기사항  
Source: Dissertations Abstracts International, Volume: 85-09, Section: A.
주기사항  
Advisor: Alwan, Abeer A.
학위논문주기  
Thesis (Ph.D.)--University of California, Los Angeles, 2024.
초록/해제  
요약Automatic speech recognition (ASR) systems have improved significantly in the last decade due to advances in deep learning algorithms and easier access to very large databases. ASR systems, however, face two major challenges. The first challenge is in low-resource situations, such as child speech, where the accuracy degrades significantly, and the second challenge concerns low inference efficiency due to the autoregressive mechanism and size of ASR models. In this dissertation, we address these challenges by introducing novel techniques to improve the accuracy and the inference efficiency for ASR tasks, especially child ASR. To address the accuracy challenge, we introduce novel self-supervised learning (SSL) methods using un-annotated adult speech data and explore how these methods can improve the downstream child ASR tasks. Specifically, a bidirectional autoregressive predictive coding (Bi-APC) method is proposed for non-causal models pretraining with the usage of adult speech data. The pretrained model is then finetuned on supervised child speech-text pairs. We also propose a novel framework, domain responsible adaptation and finetuning (DRAFT), to reduce the domain shifting in pretrained speech models. The DRAFT framework is effective for APC that uses a causal transformer as the backbone, and for Bi-APC, Wav2vec2.0 and HuBERT methods, which use a non-causal transformer as the backbone. To address the inference efficiency challenge, we introduce a novel Connectionist Temporal Classification (CTC) Alignment-based Single-Step Non-Autoregressive Transformer (CASS-NAT) for end-to-end ASR. A comprehensive evaluation of CASS-NAT is performed in this dissertation. In CASS-NAT, the word embeddings in an autoregressive transformer (AT) are substituted with token-level acoustic embeddings (TAE) that are extracted based on the encoder outputs and CTC alignments. TAE can be obtained simultaneously without recurrent operations, leading to a parallel generation of output sequences. In addition, an error-based alignment sampling method is proposed to reduce alignment mismatch between the training and inference. CASS-NAT achieves ~20x speed up during inference without significant performance degradation compared to AT. We also propose a CASS-NAT variant (UniEnc-CASSNAT) that consists of only an encoder module. Together with the proposed multi-pass CTC training and iterative decoding, UniEnc-CASSNAT can perform as well as CASS-NAT with fewer model parameters. Beyond these two challenges, and in order to facilitate the development of better child ASR, we build the first child ASR benchmark for the research community. The benchmark includes comparisons of widely-used techniques for ASR such as data augmentation, parameter efficient finetuning (PEFT), self-supervised (HuBERT and WavLM) and supervised models (such as Whisper). All codes developed in this dissertation will be available.
일반주제명  
Electrical engineering
일반주제명  
Computer science
일반주제명  
Communication
키워드  
Adapter finetuning
키워드  
Automatic speech recognition
키워드  
Non-autoregressive generation
키워드  
Self-supervised learning
키워드  
Speech foundation models
기타저자  
University of California, Los Angeles Electrical and Computer Engineering 0333
기본자료저록  
Dissertations Abstracts International. 85-09A.
전자적 위치 및 접속  
로그인 후 원문을 볼 수 있습니다.

