Pith. sign in

REVIEW 3 major objections 6 minor 7 cited by

JAM: A Tiny Flow-based Song Generator with Fine-grained Controllability and Aesthetic Alignment

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read JAM is a 530M-parameter flow-matching song generator that conditions on word- and phoneme-level timestamps, giving musicians control over the timing of every word and the total duration while reducing word and phoneme error rates by more…

desk verdict A real controllability contribution with a released model and benchmark, but the headline 3x intelligibility claim only holds with oracle timing inputs that baselines don't get. read the letter →

arxiv 2507.20880 v1 pith:C4SNRRIU submitted 2025-07-28 cs.SD cs.AI

classification cs.SDcs.AI
keywords lyrics-to-songgenerationflowmatchingword-leveltimingcontroldurationdirectpreferenceoptimizationsongevaluationbenchmarksingingvoicesynthesiscontrollablemusic
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

This paper claims that a compact 530M-parameter flow-matching model can give musicians precise control over when each word and phoneme is sung and how long the song lasts, while matching or beating far larger lyrics-to-song systems. The central claim is that conditioning on word- and phoneme-level timestamps, rather than just the lyric text, dramatically improves vocal intelligibility: on the paper's new JAME benchmark, word error rate falls to 0.151 and phoneme error rate to 0.101, a reduction of more than threefold over prior open models. It further claims that iterative preference optimization using automated aesthetic scores, with no human annotations, moves generated songs toward what listeners prefer on enjoyment, musicality, and structural clarity. The paper also argues that a public, genre-diverse, contamination-controlled evaluation set is necessary for fair comparison in this fast-moving area.

What carries the argument

The load-bearing mechanism is the word-level phoneme alignment map: for each word $(w_i, t_i^s, t_i^e)$ the model spreads IPA phonemes evenly across the latent frames between start and end, surrounded by SONG FILLER tokens for silence, instrumental passages, and padding, and VOCAL FILLER tokens for transitions inside a word. This upsampled phoneme sequence is embedded and convolutionally downsampled to a latent-aligned lyric embedding that is injected into the first eight transformer layers along with a learned padding bias that distinguishes content from silence. The training objective is rectified flow matching, and multi-condition classifier-free guidance keeps style and lyric conditioning separable at inference.

What would settle it

Manually transcribe a held-out set of generated songs with human-verified word boundaries and compute WER against a human reference instead of the automatic recognizer's transcript; if the gap between JAM and the baselines shrinks substantially, the reported 3x improvement partly reflects recognition bias rather than true lyrical fidelity.

Watch

Extended reading notes

Core claim

The paper's core discovery is that lyrics-to-song generation becomes far more intelligible and controllable when the generator receives temporally aligned phonemes instead of a bare lyric string. JAM converts each word with its start and end time into IPA phonemes, spreads them evenly across the corresponding latent frames, and marks non-vocal and intra-word gaps with distinct filler tokens; this alignment signal, combined with a global duration embedding and a learnable bias on padded regions, lets a 530M-parameter model generate full 44.1kHz songs up to 3 minutes 50 seconds while halving the word and phoneme error rates of the next-best system. A second discovery is that iterative direct preference optimization, using an automated song-quality scorer as reward and a ground-truth reconstruction term as regularizer, raises aesthetic scores across rounds even as it introduces a mild realism cost measured by Fréchet Audio Distance.

Load-bearing premise

The paper assumes the automatic speech recognizer's word timestamps, extracted from separated vocal tracks, are accurate enough to serve as ground truth for both training and evaluation; if those timestamps carry systematic bias, the model can learn and reproduce that bias, inflating the measured word and phoneme error-rate gains.

Editorial extensions

If this is right

  • A musician can specify the start and end time of every word and the overall song length, enabling precise prosody, phrasing, and pacing without any retraining.
  • The reported word error rate of about 0.15 and phoneme error rate of about 0.10 imply that generated vocals are several times more intelligible than those of previous open models, making them closer to usable production drafts.
  • JAME offers a public, five-genre benchmark built from songs released after the baseline training cutoffs, so future lyrics-to-song models can be compared on identical prompts and references without training-data leakage.
  • Each additional round of direct preference optimization improves aesthetic scores, and adding a ground-truth reconstruction term to the DPO loss reduces the accompanying drift in genre accuracy and Fréchet Audio Distance.
  • Because beat-quantized timestamps degrade quality only mildly while naive predicted timestamps degrade it sharply, practical deployment will depend on a dedicated learned duration predictor rather than a general-purpose language model.

