Staged Depth-Pruning Distillation of a Flow-Matching Text-to-Speech Teacher: A Compact Hindi Speech Synthesizer

Sivateja Trikutam

REVIEW 4 major objections 4 minor 17 references

Depth-only pruning of a 337M flow-matching TTS teacher produces a 190M Hindi voice that keeps about 96% of the teacher's predicted naturalness and speaker similarity, runs in real time on a 6 GB laptop GPU, and generalizes to unseen sentenc

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 18:43 UTC pith:SREWJI3N

load-bearing objection Depth-only warm-starting is a credible recipe for compact TTS, but the paper's real-time claim is contradicted by its own RTF table and the headline WER rests on a one-sentence demo. the 4 major comments →

arxiv 2607.18662 v1 pith:SREWJI3N submitted 2026-07-19 cs.SD cs.CLeess.AS

Staged Depth-Pruning Distillation of a Flow-Matching Text-to-Speech Teacher: A Compact Hindi Speech Synthesizer

Sivateja Trikutam This is my paper

classification cs.SD cs.CLeess.AS

keywords text-to-speechflow matchingknowledge distillationdepth pruningHindicompact modelsword error rate gatingon-device TTS

verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper tries to establish that a compact Hindi text-to-speech model can be made from a large flow-matching teacher without training from scratch, even when only about 17.6 hours of data are available. The key move is warm-starting a smaller student by removing whole transformer blocks (depth-only pruning) while keeping all widths, embeddings, and input/output heads identical, then descending gradually through block counts with re-fine-tuning and an automated intelligibility gate after each step. The claim is that this staged distillation makes a 190M-parameter student reach roughly 96% of the teacher's predicted naturalness and speaker similarity at 56% of its parameters, beat a small existing Hindi TTS on intelligibility and naturalness, and synthesize 1.8 times faster. A reader should care because the middle ground between tiny low-quality models and huge cloud models is precisely where practical on-device Hindi TTS lives, and this paper offers a reproducible path there.

Core claim

On a teacher-generated corpus of roughly 17.6 hours, training a small flow-matching TTS model from random initialization produces unintelligible speech, while warm-starting the student from a 337M-parameter teacher by retaining an evenly spaced subset of transformer blocks — with all non-block tensors copied one-to-one — makes the small model begin learning within one epoch and generalize to fresh sentences. The paper shows the teacher tolerates about a quarter of its blocks being removed (near-teacher output amplitude) but collapses past half, so it descends gradually 22 -> 16 -> 12 -> 8 -> 6 blocks, re-fine-tuning after each prune and gating each step on an ASR word-error-rate check. The r

What carries the argument

The depth-only warm-start: the student uses exactly the teacher's hidden width, text dimension, attention heads, and mel/text input-output, so every non-block tensor copies one-to-one; only the number of transformer blocks L is reduced by selecting an evenly spaced subset of the teacher's 22 blocks with endpoints kept. The staged ladder (22 -> 16 -> 12 -> 8 -> 6) then prunes from the previous fine-tuned student and re-fine-tunes after each step, with each stage gated by an automated word-error-rate measurement. The pruning-tolerance analysis (output amplitude as a proxy for whether the residual vector field still integrates to speech) motivates gradual descent instead of a single large prune

Load-bearing premise

The paper's evidence that the student produces clear, generalizing Hindi speech rests on ASR word-error rate measured on very few sentences with a single ASR judge and no human listening, so an unrepresentative sentence set or judge would undermine the central claim.

