Quantifying and Reducing Speaker Heterogeneity within the Common Voice Corpus for Phonetic Analysis

Annie Baker; Aref Farhadipour; Bogdan Pricop; Eleanor Chodroff; Jiachen Ma; Miao Zhang

REVIEW 3 major objections 4 minor 33 references

Quantifying and Reducing Speaker Heterogeneity within the Common Voice Corpus for Phonetic Analysis

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Using voice embeddings and listener judgments, the paper sets a cosine threshold of 0.354 to flag multiple speakers inside one Common Voice client ID, at a mean cost of 3.5% of utterances per language.

desk verdict A useful resource with a real caveat: the utterance-level similarity scores and the 76-language audit are worth having, but the data-loss numbers are tied to an arbitrary enrollment choice and the threshold rests on moderate non-native agreement. read the letter →

arxiv 2506.00733 v1 pith:77N3NOJV submitted 2025-05-31 eess.AS cs.SD

Miao Zhang , Aref Farhadipour , Annie Baker , Jiachen Ma , Bogdan Pricop , Eleanor Chodroff This is my paper

classification eess.AScs.SD

keywords corpusphoneticsspeakerverificationvoiceembeddingsCommoncrosslinguisticdatacleaningperceptualauditheterogeneity

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The Mozilla Common Voice Corpus assigns recordings a client ID that is widely used as a speaker ID, but several speakers can contribute under one ID. This paper claims that cosine similarity between ResNet-293 voice embeddings flags those mixed-speaker pairs, and that a perceptual audit by five listeners locates a usable cutoff at similarity 0.354. Applied across 76 languages, that cutoff removes the low-similarity, likely other-speaker utterances while discarding a mean 3.5% and median 1.6% of utterances per language. The authors release the per-utterance similarity scores so downstream phonetic analyses can choose their own cleaning threshold.

What carries the argument

The load-bearing object is a pair of voice embeddings computed by ResNet-293, a 293-layer residual neural network for speaker verification pretrained on VoxBlink2 and fine-tuned on VoxCeleb2, joined by cosine similarity. A perceptual audit maps those continuous scores onto same-speaker and different-speaker judgments, and a generalized linear mixed model with random intercepts and slopes by annotator and language fits the crossover of the logistic curve, computed as -β0/β1 = 0.354. This machinery translates an embedding-based similarity score into a concrete cleaning rule for dropping utterances from mixed-speaker client IDs.

What would settle it

A reader could test the claim by sampling utterance pairs from the 0.25-0.45 similarity range across several of the 76 languages and having native or fluent speakers label them same/different; if per-language crossover points vary widely or the pooled value shifts away from 0.354, the single threshold fails to generalize.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that speaker heterogeneity inside a Common Voice client ID can be quantified by embedding each utterance with a multilingual ResNet-293 speaker-verification model and taking the cosine similarity against the client ID's enrollment recording. Because no ground-truth speaker labels exist for Common Voice, the authors treat five listeners' same/different judgments as ground truth. The logistic crossover of those judgments, where a listener is equally likely to call a pair same or different, falls at similarity 0.354. At that threshold, across 123,737 client IDs and 9,204,867 utterance pairs, the mean utterance loss is 3.5% per language, the median is 1.6%, and 91.9% of client IDs lose no more than 10% of their utterances. The closeness of the audited threshold to the English-only VoxCeleb1-H equal-error-rate threshold of 0.405 is taken as evidence that the system's notion of voice similarity is reasonably language-independent.

Load-bearing premise

The whole threshold rests on five listeners' same/different judgments in languages most of them did not speak, with only moderate inter-annotator agreement; if those judgments mislocate the crossover, then 0.354 is an artifact of the audit rather than a property of voice similarity across the corpus.