Reading between the lines

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

  • If word-level timing is what drives lyric fidelity, then parameter scale is not the primary lever: much smaller, timing-conditioned models may continue to out-sing larger models, and existing large generators could be adapted to accept the same conditioning.
  • The evaluation loop is partly circular because the same automatic recognizer that created the training timestamps scores the outputs, so the model may be learning to match that recognizer's transcription biases; a human-verified reference transcript would be needed to confirm the 3x gain is genuinely audible.
  • A learned duration predictor trained jointly with the generator could preserve oracle-level controllability without requiring users to supply accurate word timings, widening the system beyond expert musicians.
  • The filler-token alignment scheme generalizes beyond songs: any generative task that must place discrete events on a continuous latent timeline--for example, speech with controllable emphasis or instrumental performance with note onsets--could reuse the same mechanism.
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.

Referee Report

3 major / 6 minor

Summary. The paper proposes JAM, a 530M-parameter rectified-flow song generator based on a 16-layer LLaMA-style DiT backbone, conditioned on lyrics with word- and phoneme-level timings, target duration, and style. Training proceeds through pretraining, full-song supervised fine-tuning, and iterative direct preference optimization (DPO) using SongEval scores as a synthetic reward. The authors also introduce JAME, a public evaluation set of 250 post-training-cutoff songs across five genres, and report objective and subjective comparisons against DiffRhythm, ACE-Step, LeVo, and YuE. The headline claims are that JAM is the first system with word-level timing and duration control, achieves over 3x improvement in WER/PER over prior work, and outperforms baselines on several aesthetic metrics despite being the smallest model.

Significance. If the central claims hold, the paper makes a useful contribution to lyrics-to-song generation: it demonstrates a compact model with fine-grained temporal controllability, introduces a public benchmarking dataset designed to avoid data contamination, and provides detailed ablations and an honest discussion of limitations. The release of the model, code, and evaluation set is a concrete practical benefit. However, the headline quantitative claims rest on two protocol choices that materially weaken them: JAM is evaluated with oracle word-level timestamps while baselines are not, and the same ASR system is used to create both the training annotations and the evaluation ground truth. The paper's own Section 6 results show the oracle dependence directly. The core ideas are defensible, but the evidence as presented does not yet support the strongest claims.

major comments (3)
  1. [Section 4.3 / Table 3 / Table 9] The WER/PER comparison in Table 3 is not apples-to-apples. JAM is evaluated with oracle word-level timestamps from JAME, while the baselines receive only lyrics and style and must infer timing internally. Since word-level timing is the conditioning signal JAM was trained on, the reported >3x WER/PER advantage is partly an artifact of giving JAM privileged information. The paper's own Section 6.3 and Table 9 make this explicit: replacing oracle timestamps with GPT-Dur predicted timestamps raises PER from 0.101 to 0.320, worse than DiffRhythm's 0.264. To support the abstract's 'over 3x lyrical fidelity' claim, the authors should either give the same word-level timing information to the baselines or report JAM without oracle timings as the primary configuration, ideally with a learned duration predictor.
  2. [Section 4.1 vs. Section 4.3] The intelligibility metrics are circular with respect to the ASR system. Parakeet is used both to generate the training-conditioning annotations yi=(wi,tstart_i,tend_i) in Section 4.1 and to produce the reference lyrics and evaluation transcripts in Section 4.3. JAM is therefore trained to reproduce Parakeet's transcription biases, and its outputs are then scored against Parakeet's own transcriptions. No human verification of timestamps or transcripts is reported. I recommend human verification on a subset of JAME and/or evaluation with a different ASR system, reporting WER/PER for both, so that the intelligibility numbers can be interpreted as properties of the generated audio rather than of the ASR pipeline.
  3. [Section 3.7 / Table 4] SongEval serves both as the DPO reward signal during training and as an evaluation metric in Table 4. The steady improvement in SongEval dimensions across DPO rounds is therefore expected from optimization of the reward itself, and the 'aesthetic alignment' claim is partly circular. The authors should include an evaluation signal not used in training—for example, human ratings on the same samples, or a different held-out aesthetic model—to demonstrate that the DPO procedure improves aesthetics independently of the reward used.
minor comments (6)
  1. [Algorithm 1, line 2] The loop header reads 'for each (wi, tstart_i, tstart_i)' but should presumably be 'for each (wi, tstart_i, tend_i)'.
  2. [Section 4.3] There are typographical inconsistencies: 'HDmucs' should be 'HTDemucs', and 'HDemucs' appears in the metrics paragraph.
  3. [Section 1 and Table 3] The abstract and introduction claim 'over 3x' WER/PER improvement, but Table 3 shows JAM's WER is about 2.3x better than DiffRhythm, 2.7x better than ACE-Step, and 3.4x better than YuE. The claim should be qualified as relative to the strongest baseline or rephrased.
  4. [Section 6.1] The phrase 'using the all-in-one method (Kim and Nam 2023)' is ambiguous; the method name and relevant configuration should be stated explicitly.
  5. [Section 3.7] The DPO-FM loss in Eq. (8) uses u(xw_t, t; θ) without explicitly showing the conditioning c, whereas Eq. (2) uses u(zt, t, c; θ). Aligning the notation would improve readability.
  6. [Section 7] The limitation paragraph says the system lacks 'phoneme-level duration control', which seems to contradict the abstract's claim of 'word- and phoneme-level timing and duration control.' The distinction between phoneme-level conditioning and independent phoneme-duration control should be clarified.

