The reviewed record of science sign in
Pith

arxiv: 2607.05276 · v1 · pith:SIFHR6QF · submitted 2026-07-06 · eess.AS · cs.AI

ProPS: Prompted Profile Synthesis for Natural Language-Conditioned Speaker Embedding Distributions

Thomas Thebaud , Junhyeok Lee , Laureano Moro-Velazquez , Jesus Villalba Lopez , Najim Dehak This is my paper

Reviewed by Pith2026-07-07 20:07 UTCglm-5.2pith:SIFHR6QFopen to challenge →

classification eess.AS cs.AI
keywords speakerdistributionsembeddingsprofilepropsx-vectorssynthesisaccent
0
0 comments X

The pith

Text prompts steer speaker-embedding distributions with 98% gender fidelity

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

ProPS (Prompted Profile Synthesis) maps natural-language speaker descriptions — such as 'a thirties male speaker with an Indian accent' — to full Gaussian mixture distributions over x-vectors (neural speaker embeddings). The pipeline encodes a prompt with Sentence-BERT, passes it through a three-layer MLP, and outputs mixture weights over a bank of 942×16 pre-fitted Gaussian components initialised from real speaker-profile data. A three-stage training procedure (profile-GMM initialisation, MLP pre-training on paraphrased descriptions, end-to-end fine-tuning on human-written descriptions with negative-log-likelihood loss) teaches the network to associate free-form text with the correct region of speaker space. The core claim is that natural language can serve as a direct control interface for generating diverse, profile-matching speaker embeddings without reference audio. Empirically, generated x-vectors preserve gender (98.4% classifier accuracy), accent (91.6%), and age (65.4%) at near-real-data levels, but pitch (28.2%), pace (31.6%), and tone (23.8%) fall substantially short, suggesting x-vectors encode demographic identity more faithfully than prosodic style.

Core claim

The central mechanism is a prompt-conditioned mixture density network whose Gaussian component bank is initialised from 942 profile-level GMMs (16 components each) fitted on real x-vectors from the CapSpeech dataset. Text conditioning operates entirely through learned mixture-weight selection over this fixed bank — the means and variances are trainable but structurally anchored to pre-observed speaker profiles. The paper demonstrates that this architecture suffices to reproduce demographic and accent characteristics of held-out speakers, with generated x-vectors achieving 98.4% gender accuracy and 91.6% accent accuracy when evaluated by downstream classifiers, but that prosodic attributes (3

What carries the argument

Prompt-conditioned mixture density network (MDN) with a fixed bank of 942×16 diagonal-covariance Gaussian components initialised from profile-level GMMs; Sentence-BERT text encoder; ECAPA-TDNN x-vector extractor (D=192); three-stage training (GMM initialisation → MLP pre-training on GPT-generated descriptions → end-to-end NLL fine-tuning on human descriptions)

If this is right

  • ProPS-generated x-vectors could serve as conditioning inputs for zero-shot TTS or voice conversion systems, letting producers specify desired speaker profiles in plain English rather than providing reference audio.
  • The sharp gap between demographic accuracy (gender 98.4%, accent 91.6%) and prosodic accuracy (pitch 28.2%, pace 31.6%, tone 23.8%) indicates that current x-vector representations encode who a speaker is far better than how they speak, bounding what text-conditioned generation can achieve without richer embeddings.
  • The component-bank architecture means coverage is limited to the convex hull of 942 training profiles; profiles underrepresented in CapSpeech (e.g., rare accent combinations) would receive degraded or interpolated distributions.

Load-bearing premise

The architecture assumes that a fixed bank of Gaussian components fitted to 942 training profiles provides a sufficient basis for representing arbitrary text-conditioned speaker distributions — the model can only produce mixtures of pre-fitted profile GMMs and cannot generate distributional structure outside the convex hull of what was observed in training.

What would settle it

If downstream attribute classifiers trained on real x-vectors cannot reliably distinguish requested attributes in generated samples — which is already the case for pitch (28.2%), pace (31.6%), and tone (23.8%) — the claim that ProPS preserves prosodic characteristics fails, and the method's utility narrows to demographic conditioning only.