Editorial extensions

If this is right

Applying the 0.354 cutoff to VoxCommunis-derived data gives phonetic analyses a speaker-heterogeneity-reduced subset while retaining the vast majority of utterances: 70 of 76 languages lose less than 10% of their data.
Researchers no longer need to discard entire client IDs, as in earlier ID-level cleaning; utterance-level scores preserve usable recordings from mixed client IDs.
The audited threshold being close to the English VoxCeleb1-H equal-error-rate threshold of 0.405 suggests that English-trained verification thresholds can serve as a rough default on multilingual data.
Because the authors release similarity scores for all utterance pairs, each downstream study can select its own trade-off between speaker purity and data retention.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Extension: because the similarity-score interquartile range varies across languages, with Q1 ranging from 0.37 to 0.75, a single global threshold may over-clean high-variance languages and under-clean low-variance ones; a native-speaker audit could yield per-language cutoffs and would directly test this.
Extension: the released per-pair similarity scores can double as a diagnostic for utterance quality, not just speaker identity; trials labeled 'audio quality issue' and 'missing speech' formed 12.1% of the audit sample yet were excluded from the threshold fit, so score-quality links remain unexplored.
Extension: if the 0.354 threshold is approximately language-independent, as the paper's closeness to the English-only 0.405 suggests, the same embedding-based approach could estimate effective speaker counts in other crowdsourced speech corpora that also lack ground-truth speaker IDs.
Extension: the paper focuses on multiple speakers sharing one client ID, but a complementary problem is one speaker spread across many client IDs, which would bias variance estimates in the opposite direction.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The paper ships something worth having: utterance-level similarity scores for 9.2M Common Voice utterance pairs and a 76-language perceptual audit that yields a recommended threshold of 0.354. That is a step beyond Hintz and Siegert's client-ID-level cleaning, and the released scores let other researchers apply their own thresholds. The method is documented clearly, the annotation was blind to the scores, and the GLMM crossover is a sensible way to set a threshold without ground truth.

The soft spots are real but not fatal. The stress-test concern about enrollment is valid. Section 2.2 says the enrollment is 'the final recording associated with each client ID.' In a client ID that actually contains multiple speakers, that reference is arbitrary. All similarity scores and the resulting data-loss percentages are relative to that one recording. If the final recording belongs to a minority speaker, majority-speaker utterances will look dissimilar and get discarded, inflating the reported 3.5% mean loss. If it belongs to the majority speaker, minority utterances get discarded instead. The paper reports no sensitivity analysis—no first-recording, random, or centroid enrollment—so the 'minimizing data loss' claim in the conclusion is not yet established as a stable property of the corpus. This is an internal robustness gap, not a dispute with the field.

The second weakness is the threshold itself. The five authors, mostly non-native in the languages they judged, produced moderate agreement (kappa 0.45). The paper acknowledges this and even suggests by-language thresholds may be better. That is honest, but it means 0.354 should be read as a rough crosslinguistic recommendation, not a precise estimate. The lack of a confidence interval on the crossover point adds to that. The sanity check against the English VoxCeleb1-H threshold (0.405) is weaker than it looks, since the embedding model was fine-tuned on VoxCeleb2; still, the perceptual audit is independent of the model, so the threshold is not circular.

What the paper does well: it compares against prior work fairly, it releases the scores so others can tailor cleaning, and it frames the threshold as a recommendation rather than a gold standard. The citation pattern looks appropriate.

Who is this for? Empirical researchers using Common Voice or VoxCommunis for crosslinguistic phonetics, and anyone building speaker-based technology on messy crowdsourced data. A serious referee should engage with it, but the review should ask for enrollment robustness checks and a more diverse annotator pool (or by-language thresholds). The core resource is useful; the headline numbers should be treated as provisional.

Recommendation: send to peer review. The paper is a legitimate empirical contribution with fixable weaknesses.

Referee Report

3 major / 4 minor

Summary. The paper addresses the known problem that Mozilla Common Voice client IDs can contain multiple speakers, which is problematic for phonetic analysis. The authors use a ResNet-293 speaker verification model pretrained on VoxBlink2 to compute cosine similarity between an enrollment utterance (the final recording of each client ID) and all other utterances under that ID. To set a threshold for flagging potentially different speakers, five authors perceptually audited 2,280 utterance pairs across 76 languages, and a GLMM estimated the crossover point at similarity 0.354. Applying this threshold yields a mean utterance loss of 3.5% across languages (median 1.6%, range <0.0001% to 21.8%). The authors release the similarity scores and argue that 0.354 effectively reduces speaker heterogeneity while minimizing data loss, and that it is reliable because it is close to the VoxCeleb1-H EER threshold of 0.405.