What would settle it

Run the released 190M student and its 337M teacher on a held-out Hindi test set of at least 200 sentences from a different corpus, transcribe with two independent ASR engines, and collect human MOS ratings; the central claim fails if the student's WER is not within a small margin of the teacher's or if human listeners rate the student's speech clearly less natural or less intelligible.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

If the recipe holds, a deployable 190M Hindi TTS that runs in real time on a laptop GPU can be built from only 17.6 hours of teacher-generated data, and the same staged pruning may produce compact voices for other languages covered by a strong flow-matching teacher.
The depth-only warm-start makes distillation cheap: each stage recovers in thousands rather than millions of steps, so the entire fine-tuning budget stays feasible in a data-limited regime.
The documented train/inference parity bugs (mel filterbank and rotary-embedding library versions) imply that silent quality loss in such systems can be traced to version mismatches; freezing feature extraction and pinning the libraries that define the model math are part of the recipe.
The capacity cliff at 102M parameters suggests the smallest viable model size is bounded by data rather than by the pruning method, so enlarging the teacher corpus is the stated next step toward a ~100M model.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The staged-depth recipe likely transfers to any generative model whose output is defined by integrating a neural vector field (diffusion or rectified-flow models for audio, image, or video), because the fragility of the residual stream under large prunes is a property of the ODE integration, not of TTS specifically.
The WER gate is a cheap proxy; a natural extension is to gate each prune on a combined naturalness-and-similarity objective, which might better predict human preference and catch the 'speaks but muffled' failure mode the paper documents.
Because the student inherits the teacher's voice-cloning behavior (speaker similarity ~0.75), the 190M model could serve as a lightweight voice-cloning front end for personalized audiobooks or dubbing on edge devices, a use case the paper does not explicitly explore.
The parity-failure lesson suggests a testable workflow: verify checkpoint import parity by comparing a fixed input's intermediate activations across different load paths before launching long fine-tuning runs, which would catch silent random-initialization failures early.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

Depth-only warm-starting is a credible recipe for compact TTS, but the paper's real-time claim is contradicted by its own RTF table and the headline WER rests on a one-sentence demo.

read the letter

The depth-pruning recipe is the real contribution, and it is probably right. Warm-starting a flow-matching TTS student by keeping the teacher's width and copying all non-block tensors, then descending the block count with WER-gated re-fine-tuning, is a sensible and clearly explained engineering idea. The released 190M model with an independent 50-sentence FLEURS benchmark is genuine evidence, and the paper does two useful things that often get skipped: it documents silent train/inference parity bugs (mel filterbank, rotary embedding, torch.compile prefix) and points to concrete fixes, and it releases per-sentence artifacts.

The soft spots are mostly in the evaluation, and one is a direct internal contradiction. The abstract and conclusion claim the 190M model runs in real time on a 6 GB laptop GPU. Table 4 reports median RTF 3.13 for that model, measured on an Apple M4 with MPS. RTF > 1 is slower than real time, and the M4 is not a 6 GB discrete GPU. That claim needs a correction or a real benchmark on the stated hardware. It doesn't sink the recipe, but it is exactly what a referee should flag.

The smaller concerns the reader raised are real but proportionate. The headline WER 0.00 comes from one fresh sentence with two stochastic generations; that is a demo, not a result. Table 3's known-vs-fresh pattern—fresh WER lower than known WER at every stage—is odd and suggests the two sentence sets differ in difficulty; the paper does not explain it. The from-scratch failure is anecdotal, and the attribution of the 102M cliff to data budget rather than recipe is not tested—it is plausible, but the paper only shows that more training did not help. The 50-sentence benchmark has no confidence intervals and uses a single ASR judge, but that is a standard practical limitation, and the CER numbers support the qualitative conclusion.

The depth-only warm-start idea may transfer to other flow-matching TTS systems, and the citation pattern is fair: LayerDrop, Poor Man's BERT, and ShortGPT are appropriately credited as the base techniques. No circularity in the evaluation; the ASR judges are independent of the training loop.

Who is this for? Practitioners building compact TTS for low-resource languages, especially with small in-house data budgets. It is a solid, clearly written engineering paper with a reproducible recipe, but the evaluation falls short of the deployment claims as written. I would send it to peer review, expecting a major revision: fix the RTF claim, expand the benchmark to a few hundred sentences with variance reporting, and ideally add a small human listening test on a subset. The core idea deserves a serious referee.

