{"id":"41d8e47a-74ce-4275-b3e4-7a687fd71bfe","arxiv_id":"2605.20830","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Raon-OpenTTS provides an open 510K-hour curated speech dataset and DiT-based TTS models up to 1B parameters that achieve competitive WER and speaker similarity on benchmarks versus closed models trained on millions of hours.","lead":"The paper introduces a large open dataset of 615K hours of public English speech, filtered to 510K high-quality hours, and trains diffusion transformer TTS models up to 1B parameters that match closed proprietary models on key metrics. This could enable more researchers to build competitive speech systems without access to private data.","discovery_kind":"new_application","skeptic_critique":{"model":"grok-4.3","headline":"Model-based filtering from 615K to 510K hours may reduce diversity or inject selection bias, weakening the claim that open data matches proprietary-scale performance.","rationale":"The reader’s weakest_assumption directly identifies the filtering step as the critical unverified link. Because the full manuscript is stated to be available yet still provides no quantitative validation of the filter’s effect on diversity or bias, the concern remains load-bearing and justifies moving from UNVERDICTED to CONDITIONAL pending the proposed distribution check.","tokens_in":1955,"tokens_out":379,"duration_ms":24359,"concrete_test":"Compute and compare the empirical distributions of three statistics (mean SNR, speaker-embedding cosine-distance entropy, and F0 variance) on a 10 K-hour random sample of Raon-OpenTTS-Pool versus the corresponding filtered subset; a >15 % drop in any entropy measure after filtering would indicate diversity loss sufficient to question the performance claim.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The headline comparability result (Raon-OpenTTS-1B matching Qwen3-TTS / CosyVoice 3 on Seed-TTS-Eval and CV3-Hard-EN) rests on Raon-OpenTTS-Core being a high-quality, unbiased subset. The filtering step is described only at the level of “model-based pipeline”; no details are given on the filter model architecture, decision thresholds, or post-filter statistics (e.g., speaker-embedding entropy, SNR distribution, accent coverage, or prosodic variance). If the filter systematically discards noisy, expressive, or low-resource-accent segments, the resulting 510 K hours could be easier than the raw pool yet still appear competitive on the chosen benchmarks. This is the single assumption whose failure would falsify the central “open data suffices” narrative.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript introduces Raon-OpenTTS-Pool, a 615K-hour aggregation of public English speech data, from which a model-based filtering pipeline yields the 510K-hour Raon-OpenTTS-Core subset. Diffusion transformer TTS models (0.3B–1B parameters) are trained on this core set. The 1B model is reported to match or exceed closed-source systems (Qwen3-TTS, CosyVoice 3) on Seed-TTS-Eval (WER 1.78%, SIM 0.749) and CV3-Hard-EN (WER 6.15%, SIM 0.775), with an additional robustness benchmark (Raon-OpenTTS-Eval) showing strong average WER/SIM and second-best CMOS. Data, filtering code, training code, and checkpoints are released.","tokens_in":2129,"tokens_out":583,"duration_ms":34911,"significance":"If the filtering step produces an unbiased high-quality subset that preserves diversity, the result would demonstrate that large-scale open data can reach parity with proprietary-scale training, a meaningful step toward reproducible TTS research. The explicit public release of the 510K-hour pool, the filtering pipeline, training code, and model checkpoints is a concrete strength that directly supports follow-on work.","major_comments":[{"comment":"Dataset curation section: The model-based filtering pipeline that reduces Raon-OpenTTS-Pool (615K hours) to Raon-OpenTTS-Core (510K hours) is described only at the level of “model-based pipeline.” No architecture for the filter model, decision thresholds, or post-filter statistics (speaker-embedding entropy, SNR distribution, accent coverage, prosodic variance) are supplied. Because the headline comparability claims rest on the assumption that this subset remains diverse and unbiased, the omission is load-bearing for the central “open data suffices” narrative.","section":"Dataset curation"},{"comment":"Training and evaluation sections: Hyperparameters for the 1B model (learning rate schedule, batch size, total steps, regularization) are not reported, nor are statistical tests or confidence intervals on the WER and SIM numbers. Without these, it is impossible to verify that the reported parity with Qwen3-TTS and CosyVoice 3 is robust rather than an artifact of a single run or favorable test conditions.","section":"Training and evaluation"}],"minor_comments":[{"comment":"Abstract