Significance. If the central claims hold, this work would be a useful practical resource for corpus phonetics: it quantifies speaker heterogeneity in a large multilingual corpus, provides a simple utterance-level filtering criterion, and releases the underlying similarity scores so that other researchers can choose their own thresholds. The use of a multilingual speaker verification model and a perceptual audit that is independent of the embedding model are strengths. However, the current evidence does not yet support the paper's headline claim that the 0.354 threshold 'effectively reduces speaker heterogeneity while minimizing data loss': the data-loss figures are conditional on an arbitrary enrollment choice, and the threshold is a single point estimate with no reported uncertainty and only moderate inter-annotator agreement.

major comments (3)

[Section 2.2, Section 3.2, Section 5] The data-loss estimates and the 'minimizing data loss' conclusion depend on the arbitrary choice of the final recording as the enrollment utterance for each client ID. Because Common Voice client IDs may contain multiple speakers, all cosine similarities are relative to that single reference utterance. If the final recording belongs to a minority speaker within a heterogeneous client ID, utterances from the majority speaker will receive low similarity scores and be discarded, inflating data loss; if it belongs to the majority speaker, the minority speaker's utterances are discarded instead. The reported mean of 3.5% utterance loss, the per-language figures, and the client-ID-level loss proportions are therefore not stable properties of the corpus but artifacts of one enrollment convention. The manuscript provides no sensitivity analysis (e.g., first recording, random recording, or centroid embedding as enrollment). This is a load-bearing gap for the central claim and should be addressed with robustness checks.
[Section 2.3, Section 3.2, Section 4] The threshold of 0.354 is estimated from a GLMM on 2,280 trials with only moderate inter-annotator agreement (Fleiss kappa = 0.45), and no confidence interval or measure of uncertainty is reported for the crossover point. Given that the paper recommends this as a crosslinguistically reasonable universal threshold, the authors should report the sampling uncertainty of 0.354 (e.g., bootstrap or profile-likelihood CI) and examine heterogeneity of the crossover across languages. The Discussion itself acknowledges a language-familiarity effect and suggests that by-language thresholds may be more suitable, which is in tension with the conclusion's universal framing. At minimum, the paper should quantify how much the estimated threshold varies across languages or explicitly temper the crosslinguistic claim.
[Section 4] The closeness of the audited threshold (0.354) to the VoxCeleb1-H EER threshold (0.405) is presented as evidence that the threshold is 'highly reasonable and reliable.' This comparison is not quantitatively grounded: the two thresholds come from different tasks (perceptual same/different judgments versus speaker verification equal-error-rate on a specific English benchmark), and the 0.051 difference has no stated significance. Without a distribution or standard error for the audited threshold, this concordance is anecdotal. I recommend removing this claim or replacing it with a principled comparison that accounts for uncertainty.

minor comments (4)

[Section 2.3] The GLMM description ('uncorrelated random intercepts and slopes by score for annotator and language') does not make clear whether the reported 0.354 is the fixed-effect crossover or some averaged quantity; please clarify the exact model and the formula used to derive the reported point estimate.
[Figures 4 and 5] The x-axis labels in Figures 4 and 5 contain many overlapping language codes and are unreadable; consider using a table, rotating labels, or showing only selected languages.
[Title and Abstract] The title and abstract contain 'Common V oice' with an unintended space; this should be corrected to 'Common Voice'.
[Section 4] The statement that the approach is 'superior to using a threshold trained on a single language or arbitrarily selecting a threshold' is not directly supported by the results, since no comparison to those baselines is presented; please either add such a comparison or soften the claim.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the 0.354 threshold is a fitted recommendation from a perception audit that is independent of the embedding model, and the data-loss statistics are reported as conditional outcomes, not predictions.