Referee Report

4 major / 4 minor

Summary. The paper proposes a staged depth-pruning distillation recipe for building a compact Hindi TTS model under a ~17.6 h data budget. Starting from the 337M-parameter IndicF5 flow-matching teacher, the authors keep width, text dimension, attention heads, and input/output layers fixed, retain an evenly spaced subset of transformer blocks, and fine-tune after each prune (22->16->12->8->6 blocks), gating each stage by ASR word-error rate. They report that students at 249M, 190M, and 131M parameters produce intelligible speech, with a capacity cliff at 102M that they attribute to the data budget. An independent 50-sentence FLEURS benchmark compares the released 190M student with the teacher and MMS-TTS-hin, reporting WER/CER, speaker similarity, UTMOS, and RTF. The paper also documents two train/inference parity bugs (mel filterbank and rotary-embedding library versions). The headline deployment claim is that the 190M model runs in real time on a 6 GB laptop GPU.

Significance. If the results hold, this is a practical contribution: a data-efficient way to shrink a large flow-matching TTS teacher into a deployable student while preserving intelligibility and voice similarity, with released model weights, scripts, and per-sentence benchmark artifacts. The paper is honest about limitations and the failure modes it encountered. However, the main deployment claim is contradicted by the paper's own runtime measurement, and the quality evidence relies on very small samples and a single ASR judge. The recipe itself is plausible and the independent benchmark is a strength, but the claims as currently written outrun the evidence.

major comments (4)

[Abstract and §7 Conclusion vs. Table 4] The abstract and conclusion state that the 190M model 'runs in real time on a 6 GB laptop GPU.' The only runtime measurement reported, Table 4, gives median RTF = 3.13 on an Apple M4 with the MPS backend, where RTF > 1 means synthesis is slower than real time. No benchmark on a 6 GB laptop GPU is provided, and the Apple M4's unified memory is not a 6 GB discrete GPU. This is a direct internal inconsistency, not an evaluation-representativeness issue. The deployment claim must be corrected or supported by measurements on the stated hardware.
[Table 3 and §3.6] The headline 'WER 0.00 on unseen sentences' rests on at most one or two fresh sentences, and the Table 3 pattern is the opposite of what one would expect if the known and fresh sentences were of comparable difficulty: fresh-sentence WER is consistently lower than known-sentence WER (e.g., 249M: 0.00 vs. 0.18; 190M: 0.00-0.06 vs. 0.24-0.29). This strongly suggests the two sentence sets are not matched in difficulty or ASR-transcribability, making the generalization claim fragile. The independent 50-sentence benchmark partially mitigates this, but it uses a single ASR judge and no confidence intervals; the paper should report per-sentence results with uncertainty and ideally multiple judges or human listening.
[§4, Table 3, and Conclusion] The attribution of the 102M capacity cliff to the data budget rather than to the recipe is not established. The evidence is that the 102M model remains poor after 150 epochs, but there is no controlled experiment varying data quantity or model depth while holding other factors fixed. A depth-6 model trained on a larger corpus, or trained with a different optimizer/regularization, could behave differently. The claim 'the 102M target is reachable with a larger teacher corpus' is speculative and should be labeled as such.
[§1 and §3.4] The claim that training a small model from scratch is infeasible relies on a single anecdotal attempt (a 79M model that produced unintelligible speech). No systematic from-scratch baseline is reported at comparable parameter counts, depth/width configuration, or training budget. Since the paper's central contribution is that warm-starting is essential, the absence of a controlled from-scratch comparison weakens the contrast, even if the qualitative conclusion may be correct. A small controlled experiment or, failing that, a clear statement that this is anecdotal evidence would be appropriate.

minor comments (4)

[Abstract and Table 3] The abstract says students 'reach WER 0.00 on unseen sentences at 249M and 190M parameters,' but Table 3 reports the 190M fresh-sentence WER as 0.00-0.06. The wording should reflect the range and the stochastic nature of the generations.
[Table 4 footnote] The footnote says the Vakyansh ASR judge 'inflat[es] absolute WER equally for all systems.' ASR errors are not necessarily independent of system-specific artifacts; 'equally' is a strong assumption. Suggest rewording to 'may inflate absolute WER for all systems.'
[§3.4] The learning rate is written as '5×10 −5' with a stray space; format as 5×10^-5. Also define 'NFE' at first use in Table 2.
[§4.1] The 1.8x faster claim follows from 5.56/3.13, but both numbers are median RTF on the same Apple M4; consider stating this explicitly so readers do not infer a portable speed comparison.