Circularity Check

2 steps flagged · score 6.0 of 10

Two objective evaluation loops are partly circular: Parakeet supplies both training timings and WER/PER ground truth, and SongEval is both DPO reward and reported metric.

  1. self definitional [Section 4.1 (Dataset Setup) and Section 4.3 (Objective Evaluation, Evaluation Data Preparation and Metrics)]
    "We then use Parakeet-tdt-0.6b-v2 (NVIDIA 2025) to transcribe the vocal stems and extract word-level timestamps, yielding lyric annotations of the form yi = (wi, tstart i , tend i ) for each sample x. ... Ground-truth lyrics are retrieved using HDmucs and Parakeet. ... Parakeet (NVIDIA 2025) is used to transcribe the audio into lyrics."

    The same Parakeet ASR model creates both the training-time word-level annotations yi=(wi,tstart_i,tend_i) (Section 4.1) and the JAME 'ground-truth' lyrics/transcripts (Section 4.3). JAM is trained to follow Parakeet's transcription and timing behavior, and its generated vocals are then scored by Parakeet against Parakeet's own output. Any systematic ASR bias is therefore shared by training and evaluation, so the reported >3x WER/PER improvement over baselines (which were not trained with Parakeet timings) partly measures the model's fit to the evaluator rather than independent intelligibility. No human verification of timestamps or transcripts is reported.

  2. fitted input called prediction [Section 3.7 (Aesthetic Preference Alignment) and Section 4.3 (Metrics); Table 4]
    "We employ SongEval to score each generated output. While SongEval evaluates songs across five distinct criteria on a five-point scale, we compute an averaged score across these dimensions and select the samples with the highest and lowest average SongEval scores as win and loss candidates, respectively. ... We adopt both Audiobox-aesthetic and SongEval as model-based evaluation tools."

    SongEval is not an external benchmark here: Section 3.7 uses averaged SongEval scores to select DPO win/loss pairs, and the DPO loss (Eq. 8) directly increases the relative likelihood of the high-SongEval samples. Table 4 then reports SongEval↑ as an objective evaluation result. Optimizing a metric and then presenting that same metric as evidence of improvement is circular; the gains are generated by the optimization objective, not by an independent aesthetic measurement. The same issue applies to Table 3, where SongEval is one of the reported metrics.

full rationale

The paper's architecture and training pipeline are not inherently circular: the flow-matching objective, the word-level phoneme alignment algorithm, the token-level duration control ablation, and the GPT-Dur / Direct-Quant experiments are genuine, falsifiable investigations, and the human subjective evaluation (Table 6) provides some independent support. However, two quantitative evaluation loops reduce partly by construction. First, WER/PER 'ground truth' in JAME is produced by the same Parakeet ASR pipeline that generated the word-level training timestamps in Section 4.1, so JAM is effectively trained to match the evaluator's transcription behavior; the >3x WER/PER advantage therefore conflates genuine intelligibility with fit to Parakeet. Second, SongEval serves both as the DPO reward (win/loss selection) and as an evaluation metric in Table 4, so SongEval improvements after DPO are expected from the optimization objective. The paper's own Section 6/Table 9 exposes the fragility of the headline claim: with a non-oracle duration predictor (GPT-Dur), WER/PER degrade to 0.370/0.320, worse than DiffRhythm's 0.348/0.264. Because a substantial part of the headline outperformance is not independently measured, the circularity score is 6 rather than 0-2, but the presence of a human evaluation and real ablations prevents a higher score.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central empirical claims depend mainly on the accuracy of automatic word timestamps, the validity of SongEval as a reward, and the hyperparameters of the DPO loop. No new physical or mathematical entity is introduced.