and §4: The claim that comparison models were trained on “several million hours” would benefit from explicit citations or references to the known data scales of Qwen3-TTS and CosyVoice 3.","section":"Abstract"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the constructive comments and for acknowledging the significance of our open data and model releases. We address each major comment below and will revise the manuscript to enhance transparency and reproducibility.","responses":[{"response":"We agree that additional detail on the filtering pipeline would strengthen the manuscript. The current description is intentionally high-level, with the full implementation (including model architecture, thresholds, and processing steps) provided in the publicly released code. In the revision we will expand the Dataset Curation section with a concise description of the filter model, the decision thresholds employed, and quantitative post-filter statistics covering speaker-embedding diversity, SNR distribution, accent coverage, and prosodic variance. These additions will be drawn directly from the released pipeline and our internal analysis logs, preserving the original results while improving transparency.","revision_made":"yes","referee_comment":"[Dataset curation] Dataset curation section: The model-based filtering pipeline that reduces Raon-OpenTTS-Pool (615K hours) to Raon-OpenTTS-Core (510K hours) is described only at the level of “model-based pipeline.” No architecture for the filter model, decision thresholds, or post-filter statistics (speaker-embedding entropy, SNR distribution, accent coverage, prosodic variance) are supplied. Because the headline comparability claims rest on the assumption that this subset remains diverse and unbiased, the omission is load-bearing for the central “open data suffices” narrative."},{"response":"We acknowledge that explicit hyperparameter reporting and statistical analysis would allow readers to better assess robustness. In the revised manuscript we will add a dedicated training-details subsection (or table) listing the learning-rate schedule, batch size, total steps, optimizer settings, and regularization techniques used for the 1B model. We will also report results from multiple independent runs together with confidence intervals and the results of appropriate statistical tests comparing Raon-OpenTTS-1B against the closed-source baselines. These additions will be included without changing the reported point estimates.","revision_made":"yes","referee_comment":"[Training and evaluation] Training and evaluation sections: Hyperparameters for the 1B model (learning rate schedule, batch size, total steps, regularization) are not reported, nor are statistical tests or confidence intervals on the WER and SIM numbers. Without these, it is impossible to verify that the reported parity with Qwen3-TTS and CosyVoice 3 is robust rather than an artifact of a single run or favorable test conditions."}],"tokens_in":1768,"tokens_out":537,"duration_ms":43460,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The headline result is that a 1B-parameter DiT TTS model trained on 510K hours of filtered public English speech reaches 1.78% WER and 0.749 SIM on Seed-TTS-Eval and 6.15% WER / 0.775 SIM on CV3-Hard-EN, placing it ahead of other open baselines and close to some closed models trained on millions of hours. They also release the full 615K-hour pool, the filtered core, training code, checkpoints, and a new Raon-OpenTTS-Eval benchmark covering clean, noisy, in-the-wild, and expressive conditions where their model leads on average WER and SIM.","headline":"Open 510K-hour filtered dataset plus 1B DiT model hits competitive WER/SIM numbers on public benchmarks and releases everything, but the filtering step is described too lightly to fully trust the 'open matches closed' claim.","tokens_in":2639,"tokens_out":234,"would_cite":true,"duration_ms":16987,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":{"model":"grok-4.3","evidence":[],"headline":"TTS data curation + DiT training has no structural overlap with RS forcing chain","alignment":"orthogonal","rationale":"Paper's machinery (model-based filtering on DNSMOS/WER/Speech-Ratio, 615K-hour aggregation, DiT flow-matching TTS) lives in empirical speech synthesis. RS theorems (reality_from_one_distinction, Jcost uniqueness, phi-ladder constants, 8-tick/D=3 forcing, AbsoluteFloorClosure) derive physical constants and geometry from a single logical distinction; none of these appear or are paralleled here.","tokens_in":58041,"confidence":"high","tokens_out":135,"duration_ms":6357,"cache_read_input_tokens":32896,"cache_creation_input_tokens":0},"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"Aggregating and filtering public speech data enables TTS models that