Circularity Check

0 steps flagged

No circular derivation: the central pruning recipe is self-contained and evaluated by independent ASR judges against external references.

full rationale

The paper's derivation chain is not circular. The teacher (IndicF5) is an external model, and the student is obtained by depth-only weight copying plus fine-tuning on teacher-generated audio; no equation defines the student in terms of the evaluation metric. The pruning-tolerance analysis (Table 2) uses untrained RMS as a viability proxy, an empirical measurement not derived from the final WER claims. The per-stage ASR-WER gate is a selection heuristic, but it is not the paper's central evidence: §4.1 reports an independent 50-sentence FLEURS benchmark using a different ASR judge (Vakyansh) on identical inputs across systems, and the conclusions about quality retention rest on that external comparison plus UTMOS and ECAPA-TDNN, none of which are fitted outputs of the recipe. The limitations section also openly flags the use of ASR-WER and predicted MOS rather than human evaluation, which is a validity caveat, not a circularity. There are no load-bearing self-citations (the reference list contains no papers by the author), no imported uniqueness theorems, and no fitted parameter renamed as a prediction. The main risks—small sample sizes, ASR judge choice, and the RTF-vs-'real-time' hardware claim—are external-validity or reporting issues, not circular reasoning.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

The paper is an empirical recipe; its central claim depends on hand-chosen stage depths, thresholds, and evaluation assumptions rather than a formal derivation. No new physical or architectural entities are postulated.

free parameters (4)

Stage depth ladder = 22→16→12→8→6 blocks
Hand-chosen schedule; no criterion for why these exact depths beyond the RMS tolerance measurement; central to the recipe.
Pruning tolerance threshold = ≈25% block removal OK, ≈50% collapse (RMS 0.073 vs 0.005)
Empirical cutoff used to motivate gradual descent; threshold is not derived or cross-validated.
Fine-tuning hyperparameters = lr=5e-5, warmup=500 steps, batch=24,000 mel frames, EMA, NFE=32, CFG=2.0
Standard but hand-picked; results depend on these choices.
WER gating criterion = Not quantified ("near the previous stage's WER")
Per-stage go/no-go threshold is subjective and not specified numerically.

axioms (5)

domain assumption Conditional flow matching with Euler ODE integration is a valid training/generation objective for TTS.
Invoked throughout §3.4 without questioning the flow-matching framework.
domain assumption ASR WER (IndicWav2Vec or Vakyansh) is a valid proxy for intelligibility and content fidelity.
The entire gated evaluation and benchmark rely on this; no human listening test is provided (§3.6, §4.1).
domain assumption Teacher-generated audio is a sufficient training signal for the student.
The 17.6h corpus is entirely synthesized by IndicF5 (§3.5); the paper assumes this does not fundamentally cap quality.
domain assumption Removing an evenly spaced subset of transformer blocks preserves enough of the residual field that re-fine-tuning recovers quality.
The tolerance measurement (§3.3) supports this for moderate pruning, but the recoverability after fine-tuning is an empirical premise.
domain assumption The 50-sentence FLEURS subset is representative of general Hindi TTS quality.
Used as the independent benchmark (§4.1); no sampling details or confidence analysis are given.

pith-pipeline@v1.3.0-alltime-deepseek · 5339 in / 12456 out tokens · 102400 ms · 2026-08-01T18:43:12.276252+00:00 · methodology