Figures

Figures reproduced from arXiv: 2607.05276 by Jesus Villalba Lopez, Junhyeok Lee, Laureano Moro-Velazquez, Najim Dehak, Thomas Thebaud.

Figure 1
Figure 1. Figure 1: Overview of the training pipeline of ProPS. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Negative log-likelihood with the dev and test sets as a function of [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: LDA visualization of generated and real x-vector distributions for [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: LDA visualization of generated and real x-vector distributions for the [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
read the original abstract

Speaker embeddings, or x-vectors, are widely used to represent speaker identity and speaker-related attributes, but existing embedding extractors are typically descriptive rather than generative: they map an observed speech segment to an x-vector, which is then used for downstream applications. We introduce ProPS, Prompted Profile Synthesis, a framework for generating distributions of speaker embeddings conditioned on natural language prompts such as "a thirties male speaker with an Indian accent". ProPS converts human-written profile descriptions into sentence embeddings and uses a mixture density network trained on a large-scale dataset to predict a Gaussian mixture model in the x-vector space. The model is trained by maximizing the likelihood that real speaker embeddings match the requested profile, and its generated distributions are evaluated by negative log-likelihood on held-out x-vectors and by attribute classification accuracies on sampled synthetic x-vectors. Experiments show that ProPS produces profile-conditioned distributions and generates x-vectors that preserve requested speaker attributes such as age, gender, accent, and prosodic characteristics. This design enables controllable speaker-profile synthesis for speech generation systems like Text-To-Speech (TTS) or Voice Conversion (VC) while anchoring generated distributions in observed speaker-embedding structure.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 8 minor

Summary. The paper introduces ProPS, a framework for generating speaker-embedding (x-vector) distributions conditioned on natural language prompts. The method uses SBERT sentence embeddings mapped through an MLP to predict mixture weights over a bank of N×K Gaussian components initialized from profile-level GMMs, trained on the CapSpeech dataset. The model is trained in three stages (GMM initialization, pretraining on GPT-generated descriptions, fine-tuning on human descriptions) and evaluated via negative log-likelihood and attribute classification accuracy on sampled x-vectors. Results show strong preservation of gender (98.4%) and accent (91.6%), moderate age preservation (65.4%), and weak preservation of prosodic attributes (pitch 28.2%, pace 31.6%, tone 23.8%).

Significance. The paper addresses a genuine gap: controllable generation of speaker-embedding distributions from free-form text, as opposed to sampling from unconditional priors. The three-stage training design is reasonable, and the use of a fixed component bank initialized from real x-vector statistics grounds the generated distributions in observed data. The authors provide reproducible code (anonymous repository) and report an ablation study across training stages (Table I) and a sweep over the number of Gaussian components (Fig. 2). The framework produces falsifiable predictions via downstream attribute classifiers. The contribution is incremental but solid relative to PromptSpeaker, with the key distinction being direct GMM modeling in x-vector space rather than sampling through an invertible flow.