match those trained on millions of hours of proprietary data.","keywords":["text-to-speech","open datasets","diffusion transformer","speech synthesis","data filtering","TTS benchmarks","robust evaluation","public speech data"],"falsifier":"Reproducing the training of Raon-OpenTTS-1B on the released Core dataset and obtaining word error rates or speaker similarity scores that fall substantially below the reported values or those of the compared proprietary models would indicate the claim does not hold.","tokens_in":2872,"feed_emoji":"🔊","tokens_out":740,"duration_ms":64092,"temperature":0.7,"pith_summary":"This paper shows that public English speech corpora totaling 615K hours can be aggregated into Raon-OpenTTS-Pool and then filtered via a model-based pipeline into a 510K-hour high-quality Core subset. Training diffusion transformer models up to 1B parameters on this Core yields performance comparable to leading closed models on word error rate, speaker similarity, and robustness across clean, noisy, and expressive conditions. A sympathetic reader would care because the result points to a path for high-quality, reproducible text-to-speech that does not depend on restricted private datasets.","feed_headline":"Public speech data powers TTS models matching closed systems","feed_subtitle":"A 510K-hour filtered open dataset and 1B model achieve top rankings for accuracy and speaker similarity on standard and robustness tests.","key_machinery":"The model-based filtering pipeline that derives the high-quality Raon-OpenTTS-Core subset from the aggregated Raon-OpenTTS-Pool of 615K hours of public speech data.","core_discovery":"By creating Raon-OpenTTS-Pool from publicly available corpora and web recordings, applying model-based filtering to obtain Raon-OpenTTS-Core, and training DiT-based models on it, the work shows that Raon-OpenTTS-1B achieves a word error rate of 1.78% and speaker similarity of 0.749 on Seed-TTS-Eval while ranking first on both metrics for CV3-Hard-EN, matching the performance of models trained on several million hours of proprietary data.","pith_inferences":["The same aggregation and filtering approach could be tested on public corpora in languages other than English to create comparable open datasets.","Prioritizing filtered data quality over raw volume may prove useful for training other generative audio models beyond TTS.","Community extensions of the released resources could test whether further scaling or mixing with smaller additional datasets improves results.","The work leaves open whether unfiltered larger public pools could achieve similar performance if training compute is increased accordingly."],"forward_implications":["Open-weight TTS models can reach competitive levels of naturalness and accuracy without access to proprietary speech corpora.","Releasing the data pool, filtering pipeline, training code, and checkpoints enables full reproducibility of the results.","The new Raon-OpenTTS-Eval benchmark supports structured testing of TTS robustness in clean, noisy, in-the-wild, and expressive conditions.","Scaling DiT-based TTS models to 1B parameters with filtered public data produces strong results on both word error rate and speaker similarity."],"fun_headline_variants":["Open 510K-hour pool trains TTS matching closed models","1B TTS from public data matches closed models on WER SIM","Raon-OpenTTS-1B competes with closed TTS on public data","Public pool creates robust TTS rivaling million-hour models"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"The model-based filtering pipeline selects a high-quality subset from public data that maintains diversity and avoids introducing biases that harm TTS performance.","fun_headline_variants_meta":{"raw":{"variants":["Open 510K-hour pool trains TTS matching closed models","1B TTS from public data matches closed models on WER SIM","Raon-OpenTTS-1B competes with closed TTS on public data","Public pool creates robust TTS rivaling million-hour models"]},"model":"grok-4.3","cost_usd":0.008984,"raw_usage":{"total_tokens":4172,"prompt_tokens":942,"num_sources_used":0,"completion_tokens":71,"cost_in_usd_ticks":89837000,"prompt_tokens_details":{"text_tokens":942,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":3159,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":942,"tokens_out":71,"duration_ms":26715,"temperature":1.0,"reasoning_tokens":3159,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-05-21T02:28:42.792602+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"Reproducing the training of Raon-OpenTTS-1B on the released Core dataset and obtaining word error rates or speaker similarity scores that fall substantially below the reported values or those of the compared proprietary models would indicate the claim does not hold.","supporting_citations":[],"review_version":1}