0 comments

read the original abstract

We present a practical recipe for building a compact Hindi text-to-speech (TTS) model by distilling a large flow-matching teacher (IndicF5, 337M-parameter DiT) under a severe data budget (~17.6 hours). Training a small model from scratch on this much data fails outright. Instead we warm-start the student from the teacher by pruning depth only: keeping the teacher's width, text dimension, attention heads, and mel/text I/O fixed so all non-block tensors copy one-to-one, and retaining an evenly-spaced subset of transformer blocks. We first measure how much depth the teacher tolerates (it remains near-functional at -27% blocks but collapses past -50%), then descend gradually (22 -> 16 -> 12 -> 8 -> 6 blocks), re-fine-tuning after each prune, with each step gated by an objective ASR word-error-rate (WER) check. The resulting students reach WER 0.00 on unseen sentences at 249M and 190M parameters, and remain robust down to 131M; at 102M we observe a clear capacity cliff that we attribute to the data budget rather than the recipe. We also document two train/inference feature- and library-parity failures (mel filterbank and rotary-embedding library versions) that silently degrade audio, and a version-independent fix. The method yields a high-quality Hindi voice that runs in real time on a 6 GB laptop GPU. An independent 50-sentence FLEURS benchmark compares the released 190M student against its teacher and MMS-TTS-hin.

discussion (0)

Reference graph

Works this paper leans on

17 extracted references · 6 linked inside Pith

[1]

IndicF5: High-quality text-to-speech for indian languages.https:// huggingface.co/ai4bharat/IndicF5, 2025

AI4Bharat. IndicF5: High-quality text-to-speech for indian languages.https:// huggingface.co/ai4bharat/IndicF5, 2025

2025
[2]

F5-TTS: A fairytaler that fakes fluent and faithful speech with flow matching

Yushen Chen, Zhikang Niu, Ziyang Ma, Keqi Deng, Chunhui Wang, Jian Zhao, Kai Yu, and Xie Chen. F5-TTS: A fairytaler that fakes fluent and faithful speech with flow matching. arXiv preprint arXiv:2410.06885, 2024

Pith/arXiv arXiv 2024
[3]

Reducing transformer depth on demand with structured dropout

Angela Fan, Edouard Grave, and Armand Joulin. Reducing transformer depth on demand with structured dropout. InInternational Conference on Learning Representations (ICLR), 2020

2020
[4]

Kokoro-82M: An open-weight TTS model.https://huggingface.co/hexgrad/ Kokoro-82M, 2025

Hexgrad. Kokoro-82M: An open-weight TTS model.https://huggingface.co/hexgrad/ Kokoro-82M, 2025

2025
[5]

Distilling the knowledge in a neural network

Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531, 2015

Pith/arXiv arXiv 2015
[6]

Tahir Javed, Sumanth Doddapaneni, Abhigyan Raman, Kaushal Santosh Bhogale, Gowtham Ramesh, Anoop Kunchukuttan, Pratyush Kumar, and Mitesh M. Khapra. To- wards building ASR systems for the next billion users.Proceedings of the AAAI Conference on Artificial Intelligence, 2022. 6

2022
[7]

Yoon Kim and Alexander M. Rush. Sequence-level knowledge distillation. InConference on Empirical Methods in Natural Language Processing (EMNLP), 2016

2016
[8]

Yaron Lipman, Ricky T. Q. Chen, Heli Ben-Hamu, Maximilian Nickel, and Matthew Le. Flow matching for generative modeling. InInternational Conference on Learning Represen- tations (ICLR), 2023

2023
[9]

Flow straight and fast: Learning to generate and transfer data with rectified flow

Xingchao Liu, Chengyue Gong, and Qiang Liu. Flow straight and fast: Learning to generate and transfer data with rectified flow. InInternational Conference on Learning Representa- tions (ICLR), 2023

2023
[10]

Natural language guidance of high-fidelity text-to-speech with synthetic annotations.arXiv preprint arXiv:2402.01912, 2024

Dan Lyth and Simon King. Natural language guidance of high-fidelity text-to-speech with synthetic annotations.arXiv preprint arXiv:2402.01912, 2024

Pith/arXiv arXiv 2024
[11]

ShortGPT: Layers in large language models are more redundant than you expect.arXiv preprint arXiv:2403.03853, 2024