major comments (3)
  1. Abstract, final sentence: the claim that ProPS generates x-vectors that 'preserve requested speaker attributes such as age, gender, accent, and prosodic characteristics' is not supported by Table II for prosodic characteristics. Pitch (28.2%, 7-class), pace (31.6%, 5-class), and tone (23.8%, 5-class) are near or below chance-level baselines. The abstract should be revised to scope the claim to gender, accent, and age, with prosodic characteristics acknowledged as not reliably preserved. The conclusion already states this more carefully, so the abstract should be aligned with the conclusion's own assessment.
  2. Table II / §IV.D: the evaluation of attribute preservation lacks an unconditional or random-prompt baseline. The component bank is initialized from 942 profile-level GMMs, and if the training data is imbalanced across attribute values (e.g., more American-accent profiles than Cantonese), the high gender (98.4%) and accent (91.6%) accuracies could partly reflect the marginal distribution of attributes in the component bank rather than prompt-based control. An unconditional baseline — sampling from uniform mixture weights over all components, or from the average predicted mixture weights across all prompts — would establish the floor and confirm that the text prompt contributes beyond marginal attribute frequency. Without this, H1 (prompt controls attribute region) and H2 (component bank imbalance drives accuracy) cannot be distinguished. This is load-bearing for the central claim of cont滚
  3. §III.C, Eq. (4) and §III.D.1: the architecture defines the conditional density as a mixture over N×K fixed Gaussian components with trainable means and variances. The text conditioning enters only through mixture weights π(e). This means the model can only produce distributions within the convex hull of the pre-fitted profile GMMs — it cannot generate distributional structure outside the training profiles. The paper should explicitly acknowledge this limitation (e.g., in §VI) and discuss its implications for generalization to unseen profile combinations. Currently, §VI mentions dataset coverage limitations but does not address this architectural constraint.
minor comments (8)
  1. §III.C, Eq. (4): the notation switches between k (Eq. 2) and j (Eq. 4) for component indexing; using a consistent index would help.
  2. §III.B.2: the attribute list for pace includes 6 values (slowly through very fast) but Table II reports 5 classes for pace. Similarly, tone lists 5 values and Table II reports 5 classes, which is consistent, but the pace discrepancy should be clarified.
  3. §III.D.2: the pretraining stage uses GPT-generated descriptions 3–10 for training and description 2 for validation, but descriptions 0 and 1 are reserved for test. It is unclear whether any label smoothing or regularization is applied to the target mixture weights during pretraining; a brief mention would help.
  4. Table I: the NLL values are very close across models (e.g., 303.8 vs 305.3 for test GPT-1). Adding the standard deviation or confidence interval for each entry would aid the reader in interpreting the significance markers.
  5. §V.D: the observation that generated age accuracy (65.4%) exceeds real test accuracy (59.8%) is attributed to 'cleaner or more prototypical age distributions.' An alternative explanation is that the SVR classifier, trained on real data, may be biased toward the central tendency of the age distribution, which the generated samples cluster around. This should be discussed.
  6. Fig. 2: the x-axis label and the relationship between K and the number of valid profiles could be clearer; a table companion showing exact profile counts per K value would help.
  7. §III.B.3: the GPT model identifier 'gpt-5.4-mini-2026-03-17' does not correspond to any known model release; this should be corrected or clarified.
  8. The HuggingFace link footnote (§I, contribution 2) states it will be provided after anonymization; this should be resolved before camera-ready.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for a careful and constructive review. The comments are substantive and we address each below. We agree with two of the three major comments and will revise the manuscript accordingly; for the third (convex hull limitation) we agree the point is architecturally correct and will add an explicit discussion, though we note this is a deliberate design trade-off rather than a defect.

read point-by-point responses

  1. Referee: Abstract claim about prosodic characteristics is unsupported by Table II (pitch 28.2%, pace 31.6%, tone 23.8% near or below chance). Abstract should be scoped to gender, accent, and age, with prosodic characteristics acknowledged as not reliably preserved.

    Authors: The referee is correct. The abstract's final sentence claims preservation of 'age, gender, accent, and prosodic characteristics,' but Table II shows that prosodic attribute accuracies are at or near chance: pitch 28.2% (7-class, chance ~14.3%), pace 31.6% (5-class, chance ~20%), and tone 23.8% (5-class, chance ~20%). While these are technically above uniform-random chance, they are not meaningfully above the marginal-class-frequency baselines one would expect from imbalanced data, and the conclusion section already acknowledges this honestly. We will revise the abstract to scope the claim to gender, accent, and age, and explicitly note that prosodic characteristics are not reliably preserved. This aligns the abstract with the conclusion's own assessment. revision: yes

  2. Referee: Table II / §IV.D lacks an unconditional or random-prompt baseline. High gender and accent accuracies could reflect marginal attribute distribution in the component bank rather than prompt-based control. An unconditional baseline would distinguish H1 (prompt controls attribute region) from H2 (component bank imbalance drives accuracy).

