Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 107 of the 107 references it cites were found in a registry or an index; none of the 39 sources its claims cite is checked for support yet.

Speaker Adaptation Using Spectro-Temporal Deep Features for Dysarthric and Elderly Speech Recognition

Open access
Geng, Mengzhe, Xie, Xurong, Ye, Zi, Wang, Tianzi, Li, Guinan, Hu, Shujie, Liu, Xunying, Meng, Helen · 2022 · IEEE/ACM Transactions on Audio Speech and Language Processing

Abstract

Despite the rapid progress of automatic speech recognition (ASR) technologies targeting normal speech in recent decades, accurate recognition of dysarthric and elderly speech remains highly challenging tasks to date. Sources of heterogeneity commonly found in normal speech including accent or gender, when further compounded with the variability over age and speech pathology severity level, create large diversity among speakers. To this end, speaker adaptation techniques play a key role in personalization of ASR systems for such users. Motivated by the spectro-temporal level differences between dysarthric, elderly and normal speech that systematically manifest in articulatory imprecision, decreased volume and clarity, slower speaking rates and increased dysfluencies, novel spectro-temporal subspace basis deep embedding features derived using SVD speech spectrum decomposition are proposed in this paper to facilitate auxiliary feature based speaker adaptation of state-of-the-art hybrid DNN/TDNN and end-to-end Conformer speech recognition systems. Experiments were conducted on four tasks: the English UASpeech and TORGO dysarthric speech corpora; the English DementiaBank Pitt and Cantonese JCCOCC MoCA elderly speech datasets. The proposed spectro-temporal deep feature adapted systems outperformed baseline i-Vector and x-Vector adaptation by up to 2.63% absolute (8.63% relative) reduction in word error rate (WER). Consistent performance improvements were retained after model based speaker adaptation using learning hidden unit contributions (LHUC) was further applied. The best speaker adapted system using the proposed spectral basis embedding features produced the lowest published WER of 25.05% on the UASpeech test set of 16 dysarthric speakers.

Overview

CoChat Summary

TL;DR

This paper proposes spectro-temporal deep embedding features, derived from SVD-based speech spectrum decomposition and bottleneck DNN classifiers, for speaker adaptation in dysarthric and elderly speech recognition. Across four tasks, the adapted ASR systems outperform i-Vector/x-Vector baselines, with the best reported UASpeech system reaching 25.05% WER.

Key findings

  • On the English UASpeech corpus, the proposed spectral basis embedding (SBE) adapted systems reduced overall WER by up to 2.63% absolute (8.63% relative) versus conventional i-Vector/x-Vector adaptation, and the best speaker-adapted system achieved 25.05% WER on the 16-speaker test set.
  • On the TORGO corpus, SBE-adapted TDNN systems improved over i-Vector adaptation by up to 1.27% absolute (14% relative) overall WER, and the SBE-adapted Conformer improved by 1.98% absolute (14.01% relative).
  • On the DementiaBank Pitt corpus, SBE-adapted hybrid TDNN systems improved over i-Vector adaptation by up to 2.57% absolute (8.1% relative) overall WER, and the SBE-adapted Conformer improved by 1.38% absolute (4.21% relative).
  • On the JCCOCC MoCA corpus, SBE-adapted hybrid TDNN systems improved over i-Vector adaptation by up to 1.07% absolute (4.44% relative) overall WER, and the SBE-adapted Conformer improved by 1.38% absolute (4.21% relative).

Methods

  • The paper decomposes mel-spectrograms with SVD into spectral and temporal basis vectors, then trains DNN classifiers on these bases to predict speech severity/age labels; the 25-dimensional bottleneck outputs are used as speaker embeddings.
  • Experiments use UASpeech and TORGO for dysarthric speech, and DementiaBank Pitt and JCCOCC MoCA for elderly speech, with data augmentation via speed perturbation and ASR systems including hybrid DNN, LF-MMI TDNN, and E2E Conformer.
  • Main comparisons are against i-Vector and x-Vector adaptation, with optional LHUC on top of auxiliary-feature adaptation; performance is measured in WER (and CER for JCCOCC MoCA).

Limitations

  • The paper states that SVD spectrum decomposition is performed in an unsupervised fashion, and that it is theoretically nontrivial to produce a perfect disentanglement between time-invariant and time-variant characteristics.
  • It explicitly notes that using both spectral and temporal embedding features (SBE+TBE) did not give a consistent benefit over spectral features alone in the dysarthric speech experiments, so the remaining experiments use SBE only.
  • Evident from the text: the paper’s gains are demonstrated on a limited set of four corpora and under specific experimental setups (selected models, language models, and augmentation), so generalization beyond these settings is not directly shown.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1109/taslp.2022.3195113

Cites

107 of 107 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

0 Valid · 0 Not valid · 24 Not checked · 35 Own results and interpretations

34 more claims

Create a free account to see all 59 claims and to verify the 23 references still open.

See all claims — free account
  • It may state two separate claims as one. 10
  • Probably not a substantial claim. 11

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Citations

Cites

Show all 107 references

Research works printed but not linked: 4

  1. lanier, 2010, J Speech Disorders
  2. young, 2002, The HTK Book, 3
  3. locatello, 2020, A sober look at the unsupervised learning of disentangled representations and their evaluation, J Mach Learn Res, 21, 1
  4. xu, 0, Speaker turn aware similarity scoring for diarization of speech-based cognitive assessments, Proc IEEE Asia-Pacific Signal Inf Process Assoc Annu Summit Conf, 1299

Cited by

See all 48

Metadata

Authors
Geng, M., Xie, X., Ye, Z., Wang, T., Li, G., Hu, S., Liu, X., & Meng, H.
Affiliations
  • Chinese University of Hong Kong (HK)
  • Chinese Academy of Sciences (CN)
Venue
IEEE/ACM Transactions on Audio Speech and Language Processing
Published
2022
Volume
30
Pages
2597-2611
ISSN
  • 2329-9290 (ISSN-L)
  • 2329-9304
OpenAlex
W4221162789
Cited by
48
References
107
Citations by year
2026: 12 · 2025: 21 · 2024: 8 · 2023: 7
FWCI
2.96
Type
journal article
Language
en
Open access
Locations
  • IEEE/ACM Transactions on Audio Speech and Language Processing (publishedVersion)
  • arXiv (Cornell University) (submittedVersion)
  • arXiv (Cornell University)
Funders
  • Hong Kong Research Council GRF (14200218, 14200220, 14200021)
  • Theme based Research Scheme (T45-407/19 N)
  • Innovation and Technology Fund (ITS/254/19)
  • National Natural Science Foundation of China (62106255)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Speech Recognition and Synthesis
  • Physical Sciences › Computer Science › Signal Processing › Speech and Audio Processing
  • Health Sciences › Health Professions › Speech and Hearing › Voice and Speech Disorders
Keywords
speaker adaptation, dysarthric speech recognition, conformational isomers, speech-language pathology

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-10-11, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-11 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Speaker Adaptation Using Spectro-Temporal Deep Features for Dysarthric and Elderly Speech Recognition. https://app.cochat.ai/papers/10.1109/taslp.2022.3195113
The paper
Geng, M., Xie, X., Ye, Z., Wang, T., Li, G., Hu, S., Liu, X., & Meng, H. (2022). Speaker Adaptation Using Spectro-Temporal Deep Features for Dysarthric and Elderly Speech Recognition. IEEE/ACM Transactions on Audio Speech and Language Processing. https://doi.org/10.1109/taslp.2022.3195113

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.