Xin Men, Mingyu Xu, Qingyu Zhang, Bingning Wang, Hongyu Lin, Yaojie Lu, Xianpei Han, and Weipeng Chen. ShortGPT: Layers in large language models are more redundant than you expect.arXiv preprint arXiv:2403.03853, 2024

Pith/arXiv arXiv 2024
[12]

Scalable diffusion models with transformers

William Peebles and Saining Xie. Scalable diffusion models with transformers. In IEEE/CVF International Conference on Computer Vision (ICCV), 2023

2023
[13]

Scaling speech technology to 1,000+ languages.Journal of Machine Learning Research, 2024

Vineel Pratap, Andros Tjandra, Bowen Shi, Paden Tomasello, Arun Babu, Sayani Kundu, Ali Elkahky, Zhaoheng Ni, Apoorv Vyas, Maryam Fazel-Zarandi, Alexei Baevski, Yossi Adi, Xiaohui Zhang, Wei-Ning Hsu, Alexis Conneau, and Michael Auli. Scaling speech technology to 1,000+ languages.Journal of Machine Learning Research, 2024

2024
[14]

Poor man’s BERT: Smaller and faster transformer models.arXiv preprint arXiv:2004.03844, 2020

Hassan Sajjad, Fahim Dalvi, Nadir Durrani, and Preslav Nakov. Poor man’s BERT: Smaller and faster transformer models.arXiv preprint arXiv:2004.03844, 2020

Pith/arXiv arXiv 2004
[15]

UTMOS: UTokyo-SaruLab system for VoiceMOS Challenge 2022

Takaaki Saeki, Detai Xin, Wataru Nakata, Tomoki Koriyama, Shinnosuke Takamichi, and Hiroshi Saruwatari. UTMOS: UTokyo-SaruLab system for VoiceMOS Challenge 2022. In Interspeech, 2022

2022
[16]

Vocos: Closing the gap between time-domain and fourier-based neural vocoders for high-quality audio synthesis.arXiv preprint arXiv:2306.00814, 2023

Hubert Siuzdak. Vocos: Closing the gap between time-domain and fourier-based neural vocoders for high-quality audio synthesis.arXiv preprint arXiv:2306.00814, 2023

Pith/arXiv arXiv 2023
[17]

ECAPA-TDNN: Emphasized channel attention, propagation and aggregation in TDNN based speaker verification

Brecht Desplanques, Jenthe Thienpondt, and Kris Demuynck. ECAPA-TDNN: Emphasized channel attention, propagation and aggregation in TDNN based speaker verification. In Interspeech, 2020. 7

2020

[1] [1]

IndicF5: High-quality text-to-speech for indian languages.https:// huggingface.co/ai4bharat/IndicF5, 2025

AI4Bharat. IndicF5: High-quality text-to-speech for indian languages.https:// huggingface.co/ai4bharat/IndicF5, 2025

2025

[2] [2]

F5-TTS: A fairytaler that fakes fluent and faithful speech with flow matching

Yushen Chen, Zhikang Niu, Ziyang Ma, Keqi Deng, Chunhui Wang, Jian Zhao, Kai Yu, and Xie Chen. F5-TTS: A fairytaler that fakes fluent and faithful speech with flow matching. arXiv preprint arXiv:2410.06885, 2024

Pith/arXiv arXiv 2024

[3] [3]

Reducing transformer depth on demand with structured dropout

Angela Fan, Edouard Grave, and Armand Joulin. Reducing transformer depth on demand with structured dropout. InInternational Conference on Learning Representations (ICLR), 2020

2020

[4] [4]

Kokoro-82M: An open-weight TTS model.https://huggingface.co/hexgrad/ Kokoro-82M, 2025

Hexgrad. Kokoro-82M: An open-weight TTS model.https://huggingface.co/hexgrad/ Kokoro-82M, 2025

2025

[5] [5]

Distilling the knowledge in a neural network

Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531, 2015

Pith/arXiv arXiv 2015

[6] [6]