    Authors: This is a fair and important point. We acknowledge that without an unconditional baseline, the current evaluation cannot fully distinguish between prompt-driven control (H1) and component-bank imbalance (H2). We will address this in revision by adding an unconditional baseline: sampling from uniform mixture weights over all N×K components. This establishes the floor against which prompt-conditioned accuracy can be compared. If the prompt-conditioned accuracies substantially exceed the unconditional baseline, this supports H1; if they are similar, H2 would be implicated. We will run this experiment and add the results to Table II (or a supplementary table). We note that the LDA visualizations in Figures 3–4 already provide qualitative evidence that prompt-conditioned sampling produces distributions matching the requested attribute rather than the marginal distribution — for example, the male and female generated distributions in Figure 3 are clearly separated and aligned with the real distributions — but we agree that a quantitative unconditional baseline is needed and will be added. revision: yes

  3. Referee: §III.C, Eq. (4): the architecture can only produce distributions within the convex hull of pre-fitted profile GMMs. The paper should explicitly acknowledge this limitation in §VI and discuss implications for generalization to unseen profile combinations.

    Authors: The referee's architectural observation is correct. Because the text conditioning enters only through mixture weights π(e) over fixed components, and the component means and variances are initialized from the N profile-level GMMs and then fine-tuned, the model's output distributions are constrained to reweight and adjust components that originate from observed profiles. The model cannot generate distributional structure entirely outside the span of the training profiles' x-vector statistics. We will add an explicit discussion of this limitation in §VI, noting that: (1) the convex-hull constraint is a deliberate design choice that grounds generated distributions in real x-vector statistics, avoiding the risk of generating implausible embeddings; (2) it limits generalization to unseen attribute combinations not represented in the training profiles; and (3) future work could explore continuous parameter prediction (e.g., predicting means and variances directly from text embeddings rather than only mixture weights) to relax this constraint, at the cost of potentially generating less grounded distributions. We note that the current §VI mentions dataset coverage limitations but does not address this architectural constraint, and we will add it. revision: yes

Circularity Check

1 steps flagged

Pre-training explicitly trains the MLP to retrieve attribute-grouped GMMs; Table II then evaluates whether samples from those GMMs preserve the same attributes — partially circular for high-accuracy traits, but not fully forced due to independent classifiers, fine-tuning, and generalization to minimal prompts.

specific steps
  1. fitted input called prediction [§III.D.2 (pre-training) combined with §IV.D and Table II (evaluation)]
    "The target distribution is constructed from the precomputed GMM of the corresponding profile: the K components associated with that profile receive their precomputed mixture weights, while all other components receive zero weight. [...] For each characteristic, such as gender, age, accent, pitch, pace, or tone, we construct minimal natural language prompts that specify only that characteristic. [...] We then sample 10,000 synthetic x-vectors from each generated GMM and evaluate the same classifier on these generated samples."

    The component bank is initialized from profile-level GMMs fit on x-vectors grouped by profile attributes (§III.D.1). The pre-training stage (§III.D.2) explicitly trains the MLP to select the correct profile GMMs given a text description — the target is literally the precomputed mixture weights for the matching profile. Then Table II evaluates whether x-vectors sampled from these GMMs preserve the requested attributes. For strongly x-vector-encoded attributes (gender 98.4%, accent 91.6%), the high accuracy is partly forced: the GMMs were fit on x-vectors from speakers with those exact attributes, and the MLP was trained to retrieve them. However, this is only partial circularity: (1) the SVM/SVR classifiers are independently trained on real dev-set x-vectors, not on GMM components; (2) the

