{"id":"a714a605-5826-4fc4-92f4-c54025427f94","arxiv_id":"2607.06088","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":4,"one_line_summary":"A flow-matching speech separator with biometric best-of-N candidate selection and chunk-wise channel alignment achieves competitive separation metrics and the best downstream ASR/SV error rates among evaluated systems on Libri2Mix.","lead":"This paper builds a speech separation system using conditional flow matching, where a frozen speaker-encoder model defines output order during training and selects the best candidate among multiple stochastic generations at inference. A smart generalist might read it because it shows how generative AI techniques (best-of-N sampling, familiar from LLMs) can improve practical speech separation pipelines for voice assistants and transcription systems.","discovery_kind":"unclear","skeptic_critique":{"model":"glm-5.2","headline":"The best-of-N selection criterion (Eq. 3) and the EER evaluation metric both use Wav2Vec 2.0 embeddings, creating a selection-evaluation circularity that could inflate the headline SV results; the alternative-backend ablation is mentioned but no numbers are reported.","rationale":"The reader's CONDITIONAL verdict is appropriate. The paper presents a well-engineered system with a genuinely novel best-of-N biometric criterion and competitive results. The circularity concern is real but partially mitigated by two factors: (1) the ASR results (cpWER with Whisper V3) are not subject to the same circularity and show a substantial advantage for TUnet, and (2) Figure 2 demonstrates that biometric selection closely approaches SI-SDR oracle selection, suggesting the criterion tracks genuine separation quality rather than merely inflating embedding distances. However, the absence of reported numbers for the alternative-backend ablation leaves the EER claim insufficiently validated. The abstract's claim of 'all evaluated settings' is also an overclaim since Table 2 only covers the clean condition. These issues justify the CONDITIONAL verdict but do not warrant escalation to REJECT, as the core methodological contribution is sound and the ASR results provide independent support. My concern overlaps with but is more specific than the reader's weakest_assumption: the reader focused on embedding reliability under degradation, while the more directly load-bearing issue is the selection-evaluation circularity. Both point to the same underlying dependency on the Wav2Vec 2.0 encoder, so the adjustment is partial rather than a full agreement.","tokens_in":8147,"tokens_out":3215,"duration_ms":57810,"concrete_test":"Report the actual EER numbers from the mentioned ablation using ResNet-34 and DistillWhisper SV backends for evaluation (keeping Wav2Vec 2.0 for selection). If TUnet's EER advantage over SepReformer (full) and ConvUnet remains comparable in magnitude (e.g., within 20% relative), the circularity concern is resolved. If the advantage shrinks substantially or reverses, the headline SV claim is inflated. As a stronger variant: re-run selection using the SI-SDR oracle (already available in Figure 2) and evaluate EER with a non-Wav2Vec 2.0 backend; if the biometric and oracle selection paths converge under a different evaluation backend, the circularity is non-problematic.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that TUnet achieves the lowest downstream SV error rates (Table 2: 0.39% EER). The best-of-N biometric selection (Eq. 3) selects the candidate pair minimizing cos(η(ŝ_A), η(ŝ_B)) using a frozen Wav2Vec 2.0 speaker encoder η (§3.3). The EER evaluation in Table 2 is then computed using a Wav2Vec 2.0-based SV system (§4.2). This creates a direct circularity: the selection criterion optimizes for maximum inter-channel embedding distance in the same embedding family used to evaluate EER. A generated candidate with artifacts that push Wav2Vec 2.0 embeddings apart—without genuinely improving separation—would be selected by Eq. 3 and simultaneously score well on EER. The paper states that ablations with ResNet-34 and DistillWhisper-based SV backends show 'the same trend' (§5), but no numerical results are reported, making it impossible to assess whether the EER advantage and its magnitude persist when selection and evaluation use different embedding spaces. The cpWER results (Whisper V3 ASR) are less affected by this circularity, so the ASR portion of the claim is on firmer ground. The reader's weakest_assumption focused on embedding reliability on degraded outputs; the more directly load-bearing issue is this selection-evaluation circularity, which the reader noted in the rationale but did not foreground as the weakest assumption.","agreement_with_reader":"partial"},"referee_report":{"model":"glm-5.2","summary":"This paper proposes a conditional flow-matching approach to two-speaker speech separation. The method formulates separation as a conditional