free parameters (4)
  • DPO temperature beta = 2000
    Set empirically in Section 3.7; controls DPO sharpness and affects all alignment results.
  • DPO ground-truth loss weight lambda = 0.2
    Set empirically in Section 3.7; balances reconstruction vs preference loss.
  • DPO preference margin threshold = 0.15
    Minimum average SongEval score difference required to form a win/loss pair; chosen by hand in Section 3.7.
  • CFG guidance scales alpha_s, alpha_l = not reported
    Used in Eq. (10); the paper says suitable CFG values were determined empirically for each policy but does not list them.
assumptions (5)
  • standard math Flow matching objective and DPO-Diffusion loss are valid for rectified flow song generation.
    Uses Eq. (2) and Eq. (8) from Lipman et al. and Wallace et al.; no proof given, accepted as background.
  • domain assumption The frozen VAE composed of Stable Audio Open encoder and DiffRhythm decoder reconstructs song audio from latent space well enough for training and evaluation.
    Section 3.3; if reconstruction is poor, flow matching operates on a lossy latent space.
  • domain assumption Parakeet ASR word-level timestamps are accurate ground truth for lyric timing and content.
    Sections 4.1 and 4.3; used for training conditioning and for JAME ground-truth lyrics; no human verification.
  • domain assumption SongEval scores reflect human aesthetic preference and are suitable as a reward signal.
    Section 3.7; no validation against human ratings is provided in this paper.
  • domain assumption JAME songs released after baselines' training cut-offs have no data contamination.
    Section 4.3; the assumption is based on release dates, not on verifying the baselines' actual training corpora.

how reviews work

0 comments
Cite this review

Pith. "Pith review of JAM: A Tiny Flow-based Song Generator with Fine-grained Controllability and Aesthetic Alignment." pith.science (2026). https://pith.science/paper/C4SNRRIU

@misc{pith2026250720880,
  author       = {Pith},
  title        = {Pith review of: JAM: A Tiny Flow-based Song Generator with Fine-grained Controllability and Aesthetic Alignment},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/C4SNRRIU}},
  note         = {Machine review of arXiv:2507.20880}
}
read the original abstract

Diffusion and flow-matching models have revolutionized automatic text-to-audio generation in recent times. These models are increasingly capable of generating high quality and faithful audio outputs capturing to speech and acoustic events. However, there is still much room for improvement in creative audio generation that primarily involves music and songs. Recent open lyrics-to-song models, such as, DiffRhythm, ACE-Step, and LeVo, have set an acceptable standard in automatic song generation for recreational use. However, these models lack fine-grained word-level controllability often desired by musicians in their workflows. To the best of our knowledge, our flow-matching-based JAM is the first effort toward endowing word-level timing and duration control in song generation, allowing fine-grained vocal control. To enhance the quality of generated songs to better align with human preferences, we implement aesthetic alignment through Direct Preference Optimization, which iteratively refines the model using a synthetic dataset, eliminating the need or manual data annotations. Furthermore, we aim to standardize the evaluation of such lyrics-to-song models through our public evaluation dataset JAME. We show that JAM outperforms the existing models in terms of the music-specific attributes.

Figures

Figures reproduced from arXiv: 2507.20880 by the authors.

Figure 1
Figure 1. A depiction of our proposed architecture and training pipeline. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗

Discussion (0). Sign in to comment.

Forward citations

Cited by 7 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. MIDI-Informed Singing Accompaniment Generation in a Compositional Song Pipeline

    cs.SD 2026-02 unverdicted novelty 7.0 of 10

    MIDI-SAG generates consistent long-form singing accompaniments by feeding symbolic MIDI timing, chords, and structure labels into a compositional pipeline built from pre-trained modules.

  2. MusicMark: A Robust Generative Watermarking Framework for Music Generation

    cs.SD 2026-07 conditional novelty 6.5 of 10

    Embedding watermark bits into diffusion semantic latents via a frozen-backbone adapter yields far more robust music provenance than post-hoc watermarking under codecs and cover-song attacks.