full rationale

The paper has one partially circular evaluation step: the pre-training stage trains the MLP to retrieve attribute-grouped GMMs, and Table II then checks whether samples from those GMMs have the correct attributes. For gender and accent (high x-vector encoding, high accuracy), this chain is partly forced by construction. However, the circularity is only partial because: (a) the SVM/SVR classifiers are independent models trained on real x-vectors, not on the GMM outputs; (b) the fine-tuning stage (§III.D.3) updates GMM parameters end-to-end with NLL loss, so the final distributions differ from initialization; (c) evaluation uses minimal single-attribute prompts (e.g., 'a male speaker') that differ from the multi-attribute GPT descriptions used in pre-training, requiring generalization; (d) prosodic attributes (pitch 28.2%, pace 31.6%, tone 23.8%) show low accuracy, demonstrating the results are not fully determined by the training setup. Self-citations ([6], [7], [8]) involve overlapping authors but are dataset/application citations, not load-bearing theoretical claims. The central methodological contribution (text-conditioned MDN over a component bank) and the NLL evaluation (Table I) are independent of the partially circular attribute-accuracy evaluation. Score 4 reflects one partially circular 'prediction' that reduces partly to its training setup, while the central claim retains independent content.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 0 invented entities

No new entities are invented. The framework uses existing components (SBERT, ECAPA-TDNN, MDN, GMMs) in a new combination.

free parameters (4)
  • K (Gaussian components per profile)
    Selected as K=16 based on Figure 2 sweep; controls model capacity. Not fitted to the target result but chosen by hyperparameter search.
  • MLP hidden dimensions (1024, 2048, 1024)
    Architecture choice not justified by any derivation or ablation.
  • Learning rates (1e-4 pretraining, 1e-5 fine-tuning)
    Standard hyperparameters, not derived.
  • N=942 profiles
    Determined by the 100-segment threshold on CapSpeech data; affects coverage but not fitted to results.
axioms (4)
  • domain assumption SBERT embeddings of profile descriptions contain sufficient signal to distinguish among 942 speaker profiles
    Invoked in §III.C where SBERT embeddings are fed to the MLP to predict mixture weights. The NLL results (Table I) provide partial support but the gap between oracle GMMs and MDN suggests information loss.
  • domain assumption ECAPA-TDNN x-vectors encode demographic and prosodic attributes in a way that can be conditioned on
    Invoked throughout; Table II shows this holds for gender and accent but fails for tone (23.8%, near chance for 5 classes) and is weak for pitch (28.2%) and pace (31.6%).
  • domain assumption Diagonal covariance GMMs adequately model the x-vector distribution within each profile
    Invoked in §III.C and §IV.B; no comparison with full-covariance or other distributional forms is provided.
  • domain assumption GPT-generated descriptions are semantically equivalent to human-written descriptions for the purpose of pretraining
    Invoked in §III.D.2; Table I shows a NLL gap between GPT and human descriptions, suggesting they are not fully equivalent.

pith-pipeline@v1.1.0-glm · 12642 in / 4264 out tokens · 72672 ms · 2026-07-07T20:07:28.082106+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

27 extracted references · 27 canonical work pages · 6 internal anchors

  1. [1]

    X- vectors: Robust dnn embeddings for speaker recognition,

    D. Snyder, D. Garcia-Romero, G. Sell, D. Povey, and S. Khudanpur, “X- vectors: Robust dnn embeddings for speaker recognition,” in2018 IEEE international conference on acoustics, speech and signal processing (ICASSP). IEEE, 2018, pp. 5329–5333

    work page 2018

  2. [2]

    Vbx for end-to-end neural and clustering-based diarization,

    P. P ´alka, J. Han, M. Delcroix, N. Tawara, and L. Burget, “Vbx for end-to-end neural and clustering-based diarization,” inICASSP 2026- 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2026, pp. 17 472–17 476

    work page 2026

  3. [3]

    Autovc: Zero-shot voice style transfer with only autoencoder loss,

    K. Qian, Y . Zhang, S. Chang, X. Yang, and M. Hasegawa-Johnson, “Autovc: Zero-shot voice style transfer with only autoencoder loss,” in International Conference on Machine Learning. PMLR, 2019, pp. 5210–5219

    work page 2019

  4. [4]

    Privacy and utility of x-vector based speaker anonymization,

    B. M. L. Srivastava, M. Maouche, M. Sahidullah, E. Vincent, A. Bellet, M. Tommasi, N. Tomashenko, X. Wang, and J. Yamagishi, “Privacy and utility of x-vector based speaker anonymization,”IEEE/ACM Transac- tions on Audio, Speech, and Language Processing, vol. 30, pp. 2383– 2395, 2022

    work page 2022

  5. [5]