generation task in the complex STFT domain, using a frozen speaker encoder to define a canonical source order during training. At inference, a best-of-N biometric sampling criterion selects the candidate pair whose output-channel speaker embeddings are most dissimilar, and a chunk-level alignment procedure enables processing of long recordings. The Transformer U-Net (TUnet) variant is evaluated on Libri2Mix and compared against DiffSep, SepReformer (full and chunked), and MeanFlow-TSE in terms of SI-SDR, PESQ, ESTOI, downstream ASR (cpWER via Whisper V3), and downstream SV (EER via Wav2Vec 2.0). The authors report that TUnet achieves the best downstream cpWER and EER among evaluated systems on the clean condition.","tokens_in":9073,"tokens_out":1371,"duration_ms":188799,"significance":"The paper introduces a practical combination of ideas—flow-matching-based separation, biometric source ordering, best-of-N inference-time selection, and chunk-level alignment—that is well-motivated for real-world deployment. The best-of-N biometric selection criterion (Eq. 3) is a falsifiable, parameter-light inference-time strategy that does not require ground-truth references, and the paper provides an oracle SI-SDR-based upper bound for comparison (Figure 2). The downstream evaluation on both ASR and SV is a useful contribution beyond standard separation metrics. The approach is built on the NVIDIA NeMo Toolkit, which aids reproducibility. However, the significance of the headline SV results is tempered by a selection-evaluation circularity concern (see Major Comments) that is acknowledged but not numerically resolved.","major_comments":[{"comment":"§5, Table 2 and Eq. (3): The best-of-N selection criterion (Eq. 3) selects the candidate pair minimizing cos(η(ŝ_A), η(ŝ_B)) using a frozen Wav2Vec 2.0 speaker encoder η (§3.3). The EER evaluation in Table 2 is then computed using a Wav2Vec 2.0-based SV system (§4.2). This creates a direct selection-evaluation circularity: the selection criterion optimizes for maximum inter-channel embedding distance in the same embedding family used to evaluate EER. A generated candidate with artifacts that push Wav2Vec 2.0 embeddings apart—without genuinely improving separation—would be selected by Eq. 3 and simultaneously score well on EER. The paper states that ablations with ResNet-34 and DistillWhisper-based SV backends show 'the same trend' (§5, last paragraph), but no numerical results are reported, making it impossible to assess whether the EER advantage and its magnitude persist when selection和","section":null},{"comment":"Abstract and Table 2: The abstract claims 'lowest downstream automatic speech recognition and speaker verification error rates in all evaluated settings,' but Table 2 only reports the Libri2Mix clean condition. No downstream cpWER or EER results are shown for the 'both' condition. Table 1 reports separation metrics for both conditions, so the absence of downstream results for 'both' is a gap between the claim and the evidence. Either the claim should be qualified to match the evaluated settings, or downstream results for the 'both' condition should be added to Table 2.","section":null},{"comment":"§5, last paragraph: The ablation with alternative downstream backends (Parakeet ASR, ResNet-34 and DistillWhisper-based SV models) is mentioned in a single sentence but no numbers are reported anywhere. Given that this ablation directly addresses the selection-evaluation circularity concern for the SV results, it should be reported with at least summary numbers (e.g., EER for TUnet under ResNet-34 and DistillWhisper backends, with the same Wav2Vec 2.0-based selection). Without these numbers, the reader cannot verify that the SV gains are not an artifact of the shared embedding space between the selection encoder and the evaluation backend.","section":null}],"minor_comments":[{"comment":"§4.1: The separator amplitude (0.5) and length (1000 samples / 62.5 ms) are stated but the rationale for these specific values is not discussed. A brief note on sensitivity or justification would help reproducibility.","section":null},{"comment":"§4.2: The conditional dropout probability p_cond = 0.99 is reported, but it is unclear whether this means 99% of the time the condition is retained (1% dropped) or vice versa. The phrasing 'replacing c with zeros with probability 1−p_cond' in §3.2 suggests 1% dropout, which should be stated more clearly.","section":null},{"comment":"Table 1: Several baselines (DiffSep, SepReformer chunk) report only the 'clean' condition with dashes for 'both'. It would be useful to note whether these values are unavailable or not computed.","section":null},{"comment":"§3.2: The reference speaker embedding e_ref used for source ordering during training is mentioned but not fully specified. Is it a fixed global reference, a per-utterance reference, or sampled per batch? Clarifying this would aid understanding of the ordering procedure.","section":null},{"comment":"Figure 2: The axis labels and legend are small. The top/bottom and left/right panel organization should be more clearly labeled (e.g., 'clean' vs. 