  3. MPEcho: A Melody and Phoneme-Aware Generative Framework for Controllable Cover Song Generation

    cs.SD 2026-07 conditional novelty 6.0 of 10

    Adding SVS-style phoneme conditioning and a length regulator to melody-guided cover generation cuts phoneme error rate from 45.6% to 18.7% while holding melody metrics.

  4. Genre Bias or Aesthetic Perception? Identifying and Mitigating Shortcut Learning in Music Evaluation

    cs.SD 2026-07 conditional novelty 6.0 of 10

    SongEval's aesthetic scores are shortcut by genre (pop-centric), and a focal-loss plus group-regularized training objective measurably reduces that genre bias.

  5. Auditing Training Data in Generative Music Models via Black-Box Membership Inference

    cs.LG 2026-05 unverdicted novelty 6.0 of 10

    Black-box membership inference on text-to-music models reaches up to 98.6% accuracy by training an auditor on semantic alignment patterns extracted from shadow-model generations.

  6. LaDA-Band: Language Diffusion Models for Vocal-to-Accompaniment Generation

    cs.SD 2026-04 unverdicted novelty 6.0 of 10

    LaDA-Band applies discrete masked diffusion with dual-track conditioning and progressive training to generate vocal-to-accompaniment tracks that improve acoustic authenticity, global coherence, and dynamic orchestrati...

  7. LaDA-Band: Language Diffusion Models for Vocal-to-Accompaniment Generation

    cs.SD 2026-04 conditional novelty 5.0 of 10

    LaDA-Band generates complete vocal-aligned instrumental accompaniments with discrete masked diffusion, claiming simultaneous gains in audio fidelity, long-range coherence, and orchestration quality.

Reference graph

Works this paper leans on

40 extracted references · 10 canonical work pages · cited by 6 Pith papers

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    I.; Borsos, Z.; Engel, J.; Verzetti, M.; Caillon, A.; Huang, Q.; Jansen, A.; Roberts, A.; Tagliasacchi, M.; Sharifi, M.; Zeghidour, N.; and Frank, C

    Agostinelli, A.; Denk, T. I.; Borsos, Z.; Engel, J.; Verzetti, M.; Caillon, A.; Huang, Q.; Jansen, A.; Roberts, A.; Tagliasacchi, M.; Sharifi, M.; Zeghidour, N.; and Frank, C. 2023. MusicLM: Generating Music From Text. arXiv:2301.11325

  4. [4]

    R.; Schaefer, R

    Agres, K. R.; Schaefer, R. S.; Volk, A.; Van Hooren, S.; Holzapfel, A.; Dalla Bella, S.; M \"u ller, M.; De Witte, M.; Herremans, D.; Ramirez Melendez, R.; et al. 2021. Music, computing, and health: a roadmap for the current and future roles of music technology for health care and well-being. Music & Science, 4: 2059204321997709

  5. [5]

    Chen, H.; Jiang, Y.; Ma, G.; Hao, C.; Wang, S.; Yao, J.; Ning, Z.; Meng, M.; Luan, J.; and Xie, L. 2025 a . DiffRhythm+: Controllable and Flexible Full-Length Song Generation with Preference Optimization. arXiv:2507.12890

  6. [6]

    Chen, K.; Wu, Y.; Liu, H.; Nezhurina, M.; Berg-Kirkpatrick, T.; and Dubnov, S. 2023. MusicLDM: Enhancing Novelty in Text-to-Music Generation Using Beat-Synchronous Mixup Strategies. arXiv:2308.01546

  7. [7]

    Chen, Y.; Niu, Z.; Ma, Z.; Deng, K.; Wang, C.; Zhao, J.; Yu, K.; and Chen, X. 2025 b . F5-TTS: A Fairytaler that Fakes Fluent and Faithful Speech with Flow Matching. arXiv:2410.06885

  8. [8]

    Copet, J.; Kreuk, F.; Gat, I.; Remez, T.; Kant, D.; Synnaeve, G.; Adi, Y.; and Défossez, A. 2024. Simple and Controllable Music Generation. arXiv:2306.05284

Show all 40 references
  1. [9]

    D \'e fossez, A. 2021. Hybrid Spectrogram and Waveform Source Separation. In Proceedings of the ISMIR 2021 Workshop on Music Source Separation

  2. [10]

    Esser, P.; Kulal, S.; Blattmann, A.; Entezari, R.; Müller, J.; Saini, H.; Levi, Y.; Lorenz, D.; Sauer, A.; Boesel, F.; Podell, D.; Dockhorn, T.; English, Z.; Lacey, K.; Goodwin, A.; Marek, Y.; and Rombach, R. 2024. Scaling Rectified Flow Transformers for High-Resolution Image ...