full rationale

The derivation chain is: (i) an external ResNet-293 model (pre-trained on VoxBlink2, fine-tuned on VoxCeleb2) produces cosine similarity scores; (ii) five authors, blinded to the model scores, label sampled utterance pairs as same/different speaker; (iii) a GLMM crossover at p=0.5 yields the threshold 0.354; (iv) applying 0.354 to all 9,204,867 scored pairs yields the data-loss proportions. Each link is externally anchored. The perceptual judgments are human and score-blinded, so the threshold is not defined in terms of the corpus it later filters; the data-loss figures are explicitly conditional ('Using 0.354 as the threshold, we obtained the proportion data loss'), not predictions forced by construction; and the Discussion calls the threshold 'merely a recommendation rather than a gold standard,' disclaiming any claim that it was derived from the data it cleans. The convergence with the VoxCeleb1-H EER threshold (0.405) is an external benchmark sanity check on the same model family, not the source of the threshold. Self-citations to VoxCommunis [2, 16] and [10] are used only to define the 76-language selection and corpus context; they are not load-bearing premises for the threshold or the data-loss estimates. The reader-flagged weaknesses — moderate inter-annotator agreement (Fleiss kappa = 0.45), non-native listening by annotators, the arbitrary choice of the final recording as enrollment, and the absence of uncertainty around 0.354 — are internal robustness limitations that the authors largely acknowledge in the Discussion, not circular reductions. Accordingly the circularity score is 1: minor self-citations exist, but no load-bearing circular step was found.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The central contribution is a fitted threshold and score release. The only fitted number is tau; the procedure also leans on domain assumptions about embedding language-independence and on the validity of non-native perceptual judgments, the latter explicitly flagged by the authors.

free parameters (1)

similarity threshold tau = 0.354
Crossover point of logistic GLMM fit to 2,280 auditor judgments of same versus different speaker; no confidence interval reported. Used in Section 3.2 to compute utterance data loss.

assumptions (3)

domain assumption ResNet-293 embeddings pretrained on VoxBlink2 and fine-tuned on VoxCeleb2 represent speaker identity consistently across the 76 target languages.
Invoked in Section 2.2; all similarity scores depend on this. The paper acknowledges language effects on embeddings [18] and relies on multilingual pretraining to mitigate them.
domain assumption Human 'same speaker' versus 'different speaker' judgments are a valid ground truth for speaker heterogeneity within client IDs.
Invoked in Section 2.3. Inter-annotator agreement is only moderate (Fleiss kappa = 0.45) and annotators are non-native for most languages, as the Discussion notes.
domain assumption A single global similarity threshold is appropriate across all 76 languages.
The GLMM pools languages with random effects and yields one crossover value. The paper itself concedes by-language thresholds may be better; crosslinguistic robustness is not tested with held-out languages.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Quantifying and Reducing Speaker Heterogeneity within the Common Voice Corpus for Phonetic Analysis." pith.science (2026). https://pith.science/paper/77N3NOJV

@misc{pith2026250600733,
  author       = {Pith},
  title        = {Pith review of: Quantifying and Reducing Speaker Heterogeneity within the Common Voice Corpus for Phonetic Analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/77N3NOJV}},
  note         = {Machine review of arXiv:2506.00733}
}

read the original abstract

With its crosslinguistic and cross-speaker diversity, the Mozilla Common Voice Corpus (CV) has been a valuable resource for multilingual speech technology and holds tremendous potential for research in crosslinguistic phonetics and speech sciences. Properly accounting for speaker variation is, however, key to the theoretical and statistical bases of speech research. While CV provides a client ID as an approximation to a speaker ID, multiple speakers can contribute under the same ID. This study aims to quantify and reduce heterogeneity in the client ID for a better approximation of a true, though still anonymous speaker ID. Using ResNet-based voice embeddings, we obtained a similarity score among recordings with the same client ID, then implemented a speaker discrimination task to identify an optimal threshold for reducing perceived speaker heterogeneity. These results have major downstream applications for phonetic analysis and the development of speaker-based speech technology.

Figures

Figures reproduced from arXiv: 2506.00733 by the authors.