    Zero-shot multi-speaker text-to-speech with state-of-the-art neural speaker embeddings,

    E. Cooper, C.-I. Lai, Y . Yasuda, F. Fang, X. Wang, N. Chen, and J. Ya- magishi, “Zero-shot multi-speaker text-to-speech with state-of-the-art neural speaker embeddings,” inICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2020, pp. 6184–6188

    work page 2020

  6. [6]

    Demographic attributes prediction from speech using wavlm embeddings,

    Y . Yang, T. Thebaud, and N. Dehak, “Demographic attributes prediction from speech using wavlm embeddings,” in2025 59th Annual Conference on Information Sciences and Systems (CISS). IEEE, 2025, pp. 1–6

    work page 2025

  7. [7]

    Interpretable speech features vs. dnn embeddings: What to use in the automatic assessment of parkinson’s disease in multi- lingual scenarios,

    A. Favaro, Y .-T. Tsai, A. Butala, T. Thebaud, J. Villalba, N. Dehak, and L. Moro-Vel´azquez, “Interpretable speech features vs. dnn embeddings: What to use in the automatic assessment of parkinson’s disease in multi- lingual scenarios,”Computers in Biology and Medicine, vol. 166, p. 107559, 2023

    work page 2023

  8. [8]

    Capspeech: En- abling downstream applications in style-captioned text-to-speech,

    H. Wang, J. Hai, D. Chong, K. Thakkar, T. Feng, D. Yang, J. Lee, T. Thebaud, L. M. Velazquez, J. Villalbaet al., “Capspeech: En- abling downstream applications in style-captioned text-to-speech,”arXiv preprint arXiv:2506.02863, 2025

    work page arXiv 2025

  9. [9]

    Front- end factor analysis for speaker verification,

    N. Dehak, P. J. Kenny, R. Dehak, P. Dumouchel, and P. Ouellet, “Front- end factor analysis for speaker verification,”IEEE Transactions on Audio, Speech, and Language Processing, vol. 19, no. 4, pp. 788–798, 2010

    work page 2010

  10. [10]

    ECAPA-TDNN: Emphasized Channel Attention, Propagation and Aggregation in TDNN Based Speaker Verification

    B. Desplanques, J. Thienpondt, and K. Demuynck, “Ecapa-tdnn: Em- phasized channel attention, propagation and aggregation in tdnn based speaker verification,”arXiv preprint arXiv:2005.07143, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2005

  11. [11]

    Mixture density networks,

    C. M. Bishop, “Mixture density networks,” 1994

    work page 1994

  12. [12]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,”Advances in neural information processing systems, vol. 30, 2017

    work page 2017

  13. [13]

    Bert: Pre-training of deep bidirectional transformers for language understanding,

    J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “Bert: Pre-training of deep bidirectional transformers for language understanding,” inPro- ceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies, volume 1 (long and short papers), 2019, pp. 4171–4186

    work page 2019

  14. [14]

    Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks

    N. Reimers and I. Gurevych, “Sentence-bert: Sentence embeddings using siamese bert-networks,” inProceedings of the 2019 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics, 11 2019. [Online]. Available: https: //arxiv.org/abs/1908.10084

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  15. [15]

    Promptspeaker: Speaker generation based on text descriptions,

    Y . Zhang, G. Liu, Y . Lei, Y . Chen, H. Yin, L. Xie, and Z. Li, “Promptspeaker: Speaker generation based on text descriptions,” in 2023 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU). IEEE, 2023, pp. 1–7

    work page 2023

  16. [16]

    Glow: Generative flow with invertible 1x1 convolutions,

    D. P. Kingma and P. Dhariwal, “Glow: Generative flow with invertible 1x1 convolutions,”Advances in neural information processing systems, vol. 31, 2018

    work page 2018

  17. [17]

    GigaSpeech: An Evolving, Multi-domain ASR Corpus with 10,000 Hours of Transcribed Audio