  3. [11]

    D.; Carr, C.; Zukowski, Z.; Taylor, J.; and Pons, J

    Evans, Z.; Parker, J. D.; Carr, C.; Zukowski, Z.; Taylor, J.; and Pons, J. 2024. Stable Audio Open. arXiv:2407.14358

  4. [12]

    Freeman III, W. J. 1998. A neurobiological role of music in social bonding

  5. [13]

    Ghosal, D.; Majumder, N.; Mehrish, A.; and Poria, S. 2023. Text-to-Audio Generation using Instruction Tuned LLM and Latent Diffusion Model. arXiv preprint arXiv:2304.13731

  6. [14]

    Gong, J.; Zhao, S.; Wang, S.; Xu, S.; and Guo, J. 2025. ACE-Step: A Step Towards Music Generation Foundation Model. arXiv:2506.00045

  7. [15]

    Ho, J.; Jain, A.; and Abbeel, P. 2020. Denoising Diffusion Probabilistic Models. arXiv:2006.11239

  8. [16]

    Hong, Z.; Huang, R.; Cheng, X.; Wang, Y.; Li, R.; You, F.; Zhao, Z.; and Zhang, Z. 2024. Text-to-Song: Towards Controllable Music Generation Incorporating Vocals and Accompaniment. arXiv:2404.09313

  9. [18]

    A.; Li, C.; Valle, R.; Catanzaro, B.; and Poria, S

    Hung, C.-Y.; Majumder, N.; Kong, Z.; Mehrish, A.; Bagherzadeh, A. A.; Li, C.; Valle, R.; Catanzaro, B.; and Poria, S. 2025 b . TangoFlux: Super Fast and Faithful Text to Audio Generation with Flow Matching and Clap-Ranked Preference Optimization. arXiv:2412.21037

  10. [19]

    Jiang, Z.; Ren, Y.; Li, R.; Ji, S.; Zhang, B.; Ye, Z.; Zhang, C.; Jionghao, B.; Yang, X.; Zuo, J.; Zhang, Y.; Liu, R.; Yin, X.; and Zhao, Z. 2025. MegaTTS 3: Sparse Alignment Enhanced Latent Diffusion Transformer for Zero-Shot Speech Synthesis. arXiv:2502.18924

  11. [20]

    Kim, T.; and Nam, J. 2023. All-In-One Metrical And Functional Structure Analysis With Neighborhood Attentions on Demixed Audio. arXiv:2307.16425

  12. [21]

    Lam, M. W. Y.; Tian, Q.; Li, T.; Yin, Z.; Feng, S.; Tu, M.; Ji, Y.; Xia, R.; Ma, M.; Song, X.; Chen, J.; Wang, Y.; and Wang, Y. 2023. Efficient Neural Music Generation. arXiv:2305.15719

  13. [22]

    Lei, S.; Xu, Y.; Lin, Z.; Zhang, H.; Tan, W.; Chen, H.; Yu, J.; Zhang, Y.; Yang, C.; Zhu, H.; Wang, S.; Wu, Z.; and Yu, D. 2025. LeVo: High-Quality Song Generation with Multi-Preference Alignment. arXiv:2506.07520

  14. [23]

    Lei, S.; Zhou, Y.; Tang, B.; Lam, M. W. Y.; Liu, F.; Liu, H.; Wu, J.; Kang, S.; Wu, Z.; and Meng, H. 2024. SongCreator: Lyrics-based Universal Song Generation. arXiv:2409.06029

  15. [24]

    Lipman, Y.; Chen, R. T. Q.; Ben-Hamu, H.; Nickel, M.; and Le, M. 2023. Flow Matching for Generative Modeling. arXiv:2210.02747

  16. [25]

    Liu, H.; Yuan, Y.; Liu, X.; Mei, X.; Kong, Q.; Tian, Q.; Wang, Y.; Wang, W.; Wang, Y.; and Plumbley, M. D. 2024. AudioLDM 2: Learning Holistic Audio Generation with Self-supervised Pretraining. arXiv:2308.05734

  17. [26]