'both' as panel titles rather than relying on caption text).","section":null},{"comment":"§4.2: MeanFlow-TSE is described as addressing target speaker extraction, which is a different problem setting. The comparison is informative but the paper could more clearly flag this as a non-blind reference for readers unfamiliar with the distinction, perhaps by adding a footnote.","section":null},{"comment":"The paper would benefit from a brief discussion of computational cost: the best-of-N procedure with N=4 requires 4x generation per chunk. Reporting wall-clock or RTF comparisons with baselines would strengthen the practical deployment framing.","section":null}],"recommendation":"major_revision","confidential_remarks":"The selection-evaluation circularity is the central concern. The authors appear to have already run the alternative-backend ablations (they mention 'the same trend'), so resolving this may simply require reporting those numbers. If the alternative-backend results confirm the EER advantage, the paper could move to minor revision. I recommend asking the authors to include these numbers in a revision rather than treating this as a fundamental flaw. The ASR results (cpWER via Whisper V3) are less affected by the circularity and provide a more independent signal."},"author_rebuttal":{"model":"glm-5.2","summary":"We thank the referee for a careful and constructive review. The referee raises three major comments, all of which are legitimate. (1) The selection-evaluation circularity concern for the SV results is valid: the best-of-N selection criterion (Eq. 3) uses a Wav2Vec 2.0 speaker encoder, and the EER evaluation in Table 2 also uses a Wav2Vec 2.0-based SV system. We agree that numerical results from alternative SV backends are needed to resolve this concern and will add them. (2) The abstract overclaims by stating 'all evaluated settings' when Table 2 only covers the clean condition. We will either qualify the claim or add downstream results for the 'both' condition. (3) The ablation sentence mentioning ResNet-34 and DistillWhisper backends without numbers is insufficient given that it directly addresses the circularity concern. We will expand it into a proper table. All three points require revision, and we can address all of them.","responses":[{"response":"The referee correctly identifies a selection-evaluation circularity: the selection encoder η (Wav2Vec 2.0-based, Novoselov et al. 2022) and the EER evaluation backend (Wav2Vec 2.0, Khmelev et al. 2025) are from the same model family. We agree this is a genuine concern and that the current manuscript does not provide sufficient evidence to rule it out. In the revision, we will add a new table reporting EER for TUnet (and key baselines) under two alternative SV backends—ResNet-34 and DistillWhisper (Khmelev et al. 2026)—while keeping the Wav2Vec 2.0-based selection criterion unchanged. This cross-backend evaluation directly tests whether the EER advantage persists when the selection and evaluation embedding spaces are decoupled. We will also add an explicit discussion of the circularity concern in §5, acknowledging the shared model family and explaining why the cross-backend results mitigate it. We note that the ASR results (cpWER via Whisper V3) are not subject to this circularity, since the selection criterion operates purely on speaker embeddings and the ASR backend is a separate model.","revision_made":"yes","referee_comment":"§5, Table 2 and Eq. (3): The best-of-N selection criterion (Eq. 3) selects the candidate pair minimizing cos(η(ŝ_A), η(ŝ_B)) using a frozen Wav2Vec 2.0 speaker encoder η (§3.3). The EER evaluation in Table 2 is then computed using a Wav2Vec 2.0-based SV system (§4.2). This creates a direct selection-evaluation circularity: the selection criterion optimizes for maximum inter-channel embedding distance in the same embedding family used to evaluate EER. A generated candidate with artifacts that push Wav2Vec 2.0 embeddings apart—without genuinely improving separation—would be selected by Eq. 3 and simultaneously score well on EER. The paper states that ablations with ResNet-34 and DistillWhisper-based SV backends show 'the same trend' (§5, last paragraph), but no numerical results are reported, making it impossible to assess whether the EER advantage and its magnitude persist when selection和"},{"response":"The referee is correct that the abstract's claim of 'all evaluated settings' is not supported by Table 2, which only covers the clean condition. This is an overclaim relative to the evidence presented. We will resolve this by adding downstream cpWER and EER results for the 'both' condition to Table 2, so that the claim is fully supported. If