**Figure 1.** The distribution of the similarity scores across languages. 3.2. Auditing results In the first round of auditing, the overall distribution of validation labels in all the trials was: same speaker (40.3%), different speaker (45.2%), audio quality issue (8.6%), missing speech (3.5%), and not sure (2.4%). The result of the first round in each score bin is shown in [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

**Figure 2.** Counts of speaker validation audit by binned similarity score in the first round. The estimated crossover point from the logistic mixed model was at a similarity score of 0.354 ( [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** The distribution of same (top) and different (bottom) speaker responses, and the GLMM fit in the blue curve. The vertical line represents the similarity score (0.354) at which the probability of a same or different speaker judgment is 0.5. subset ranged from 2% to 9% of client IDs per language with a median of 4% and a mean of 8%. 0.00 0.05 0.10 0.15 0.20 0.25 nan−tw luo sw rw uz zh−CN cv lij ky kab eu ab ug mn or c… view at source ↗

Figures from the paper (2 more)

**Figure 4.** Figure 4: The proportion of utterances with a similarity score under 0.354 across languages. 0.0 0.2 0.4 0.6 or cs nan−tw ky kab ab zh−CN cv luo rw lij fr sw vi ja bas sr cyml eu zza gn kmr ka sk mn lggl ca el mk sah pa−INba dv dav tr bgid ta yoko sq sv−SE ug ckb mr pt pl kk uz …

**Figure 5.** Figure 5: The proportion of client IDs that have more than 10% of utterances with a similarity score under 0.354 across languages. 4. Discussion The goal of this paper was to quantify the speaker heterogeneity between recordings and reduce the heterogeneity within the client I…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

33 extracted references · 30 canonical work pages

[1]

Quantifying and Reducing Speaker Heterogeneity within the Common Voice Corpus for Phonetic Analysis

Introduction As a massively multilingual spoken corpus, the Mozilla Com- mon V oice Corpus has been pivotal in the development of novel and advanced speech technologies and holds tremendous po- tential for advancing research in crosslinguistic phonetics and speech sciences [1, 2]. Properly accounting for speaker vari- ation is, however, key to the theoret...

work page Pith review arXiv 2025
[2]

same speaker

Method 2.1. Data For this study, we selected 76 languages from the Mozilla Com- mon V oice Corpus. These languages were chosen based on their representation in the V oxCommunis Corpus [2, 16]. The V ox- Communis Corpus, derived from the Common V oice Corpus, includes word- and phone-level forced alignments to facilitate further phonetic analysis. It consi...

work page
[3]

same speaker

Results 3.1. Similarity scores The overall distribution of similarity across languages shows a strong left skew, indicating that most similarity scores are rea- sonably high (Figure 1). Across all languages, the inter-quartile range (IQR) of the similarity scores was 0.62 to 0.80 with a me- dian of 0.72. Across individual languages, the similarity score Q...

work page
[4]

To quantify heterogeneity, we used cosine similarity calculated from the ResNet-293 model pre-trained on the mul- tilingual V oxBlink2 dataset (Figure 1)

Discussion The goal of this paper was to quantify the speaker heterogene- ity between recordings and reduce the heterogeneity within the client IDs. To quantify heterogeneity, we used cosine similarity calculated from the ResNet-293 model pre-trained on the mul- tilingual V oxBlink2 dataset (Figure 1). To reduce heterogene- ity, as we did not have the gro...

work page
[5]

Our results indicate that a threshold of 0.354 effectively reduces speaker heterogeneity while minimizing data loss

Conclusion This study quantified and addressed speaker heterogeneity within the Mozilla Common V oice Corpus by using ResNet- based voice embeddings and an auditing procedure to establish a crosslinguistically reasonable heterogeneity threshold. Our results indicate that a threshold of 0.354 effectively reduces speaker heterogeneity while minimizing data ...

work page
[6]

Acknowledgements This research was supported by SNF Grant PR00P1 208460 to EC

work page
[7]