Tahir Javed, Sumanth Doddapaneni, Abhigyan Raman, Kaushal Santosh Bhogale, Gowtham Ramesh, Anoop Kunchukuttan, Pratyush Kumar, and Mitesh M. Khapra. To- wards building ASR systems for the next billion users.Proceedings of the AAAI Conference on Artificial Intelligence, 2022. 6

2022

[7] [7]

Yoon Kim and Alexander M. Rush. Sequence-level knowledge distillation. InConference on Empirical Methods in Natural Language Processing (EMNLP), 2016

2016

[8] [8]

Yaron Lipman, Ricky T. Q. Chen, Heli Ben-Hamu, Maximilian Nickel, and Matthew Le. Flow matching for generative modeling. InInternational Conference on Learning Represen- tations (ICLR), 2023

2023

[9] [9]

Flow straight and fast: Learning to generate and transfer data with rectified flow

Xingchao Liu, Chengyue Gong, and Qiang Liu. Flow straight and fast: Learning to generate and transfer data with rectified flow. InInternational Conference on Learning Representa- tions (ICLR), 2023

2023

[10] [10]

Natural language guidance of high-fidelity text-to-speech with synthetic annotations.arXiv preprint arXiv:2402.01912, 2024

Dan Lyth and Simon King. Natural language guidance of high-fidelity text-to-speech with synthetic annotations.arXiv preprint arXiv:2402.01912, 2024

Pith/arXiv arXiv 2024

[11] [11]

ShortGPT: Layers in large language models are more redundant than you expect.arXiv preprint arXiv:2403.03853, 2024

Xin Men, Mingyu Xu, Qingyu Zhang, Bingning Wang, Hongyu Lin, Yaojie Lu, Xianpei Han, and Weipeng Chen. ShortGPT: Layers in large language models are more redundant than you expect.arXiv preprint arXiv:2403.03853, 2024

Pith/arXiv arXiv 2024

[12] [12]

Scalable diffusion models with transformers

William Peebles and Saining Xie. Scalable diffusion models with transformers. In IEEE/CVF International Conference on Computer Vision (ICCV), 2023

2023

[13] [13]

Scaling speech technology to 1,000+ languages.Journal of Machine Learning Research, 2024

Vineel Pratap, Andros Tjandra, Bowen Shi, Paden Tomasello, Arun Babu, Sayani Kundu, Ali Elkahky, Zhaoheng Ni, Apoorv Vyas, Maryam Fazel-Zarandi, Alexei Baevski, Yossi Adi, Xiaohui Zhang, Wei-Ning Hsu, Alexis Conneau, and Michael Auli. Scaling speech technology to 1,000+ languages.Journal of Machine Learning Research, 2024

2024

[14] [14]

Poor man’s BERT: Smaller and faster transformer models.arXiv preprint arXiv:2004.03844, 2020

Hassan Sajjad, Fahim Dalvi, Nadir Durrani, and Preslav Nakov. Poor man’s BERT: Smaller and faster transformer models.arXiv preprint arXiv:2004.03844, 2020

Pith/arXiv arXiv 2004

[15] [15]

UTMOS: UTokyo-SaruLab system for VoiceMOS Challenge 2022

Takaaki Saeki, Detai Xin, Wataru Nakata, Tomoki Koriyama, Shinnosuke Takamichi, and Hiroshi Saruwatari. UTMOS: UTokyo-SaruLab system for VoiceMOS Challenge 2022. In Interspeech, 2022

2022

[16] [16]

Vocos: Closing the gap between time-domain and fourier-based neural vocoders for high-quality audio synthesis.arXiv preprint arXiv:2306.00814, 2023

Hubert Siuzdak. Vocos: Closing the gap between time-domain and fourier-based neural vocoders for high-quality audio synthesis.arXiv preprint arXiv:2306.00814, 2023

Pith/arXiv arXiv 2023

[17] [17]

ECAPA-TDNN: Emphasized channel attention, propagation and aggregation in TDNN based speaker verification

Brecht Desplanques, Jenthe Thienpondt, and Kris Demuynck. ECAPA-TDNN: Emphasized channel attention, propagation and aggregation in TDNN based speaker verification. In Interspeech, 2020. 7

2020