computational constraints prevent completing the 'both' condition downstream evaluation in time for revision, we will instead qualify the abstract to say 'in the clean condition' and note the gap explicitly. Our preference is to add the results.","revision_made":"yes","referee_comment":"Abstract and Table 2: The abstract claims 'lowest downstream automatic speech recognition and speaker verification error rates in all evaluated settings,' but Table 2 only reports the Libri2Mix clean condition. No downstream cpWER or EER results are shown for the 'both' condition. Table 1 reports separation metrics for both conditions, so the absence of downstream results for 'both' is a gap between the claim and the evidence. Either the claim should be qualified to match the evaluated settings, or downstream results for the 'both' condition should be added to Table 2."},{"response":"We agree. Mentioning these ablation results without reporting any numbers is insufficient, especially given that they directly address the circularity concern raised in the first major comment. In the revision, we will expand the single sentence in §5 into a proper table (or at minimum a clearly structured paragraph with full numerical results) reporting EER for TUnet and relevant baselines under ResNet-34 and DistillWhisper SV backends, with the Wav2Vec 2.0-based selection criterion held fixed. We will also report cpWER under the Parakeet ASR backend. This will allow readers to verify that the downstream gains are not artifacts of the shared embedding space.","revision_made":"yes","referee_comment":"§5, last paragraph: The ablation with alternative downstream backends (Parakeet ASR, ResNet-34 and DistillWhisper-based SV models) is mentioned in a single sentence but no numbers are reported anywhere. Given that this ablation directly addresses the selection-evaluation circularity concern for the SV results, it should be reported with at least summary numbers (e.g., EER for TUnet under ResNet-34 and DistillWhisper backends, with the same Wav2Vec 2.0-based selection). Without these numbers, the reader cannot verify that the SV gains are not an artifact of the shared embedding space between the selection encoder and the evaluation backend."}],"tokens_in":8200,"tokens_out":1266,"duration_ms":206282,"standing_objections":[]},"desk_editor":{"model":"glm-5.2","letter":"The main thing to know: this paper combines conditional flow matching with a biometric best-of-N selection criterion for speech separation, and the downstream ASR results (3.84% cpWER on Libri2Mix clean) are genuinely strong, especially compared to chunked SepReformer (12.90%). The best-of-N idea — selecting the candidate pair whose speaker embeddings are most dissimilar — is a clean, practical contribution to inference-time scaling for generative separation. The chunk-level alignment via accumulated embedding centroids is also a sensible solution to the permutation problem in long-form processing. Figure 2 showing biometric selection approaching oracle SI-SDR selection is the most convincing result in the paper. Credit is earned for the engineering: the pipeline is coherent and the experimental setup is reasonable, with appropriate baselines (DiffSep, SepReformer, MeanFlow-TSE) and meaningful augmentation (reverb, WHAM!, MUSAN, CHiME-8 noise). The framing of separation as structured denoising (concatenating sources with a non-speech separator) is elegant. The ASR results are on firmer ground because Whisper V3 is independent of the selection criterion. The cpWER gains are real and practically meaningful. Now the soft spots. The stress-test concern about selection-evaluation circularity is the most important issue. The best-of-N criterion (Eq. 3) selects candidates maximizing inter-channel Wav2Vec 2.0 embedding distance, and the EER evaluation uses a Wav2Vec 2.0-based SV system. A candidate with artifacts that push embeddings apart would be selected and simultaneously score well on EER. The paper mentions ablations with ResNet-34 and DistillWhisper SV backends showing 'the same trend,' but reports no numbers — this is insufficient. The reader's concern about embedding reliability on degraded outputs is valid but secondary; the circularity is the load-bearing issue. The abstract overclaims 'lowest error rates in all evaluated settings' when Table 2 only reports the clean condition. The 'both' condition downstream results are absent. This should be corrected. Evaluation is limited to Libri2Mix only, which constrains generalizability claims. No code or data release is mentioned. The free parameters (N, separator amplitude, p_cond, STFT parameters) are not extensively ablated, though Figure 2 covers N adequately. These are fixable issues. The core contribution — biometric best-of-N for generative separation — is new and the ASR evidence supports it. The SV evidence needs the alternative-backend numbers to be credible. This paper is