Common V oice: A massively-multilingual speech corpus,

R. Ardila, M. Branson, K. Davis, M. Kohler, J. Meyer, M. Henretty, R. Morais, L. Saunders, F. Tyers, and G. Weber, “Common V oice: A massively-multilingual speech corpus,” in Proceedings of the Twelfth Language Resources and Evaluation Conference, N. Calzolari, F. B ´echet, P. Blache, K. Choukri, C. Cieri, T. Declerck, S. Goggi, H. Isahara, B. Maegaard, J...

work page 2020
[8]

V oxCommunis: A corpus for cross- linguistic phonetic analysis,

E. Ahn and E. Chodroff, “V oxCommunis: A corpus for cross- linguistic phonetic analysis,” in Proceedings of the Thirteenth Language Resources and Evaluation Conference, 2022, pp. 5286– 5294

work page 2022

Show all 33 references

[9]

MLS: A large-scale multilingual dataset for speech research,

V . Pratap, Q. Xu, A. Sriram, G. Synnaeve, and R. Collobert, “MLS: A large-scale multilingual dataset for speech research,” in Interspeech 2020, 2020, pp. 2757–2761

2020
[10]

VoxPopuli: A large- scale multilingual speech corpus for representation learning, semi-supervised learning and interpretation,

C. Wang, M. Riviere, A. Lee, A. Wu, C. Talnikar, D. Haziza, M. Williamson, J. Pino, and E. Dupoux, “VoxPopuli: A large- scale multilingual speech corpus for representation learning, semi-supervised learning and interpretation,” in Proceedings of the 59th Annual Meeting of the ...

2021
[11]

V oxlingua107: A dataset for spoken lan- guage recognition,

J. Valk and T. Alum ¨ae, “V oxlingua107: A dataset for spoken lan- guage recognition,” in 2021 IEEE Spoken Language Technology Workshop (SLT), 2021, pp. 652–658

2021
[12]

Yodas: Youtube-oriented dataset for audio and speech,

X. Li, S. Takamichi, T. Saeki, W. Chen, S. Shiota, and S. Watan- abe, “Yodas: Youtube-oriented dataset for audio and speech,” in 2023 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU). IEEE, 2023, pp. 1–8

2023
[13]

The UCLA phonetics lab archive,

P. Ladefoged, B. Blankenship, R. G. Schuh, P. Jones, N. Gfroerer, E. Griffiths, L. Harrington, C. Hipp, P. Jones, M. M. Kaneko, C. Moore-Cantwell, G. Oh, K. Pfister, K. Vaughan, R. Videc, S. Weismuller, S. Weiss, J. White, S. Conlon, W. J. Lee, and R. Toribio, “The UCLA phonet...
[14]

Becoming a high-resource language in speech: The Catalan case in the Com- mon V oice corpus,

C. Armentano-Oller, M. Marimon, and M. Villegas, “Becoming a high-resource language in speech: The Catalan case in the Com- mon V oice corpus,” inProceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING...

2024
[15]

CMU Wilderness Multilingual Speech Dataset,

A. W. Black, “CMU Wilderness Multilingual Speech Dataset,” in ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2019, pp. 5971– 5975

2019
[16]

FLEURS: Few-shot learning evaluation of universal representations of speech,

A. Conneau, M. Ma, S. Khanuja, Y . Zhang, V . Axelrod, S. Dalmia, J. Riesa, C. Rivera, and A. Bapna, “FLEURS: Few-shot learning evaluation of universal representations of speech,” in 2022 IEEE Spoken Language Technology Workshop (SLT). IEEE, 2023, pp. 798–805

2022
[17]

Phonetic segmentation of the UCLA Phonetics Lab Archive,

E. Chodroff, B. Pa ˇzon, A. Baker, and S. Moran, “Phonetic segmentation of the UCLA Phonetics Lab Archive,” in Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024) , N. Calzolari, M.-Y . Kan, ...