    Liu, Z.; Ding, S.; Zhang, Z.; Dong, X.; Zhang, P.; Zang, Y.; Cao, Y.; Lin, D.; and Wang, J. 2025. SongGen: A Single Stage Auto-regressive Transformer for Text-to-Song Generation. arXiv:2502.13128

  18. [28]

    Majumder, N.; Hung, C.-Y.; Ghosal, D.; Hsu, W.-N.; Mihalcea, R.; and Poria, S. 2024 b . Tango 2: Aligning Diffusion-based Text-to-Audio Generations through Direct Preference Optimization. arXiv:2404.09956

  19. [29]

    Melechovsky, J.; Guo, Z.; Ghosal, D.; Majumder, N.; Herremans, D.; and Poria, S. 2024. Mustango: Toward Controllable Text-to-Music Generation. arXiv:2311.08355

  20. [30]

    Ning, Z.; Chen, H.; Jiang, Y.; Hao, C.; Ma, G.; Wang, S.; Yao, J.; and Xie, L. 2025. DiffRhythm: Blazingly Fast and Embarrassingly Simple End-to-End Full-Length Song Generation with Latent Diffusion. arXiv:2503.01183

  21. [31]

    NVIDIA . 2025. Parakeet TDT 0.6B V2. https://huggingface.co/nvidia/parakeet-tdt-0.6b-v2. Hugging Face model. 600M‑parameter ASR, CC‑BY‑4.0, released May 1 2025

  22. [32]

    Peebles, W.; and Xie, S. 2023. Scalable Diffusion Models with Transformers. arXiv:2212.09748

  23. [33]

    D.; and Finn, C

    Rafailov, R.; Sharma, A.; Mitchell, E.; Ermon, S.; Manning, C. D.; and Finn, C. 2024. Direct Preference Optimization: Your Language Model is Secretly a Reward Model. arXiv:2305.18290

  24. [34]

    Rouard, S.; Massa, F.; and D \'e fossez, A. 2023. Hybrid Transformers for Music Source Separation. In ICASSP 23

  25. [35]

    Song, J.; Meng, C.; and Ermon, S. 2022. Denoising Diffusion Implicit Models. arXiv:2010.02502

  26. [36]

    Spring-Media . 2025. DeepPhonemizer: Grapheme-to-Phoneme Conversion. https://github.com/spring-media/DeepPhonemizer. GitHub repository, MIT License, current version as of July 2025

  27. [37]

    van den Oord, A.; Vinyals, O.; and Kavukcuoglu, K. 2018. Neural Discrete Representation Learning. arXiv:1711.00937

  28. [38]

    Wallace, B.; Dang, M.; Rafailov, R.; Zhou, L.; Lou, A.; Purushwalkam, S.; Ermon, S.; Xiong, C.; Joty, S.; and Naik, N. 2023. Diffusion Model Alignment Using Direct Preference Optimization. arXiv:2311.12908

  29. [39]

    Wu, H.; Xiao, B.; Codella, N.; Liu, M.; Dai, X.; Yuan, L.; and Zhang, L. 2021. CvT: Introducing Convolutions to Vision Transformers. arXiv:2103.15808

  30. [40]

    Yao, J.; Ma, G.; Xue, H.; Chen, H.; Hao, C.; Jiang, Y.; Liu, H.; Yuan, R.; Xu, J.; Xue, W.; Liu, H.; and Xie, L. 2025. SongEval: A Benchmark Dataset for Song Aesthetics Evaluation. arXiv:2505.10793

  31. [41]

    Yuan, R.; Lin, H.; Guo, S.; Zhang, G.; Pan, J.; Zang, Y.; Liu, H.; Liang, Y.; Ma, W.; Du, X.; Du, X.; Ye, Z.; Zheng, T.; Ma, Y.; Liu, M.; Tian, Z.; Zhou, Z.; Xue, L.; Qu, X.; Li, Y.; Wu, S.; Shen, T.; Ma, Z.; Zhan, J.; Wang, C.; Wang, Y.; Chi, X.; Zhang, X.; Yang, Z.; Wang, X....

  32. [42]

    Zhu, H.; Zhou, Y.; Chen, H.; Yu, J.; Ma, Z.; Gu, R.; Luo, Y.; Tan, W.; and Chen, X. 2025. MuQ: Self-Supervised Music Representation Learning with Mel Residual Vector Quantization. arXiv:2501.01108

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.