for researchers working on generative speech separation and practical deployment of separation systems. It deserves a serious referee who should require: (1) numerical results for the alternative-backend SV ablation, (2) downstream evaluation on the 'both' condition, and (3) a correction to the abstract's scope claim.","headline":"Well-engineered flow-matching separator with a novel best-of-N biometric selection criterion; competitive downstream results, but the SV evaluation has a selection-evaluation circularity that needs addressing.","tokens_in":9187,"tokens_out":671,"would_cite":false,"duration_ms":59706,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"glm-5.2","headline":"Flow matching plus biometric selection cuts speech separation errors","keywords":["speech separation","flow matching","best-of-N sampling","speaker verification","biometric selection","permutation ambiguity","chunk-wise processing","conditional generation"],"falsifier":"Run the system on mixtures where one speaker is heavily masked or where the SNR between sources is highly asymmetric (e.g., -5 dB). If the speaker encoder produces unreliable embeddings for the dominated speaker, the best-of-N selection criterion (which relies on inter-channel embedding dissimilarity) and the chunk alignment (which relies on embedding clustering) would both degrade, potentially producing worse downstream ASR and speaker verification than a deterministic baseline that does not depend on biometric feedback.","tokens_in":8424,"feed_emoji":"🗣","tokens_out":1295,"duration_ms":305944,"temperature":0.7,"pith_summary":"The paper tackles three practical problems in single-channel speech separation: the source permutation ambiguity (which output channel corresponds to which speaker), the run-to-run variability inherent in generative models (where each stochastic sample yields a slightly different separation), and the difficulty of processing long recordings that must be split into chunks. The authors propose a conditional flow-matching separator that generates both separated sources simultaneously in a fixed, speaker-ordered layout. A frozen pretrained speaker encoder (based on Wav2Vec 2.0) is used at three critical points: during training to define a consistent source ordering, at inference to select the best of N stochastically generated candidate separations by choosing the one whose two output channels are most speaker-distinct, and across chunks to track and align speaker identity over time. On the Libri2Mix benchmark, the Transformer U-Net variant achieves 3.84% concatenated permutation word error rate for downstream speech recognition and 0.39% equal error rate for speaker verification, outperforming all evaluated baselines including SepReformer (which degrades substantially in chunked mode) and MeanFlow-TSE (which requires target-speaker reference information). The separation quality metrics (SI-SDR, PESQ, ESTOI) are competitive rather than leading, but the downstream ASR and speaker verification errors are the lowest among all compared systems, suggesting the method produces separations that are more useful for practical speech processing even when raw signal-level metrics do not top the leaderboard.","feed_headline":"Flow matching plus biometric selection cuts speech separation errors","feed_subtitle":"Generating multiple candidate separations and picking the one with most distinct speakers yields lowest downstream ASR and verification","key_machinery":"The method has four interlocking components. First, conditional flow matching formulates two-speaker separation as a denoising task in the complex STFT domain, where the target is an ordered concatenation of both sources separated by a fixed non-speech spacer and the condition is the mixture duplicated in the same layout. Second, a frozen Wav2Vec 2.0-based speaker encoder extracts embeddings from each source to define a canonical ordering during training (the source closer to a reference embedding is placed first). Third, at inference, N independent stochastic separations are generated per chunk and the pair with the most dissimilar inter-channel speaker embeddings is selected. Fourth, for長長","core_discovery":"The central discovery is that a speaker-embedding model can serve as an effective reference-free verifier for generative speech separation. By generating N candidate separations from the same mixture and selecting the candidate whose two output channels have the most dissimilar speaker embeddings (lowest cosine similarity), the system approximates oracle-quality selection (which requires ground-truth references) without needing them. The biometric best-of-N criterion closely tracks the oracle SI-SDR-based selection curve as N increases, with both saturating around N=4. This means the sampling variance of flow-matching models, typically viewed as a drawback, can be exploited: more samples can","pith_inferences":["The reliance on