MARC

 008250123s2024        us                              c    eng  d
■001000017160589
■00520250211151045
■006m          o    d                
■007cr#unu||||||||
■020    ▼a9798381968996
■035    ▼a(MiAaPQ)AAI31140699
■040    ▼aMiAaPQ▼cMiAaPQ
■0820  ▼a621.3
■1001  ▼aFan,  Ruchao.
■24510▼aImproving  the  Accuracy  and  Inference  Efficiency  for  Low-Resource  Automatic  Speech  Recognition
■260    ▼a[Sl]▼bUniversity  of  California,  Los  Angeles▼c2024
■260  1▼aAnn  Arbor▼bProQuest  Dissertations  &  Theses▼c2024
■300    ▼a130  p
■500    ▼aSource:  Dissertations  Abstracts  International,  Volume:  85-09,  Section:  A.
■500    ▼aAdvisor:  Alwan,  Abeer  A.
■5021  ▼aThesis  (Ph.D.)--University  of  California,  Los  Angeles,  2024.
■520    ▼aAutomatic  speech  recognition  (ASR)  systems  have  improved  significantly  in  the  last  decade  due  to  advances  in  deep  learning  algorithms  and  easier  access  to  very  large  databases.  ASR  systems,  however,  face  two  major  challenges.  The  first  challenge  is  in  low-resource  situations,  such  as  child  speech,  where  the  accuracy  degrades  significantly,  and  the  second  challenge  concerns  low  inference  efficiency  due  to  the  autoregressive  mechanism  and  size  of  ASR  models.  In  this  dissertation,  we  address  these  challenges  by  introducing  novel  techniques  to  improve  the  accuracy  and  the  inference  efficiency  for  ASR  tasks,  especially  child  ASR.  To  address  the  accuracy  challenge,  we  introduce  novel  self-supervised  learning  (SSL)  methods  using  un-annotated  adult  speech  data  and  explore  how  these  methods  can  improve  the  downstream  child  ASR  tasks.  Specifically,  a  bidirectional  autoregressive  predictive  coding  (Bi-APC)  method  is  proposed  for  non-causal  models  pretraining  with  the  usage  of  adult  speech  data.  The  pretrained  model  is  then  finetuned  on  supervised  child  speech-text  pairs.  We  also  propose  a  novel  framework,  domain  responsible  adaptation  and  finetuning  (DRAFT),  to  reduce  the  domain  shifting  in  pretrained  speech  models.  The  DRAFT  framework  is  effective  for  APC  that  uses  a  causal  transformer  as  the  backbone,  and  for  Bi-APC,  Wav2vec2.0  and  HuBERT  methods,  which  use  a  non-causal  transformer  as  the  backbone.  To  address  the  inference  efficiency  challenge,  we  introduce  a  novel  Connectionist  Temporal  Classification  (CTC)  Alignment-based  Single-Step  Non-Autoregressive  Transformer  (CASS-NAT)  for  end-to-end  ASR.  A  comprehensive  evaluation  of  CASS-NAT  is  performed  in  this  dissertation.  In  CASS-NAT,  the  word  embeddings  in  an  autoregressive  transformer  (AT)  are  substituted  with  token-level  acoustic  embeddings  (TAE)  that  are  extracted  based  on  the  encoder  outputs  and  CTC  alignments.  TAE  can  be  obtained  simultaneously  without  recurrent  operations,  leading  to  a  parallel  generation  of  output  sequences.  In  addition,  an  error-based  alignment  sampling  method  is  proposed  to  reduce  alignment  mismatch  between  the  training  and  inference.  CASS-NAT  achieves  ~20x  speed  up  during  inference  without  significant  performance  degradation  compared  to  AT.  We  also  propose  a  CASS-NAT  variant  (UniEnc-CASSNAT)  that  consists  of  only  an  encoder  module.  Together  with  the  proposed  multi-pass  CTC  training  and  iterative  decoding,  UniEnc-CASSNAT  can  perform  as  well  as  CASS-NAT  with  fewer  model  parameters.  Beyond  these  two  challenges,  and  in  order  to  facilitate  the  development  of  better  child  ASR,  we  build  the  first  child  ASR  benchmark  for  the  research  community.  The  benchmark  includes  comparisons  of  widely-used  techniques  for  ASR  such  as  data  augmentation,  parameter  efficient  finetuning  (PEFT),  self-supervised  (HuBERT  and  WavLM)  and  supervised  models  (such  as  Whisper).  All  codes  developed  in  this  dissertation  will  be  available.
■590    ▼aSchool  code:  0031.
■650  4▼aElectrical  engineering
■650  4▼aComputer  science
■650  4▼aCommunication
■653    ▼aAdapter  finetuning
■653    ▼aAutomatic  speech  recognition
■653    ▼aNon-autoregressive  generation
■653    ▼aSelf-supervised  learning
■653    ▼aSpeech  foundation  models
■690    ▼a0544
■690    ▼a0984
■690    ▼a0459
■71020▼aUniversity  of  California,  Los  Angeles▼bElectrical  and  Computer  Engineering  0333.
■7730  ▼tDissertations  Abstracts  International▼g85-09A.
■790    ▼a0031
■791    ▼aPh.D.
■792    ▼a2024
■793    ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17160589▼nKERIS▼z이  자료의  원문은  한국교육학술정보원에서  제공합니다.

미리보기

내보내기

chatGPT토론

Ai 추천 관련 도서


    신착도서 더보기
    최근 3년간 통계입니다.

    소장정보

    • 예약
    • 소재불명신고
    • 나의폴더
    • 우선정리요청
    • 비도서대출신청
    • 야간 도서대출신청
    소장자료
    등록번호 청구기호 소장처 대출가능여부 대출정보
    TF11293 전자도서 대출가능 마이폴더 부재도서신고 비도서대출신청 야간 도서대출신청

    * 대출중인 자료에 한하여 예약이 가능합니다. 예약을 원하시면 예약버튼을 클릭하십시오.

    해당 도서를 다른 이용자가 함께 대출한 도서

    관련 인기도서

    로그인 후 이용 가능합니다.