2024
[18]

CommonBench: A larger scale speaker verification benchmark,

J. Hintz and I. Siegert, “CommonBench: A larger scale speaker verification benchmark,” in 4th Symposium on Security and Pri- vacy in Speech Communication, 2024, pp. 17–20

2024
[19]

Bengali Common V oice speech dataset for automatic speech recognition,

S. Alam, A. Sushmit, Z. Abdullah, S. Nakkhatra, M. Ansary, S. M. Hossen, S. M. Mehnaz, T. Reasat, and A. I. Humayun, “Bengali Common V oice speech dataset for automatic speech recognition,” arXiv preprint arXiv:2206.14053, 2022

2022 arXiv
[20]

Findings of the VarDial Evaluation Campaign 2022,

N. Aepli, A. Anastasopoulos, A.-G. Chifu, W. Domingues, F. Faisal, M. Gaman, R. T. Ionescu, and Y . Scherrer, “Findings of the VarDial Evaluation Campaign 2022,” in Proceedings of the Ninth Workshop on NLP for Similar Languages, Varieties and Dialects , Y . Scherrer, T. Jauhia...

2022
[21]

PyThaiNLP: Thai natural language processing in Python,

W. Phatthiyaphaibun, K. Chaovavanich, C. Polpanumas, A. Suriyawongkul, L. Lowphansirikul, and P. Chormai, “PyThaiNLP: Thai natural language processing in Python,” Jun. 2024. [Online]. Available: https://github.com/PyThaiNLP/ pythainlp/

2024
[22]

V oxblink2: A 100k+ speaker recognition corpus and the open- set speaker-identification benchmark,

Y . Lin, M. Cheng, F. Zhang, Y . Gao, S. Zhang, and M. Li, “V oxblink2: A 100k+ speaker recognition corpus and the open- set speaker-identification benchmark,” inInterspeech 2024, 2024, pp. 4263–4267

2024
[23]

Zhang, E

M. Zhang, E. Ahn, and E. Chodroff. pacscilab/V oxCommunis · datasets at hugging face. [Online]. Available: https://huggingface. co/datasets/pacscilab/V oxCommunis
[24]

V oxceleb: Large-scale speaker verification in the wild,

A. Nagrani, J. S. Chung, W. Xie, and A. Zisserman, “V oxceleb: Large-scale speaker verification in the wild,”Computer Speech & Language, vol. 60, p. 101027, 2020

2020
[25]

Analysis of deep generative model impact on feature extraction and dimension reduction for short ut- terance text-independent speaker verification,

A. Farhadipour and H. Veisi, “Analysis of deep generative model impact on feature extraction and dimension reduction for short ut- terance text-independent speaker verification,” Circuits, Systems, and Signal Processing, pp. 1–18, 2024

2024
[26]

PyCantonese: Cantonese linguistics and NLP in python,

J. Lee, L. Chen, C. Lam, C. M. Lau, and T.-H. Tsui, “PyCantonese: Cantonese linguistics and NLP in python,” in Proceedings of the Thirteenth Language Resources and Evaluation Conference , N. Calzolari, F. B ´echet, P. Blache, K. Choukri, C. Cieri, T. Declerck, S. Goggi, H. Isa...

2022
[27]

Pkuseg: A toolkit for multi-domain chinese word segmentation

R. Luo, J. Xu, Y . Zhang, Z. Zhang, X. Ren, and X. Sun, “Pkuseg: A toolkit for multi-domain chinese word segmentation.” CoRR, vol. abs/1906.11455, 2019. [Online]. Available: https: //arxiv.org/abs/1906.11455

1906 arXiv
[29]

fugashi, a tool for tokenizing Japanese in python,

P. McCann, “fugashi, a tool for tokenizing Japanese in python,” in Proceedings of Second Workshop for NLP Open Source Software (NLP-OSS). Online: Association for Computational Linguistics, Nov. 2020, pp. 44–51. [Online]. Available: https: //www.aclweb.org/anthology/2020.nlposs-1.7

2020
[30]

Introducing mul- tilingual phonetic information to speaker embedding for speaker verification,

Z. Song, L. He, P. Wang, Y . Hu, and H. Huang, “Introducing mul- tilingual phonetic information to speaker embedding for speaker verification,” in ICASSP 2024-2024 IEEE International Con- ference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2024, pp. 10 091–10 095

2024
[31]

Speak- erStew: Scaling to many languages with a triaged multilingual text-dependent and text-independent speaker verification system,