a frozen speaker encoder creates a coupling between separation quality and verification quality: if the speaker encoder produces poor embeddings on degraded outputs, the selection, ordering, and alignment mechanisms all fail simultaneously with no fallback. This could be tested by injecting controlled artifacts into separated outputs and measuring embedding reliability.","The best-of-N biometric criterion implicitly assumes that speaker distinctiveness correlates with overall separation quality. This could be tested on mixtures with three or more speakers, where pairwise dissimilarity may not select the globally best separation.","The 1-second chunk size with 0.5-second hop is likely tuned for the Libri2Mix distribution; longer chunks might improve biometric embedding reliability at the cost of latency, and the optimal chunk size may vary with speaker overlap patterns and noise conditions.","The saturation of best-of-N gains around N=4 suggests diminishing returns from additional inference compute, but this saturation point may shift with more challenging acoustic conditions or more speakers, where a single good candidate is less likely among few samples."],"forward_implications":["Generative separators with biometric best-of-N selection could replace deterministic separators in practical ASR and speaker verification pipelines, especially for long-form audio where chunked processing is required.","The principle of using a domain-specific pretrained encoder (here, speaker recognition) as an inference-time verifier for generative model outputs could transfer to other structured generation tasks where output quality is hard to measure directly but downstream task performance is the real objective.","The gap between competitive signal-level metrics (SI-SDR, PESQ) and best-in-class downstream metrics (cpWER, EER) suggests that standard separation benchmarks may under-reward perceptual and semantic qualities that matter for real applications.","Chunk-wise processing with biometric channel tracking provides a path to deploying flow-matching separators on streaming or real-time audio, where full-utterance models like SepReformer are inapplicable."],"fun_headline_variants":["Biometric best-of-N selection turns sampling variance into a separation advantage","Picking the most distinct speaker pair from N candidates improves flow-matching separation","Speaker dissimilarity scoring enables oracle-free best-of-N speech separation","Flow-matching separation improved by selecting candidates with most dissimilar speakers","Reference-free biometric selection narrows the gap to oracle speech separation"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The frozen speaker encoder is assumed to produce reliable speaker embeddings from partially separated, potentially degraded speech outputs at inference time. The entire pipeline (source ordering during training, best-of-N candidate selection, and chunk-level channel alignment) depends on these embeddings being meaningful even when the generated separation quality is poor. If the speaker encoder fails on noisy or artifact-laden outputs, the ordering, selection, and alignment","fun_headline_variants_meta":{"raw":{"variants":["Biometric best-of-N selection turns sampling variance into a separation advantage","Picking the most distinct speaker pair from N candidates improves flow-matching separation","Speaker dissimilarity scoring enables oracle-free best-of-N speech separation","Flow-matching separation improved by selecting candidates with most dissimilar speakers","Reference-free biometric selection narrows the gap to oracle speech separation","Sampling variance exploited: best-of-N biometric selection improves separation","No reference needed: speaker embeddings verify flow-matching separation candidates","Biometric candidate selection tracks oracle performance for flow-matching separation","Two-source flow matching with biometric best-of-N cuts downstream ASR and SV errors"]},"model":"glm-5.2","effort":"high","cost_usd":0.0,"raw_usage":{"total_tokens":1143,"prompt_tokens":453,"completion_tokens":690,"prompt_tokens_details":null},"tokens_in":453,"tokens_out":690,"duration_ms":33261,"temperature":1.0,"reasoning_tokens":635,"cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-08T16:50:01.210715+00:00","model_set":{"reader":"glm-5.2"},"falsifier":"Run the system on mixtures where one speaker is heavily masked or where the SNR between sources is highly asymmetric (e.g., -5 dB). If the speaker encoder produces unreliable embeddings for the dominated speaker, the best-of-N selection criterion (which relies on inter-channel embedding dissimilarity) and the chunk alignment (which relies on embedding clustering) would both degrade, potentially producing worse downstream ASR and speaker verification than a deterministic baseline that does not depend on biometric feedback.","supporting_citations":[],"review_version":1}