    G. Chen, S. Chai, G. Wang, J. Du, W.-Q. Zhang, C. Weng, D. Su, D. Povey, J. Trmal, J. Zhanget al., “Gigaspeech: An evolving, multi- domain asr corpus with 10,000 hours of transcribed audio,”arXiv preprint arXiv:2106.06909, 2021

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  18. [18]

    Common Voice: A Massively-Multilingual Speech Corpus

    R. Ardila, M. Branson, K. Davis, M. Henretty, M. Kohler, J. Meyer, R. Morais, L. Saunders, F. M. Tyers, and G. Weber, “Com- mon voice: A massively-multilingual speech corpus,”arXiv preprint arXiv:1912.06670, 2019

    work page internal anchor Pith review Pith/arXiv arXiv 1912

  19. [19]

    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,”arXiv preprint arXiv:2012.03411, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2012

  20. [20]

    Emilia: An extensive, multilingual, and diverse speech dataset for large-scale speech generation,

    H. He, Z. Shang, C. Wang, X. Li, Y . Gu, H. Hua, L. Liu, C. Yang, J. Li, P. Shiet al., “Emilia: An extensive, multilingual, and diverse speech dataset for large-scale speech generation,” in2024 IEEE Spoken Language Technology Workshop (SLT). IEEE, 2024, pp. 885–890

    work page 2024

  21. [21]

    Gpt-4 technical report,

    OpenAI, J. Achiam, S. Adler, S. Agarwal, L. Ahmadet al., “Gpt-4 technical report,” 2024. [Online]. Available: https://arxiv.org/abs/2303. 08774

    work page 2024

  22. [22]

    SpeechBrain: A General-Purpose Speech Toolkit

    M. Ravanelli, T. Parcollet, P. Plantinga, A. Rouhe, S. Cornell, L. Lu- gosch, C. Subakan, N. Dawalatabad, A. Heba, J. Zhong, J.-C. Chou, S.-L. Yeh, S.-W. Fu, C.-F. Liao, E. Rastorgueva, F. Grondin, W. Aris, H. Na, Y . Gao, R. D. Mori, and Y . Bengio, “SpeechBrain: A general- purpose speech toolkit,” 2021, arXiv:2106.04624

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  23. [23]

    Scikit-learn,

    O. Kramer, “Scikit-learn,” inMachine learning for evolution strategies. Springer, 2016, pp. 45–53

    work page 2016

  24. [24]

    Scipy 1.0: fundamental algorithms for scientific computing in python,

    P. Virtanen, R. Gommers, T. E. Oliphant, M. Haberland, T. Reddy, D. Cournapeau, E. Burovski, P. Peterson, W. Weckesser, J. Brightet al., “Scipy 1.0: fundamental algorithms for scientific computing in python,” Nature methods, vol. 17, no. 3, pp. 261–272, 2020

    work page 2020

  25. [25]

    A study of computation speed-ups of the gmm-ubm speaker recognition system

    J. McLaughlin, D. A. Reynolds, and T. P. Gleason, “A study of computation speed-ups of the gmm-ubm speaker recognition system.” inEUROSPEECH, vol. 99, 1999, pp. 1215–1218

    work page 1999

  26. [26]

    Text-independent speaker identification using gmm-ubm and frame level likelihood normalization,

    R. Zheng, S. Zhang, and B. Xu, “Text-independent speaker identification using gmm-ubm and frame level likelihood normalization,” in2004 International Symposium on Chinese Spoken Language Processing. IEEE, 2004, pp. 289–292

    work page 2004

  27. [27]

    Com- parison of i-vector and gmm-ubm approaches to speaker identification with timit and nist 2008 databases in challenging environments,

    M. T. Al-Kaltakchi, W. L. Woo, S. S. Dlay, and J. A. Chambers, “Com- parison of i-vector and gmm-ubm approaches to speaker identification with timit and nist 2008 databases in challenging environments,” in2017 25th European signal processing conference (EUSIPCO). IEEE, 2017, pp. 533–537

    work page 2008