R. Chojnacka, J. Pelecanos, Q. Wang, and I. L. Moreno, “Speak- erStew: Scaling to many languages with a triaged multilingual text-dependent and text-independent speaker verification system,” in Interspeech 2021, 2021, pp. 1064–1068

2021
[32]

Language dependence in multi- lingual speaker verification,

N. T. Kleynhans and E. Barnard, “Language dependence in multi- lingual speaker verification,” in Proceedings of the Sixteenth An- nual Symposium of the Pattern Recognition Association of South Africa (PRASA), F. Nicolls, Ed. University of Cape Town, 2005

2005
[33]

Speaker recognition across languages,

T. K. Perrachione, “Speaker recognition across languages,” in The Oxford Handbook of Voice Perception, S. Fr ¨uhholz and P. Belin, Eds. Oxford University Press, 2017

2017
[2009]

Available: http://archive.phonetics.ucla.edu

[Online]. Available: http://archive.phonetics.ucla.edu

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

Quantifying and Reducing Speaker Heterogeneity within the Common Voice Corpus for Phonetic Analysis

Introduction As a massively multilingual spoken corpus, the Mozilla Com- mon V oice Corpus has been pivotal in the development of novel and advanced speech technologies and holds tremendous po- tential for advancing research in crosslinguistic phonetics and speech sciences [1, 2]. Properly accounting for speaker vari- ation is, however, key to the theoret...

work page Pith review arXiv 2025

[2] [2]

same speaker

Method 2.1. Data For this study, we selected 76 languages from the Mozilla Com- mon V oice Corpus. These languages were chosen based on their representation in the V oxCommunis Corpus [2, 16]. The V ox- Communis Corpus, derived from the Common V oice Corpus, includes word- and phone-level forced alignments to facilitate further phonetic analysis. It consi...

work page

[3] [3]

same speaker

Results 3.1. Similarity scores The overall distribution of similarity across languages shows a strong left skew, indicating that most similarity scores are rea- sonably high (Figure 1). Across all languages, the inter-quartile range (IQR) of the similarity scores was 0.62 to 0.80 with a me- dian of 0.72. Across individual languages, the similarity score Q...

work page

[4] [4]

To quantify heterogeneity, we used cosine similarity calculated from the ResNet-293 model pre-trained on the mul- tilingual V oxBlink2 dataset (Figure 1)

Discussion The goal of this paper was to quantify the speaker heterogene- ity between recordings and reduce the heterogeneity within the client IDs. To quantify heterogeneity, we used cosine similarity calculated from the ResNet-293 model pre-trained on the mul- tilingual V oxBlink2 dataset (Figure 1). To reduce heterogene- ity, as we did not have the gro...

work page

[5] [5]

Our results indicate that a threshold of 0.354 effectively reduces speaker heterogeneity while minimizing data loss

Conclusion This study quantified and addressed speaker heterogeneity within the Mozilla Common V oice Corpus by using ResNet- based voice embeddings and an auditing procedure to establish a crosslinguistically reasonable heterogeneity threshold. Our results indicate that a threshold of 0.354 effectively reduces speaker heterogeneity while minimizing data ...

work page

[6] [6]

Acknowledgements This research was supported by SNF Grant PR00P1 208460 to EC

work page

[7] [7]

Common V oice: A massively-multilingual speech corpus,

R. Ardila, M. Branson, K. Davis, M. Kohler, J. Meyer, M. Henretty, R. Morais, L. Saunders, F. Tyers, and G. Weber, “Common V oice: A massively-multilingual speech corpus,” in Proceedings of the Twelfth Language Resources and Evaluation Conference, N. Calzolari, F. B ´echet, P. Blache, K. Choukri, C. Cieri, T. Declerck, S. Goggi, H. Isahara, B. Maegaard, J...

work page 2020

[8] [8]

V oxCommunis: A corpus for cross- linguistic phonetic analysis,

E. Ahn and E. Chodroff, “V oxCommunis: A corpus for cross- linguistic phonetic analysis,” in Proceedings of the Thirteenth Language Resources and Evaluation Conference, 2022, pp. 5286– 5294

work page 2022