LoopGen: Training-Free Loopable Music Generation

LOOPGEN: TRAINING-FREE LOOPABLE MUSIC GENERATION Davide Marincione⋆Giorgio Strano⋆Donato Crisostomi Roberto Ribuoli Emanuele Rodolà Sapienza University of Rome {marincione, strano}@di.uniroma1.it ABSTRACT Loops–short audio segments designed for seamless repetition–are central to many music genres, particularly those rooted in dance and electronic styles. However, current generative music models struggle to produce truly loopable audio, as generating a short waveform alone does not guarantee a smooth transition from its endpoint back to its start, often resulting in audible discontinuities. We address this gap by modifying a non-autoregressive model (MAGNeT) to generate tokens in a circular pattern, letting the model attend to the beginning of the audio when creating its ending. This inference-only approach results in generations that are aware of future context and loop naturally, without the need for any additional training or data. We evaluate the consistency of loop transitions by computing token perplexity around the seam of the loop, observing a 55% improvement. Blind listening tests further confirm significant perceptual gains over baseline methods, improving mean ratings by 70%. Taken together, these results highlight the effectiveness of inference-only approaches in improving generative models and underscore the advantages of non-autoregressive methods for contextaware music generation. github.com/gladia-research-group/loopgen gladia-research-group.github.io/loopgen-demo 1. INTRODUCTION Loops play a critical role in music production across a broad range of genres, from hip-hop to electronic dance music. By definition, a loop is a segment of audio that can be repeated indefinitely without noticeably jarring transitions between consecutive repetitions. These short segments function as building blocks in many compositions, providing rhythmic and harmonic foundations that can be layered, remixed, and manipulated. Indeed, entire online platforms (e.g., Splice 1) revolve around sharing and curating loops, underscoring their commercial and creative significance in contemporary music-making. However, despite their ubiquity in practice, loops remain an underexplored challenge for generative music models. The primary issue lies in the disconnect between generating a short audio sample and ensuring that it loops correctly. Many existing generative approaches focus on ⋆denotes equal contribution. 1https://splice.com/ MAGNeT window main tile left padding right padding                  Figure 1. Our proposed circular padding framework for loopable sample generation. producing samples that sound coherent when played from start to finish [1,2,3,4,5,6], but they do not explicitly consider the transition point from the end of the sample back to its beginning. As a result, naive repetition of these segments often yields abrupt discontinuities, limiting their practical utility for musicians and producers who rely on seamless repetition. In this paper, we introduce a loop-aware generation framework that modifies the iterative inference of a nonautoregressive (NAR) model to produce seamless loops. Concretely, we adopt a circular padding strategy, replicating partial portions of the loop at both ends of the generation window, so that the model attends to the loop’s beginning while generating its ending (Figure 1). This ensures a smooth endpoint-to-onset transition, effectively creating “bridging tokens” that align the tail of the sample with its onset. Our method can be used in two ways: (1) to generate entire loopable segments from scratch, or (2) to refine the end of an existing audio sample so that it loops seamlessly. Additionally, we implement a beat-aware technique that constrains the total length of the loop to align with musical bars, further promoting coherent repetition. To evaluate this approach, we propose a new perplexity-based metric that quantifies the harshness of the cut at the seam of the loop. Intuitively, if the loop boundary is truly coherent, then it should not be perceived as irregular or dissonant, neither for a human listener, nor for an audio model, as a well-trained network should roughly match human perception. Our contributions are: •Loop-Aware Generation via Audio Tiling: We propose a new inference procedure that can be applied to a NAR music transformer, such as MAGNeT, 536 to create seamlessly loopable audio samples. We call this method, and the resulting model, LoopGen. •Perplexity-Based Seamlessness Metric: We introduce a metric to quantify the quality of loop boundaries, retrieving the entropy in the “seam” region of a track. •Empirical Validation and Code Release: We show that our system yields superior results according to both quantitative metrics and human listening tests, and we release our code to foster future research on the generation of musical loops. 2. RELATED WORK Recent advances in music generation leverage large-scale transformer-based architectures, which have displaced traditional recurrent neural networks for long-range sequence modeling. Pioneering systems like MuseNet [7] and Music Transformer [8] showed that attention-based models [9] could capture rich compositional structure in symbolic formats. More recently, state-of-the-art audio transformer models such as MusicGen [1] and [3,10,11,12], have demonstrated high-quality generation of waveforms, capable of handling minutes-long clips conditioned on text or user-provided melodies. Another wave of expressive and accurate models has come with the advent of diffusion models [13] such as AudioLDM [4] and [5,6,14,15,16]. Their application has also reached audio and music and, in this, they are giving high quality results on-par with the transformer models. Parallel decoding has emerged as a promising alternative to speed up generation. MAGNeT [2] employs a single non-autoregressive transformer, such as those used in NLP tasks [17,18], to predict masked audio tokens iteratively, showing that a well-designed masking and rescoring strategy can close the quality gap with autoregressive baselines at a fraction of the inference cost. VampNet [19], another non-autoregressive approach, introduces inpainting capabilities and partial rewriting to refine music segments, including short repeated “vamps,” demonstrating promise for loop-centric workflows. Likewise, SoundStorm [20] applies a bidirectional transformer on semantic tokens for efficient speech and music synthesis, further illustrating the viability of non-autoregressive methods for audio. Loopable music remains comparatively underexplored. LoopNet [21] specifically targets the generation of seamless music loops, but it is tied to a limited dataset of loops which falls short of the general-purpose of free-form approaches. Other work has focused on symbolic loops in MIDI [22,23], proposing architectures that ensures segments are musically consistent when repeated. However, these methods are intrinsically different from raw audio tokens; MIDI loops require explicit pitch and instrument representations, which do not transfer to audio-generation tasks. Recently, DITTO [24] introduced an inferencetime optimization that allows fine-grained control, including looping, over text-to-music diffusion models. While DITTO is notable for its high output quality, it requires memory comparable to a full fine-tuning, and it slows down inference by a factor of ∼100×. Finally, loopable media generation is being tackled in computer vision with tiling techniques. Models like TileGAN [25] and [26,27] synthesize textures or images that repeat edge-to-edge without seams. While these visual approaches share the overarching idea of boundary alignment, they do not directly address audio continuity or musical structure. In this paper, we build on MAGNeT’s nonautoregressive design to propose an inference-time approach for loopable music generation, avoiding additional training or data requirements. By treating time in a “circular” manner, our method enforces continuity at the loop boundary, substantially improving perceptual seamlessness in raw audio. 3. BACKGROUND 3.1 MAGNeT’s inference Unlike typical NAR models, MAGNeT’s inference does not emit all output tokens in a single inference pass. Instead, it develops the audio clip iteratively. In particular, at each iteration, MAGNeT: 1. Generates logits for each empty token in the sequence. 2. Samples a value for each token. 3. Selects the highest scoring tokens, and marks them as fixed. 4. Re-empties the remaining non-fixed tokens and, if no empty tokens are left, terminates; otherwise, it starts the next iteration. Following [2], we use MAGNeT’s own logits to select the tokens for the first iteration. MAGNeT’s inference can be viewed as a generalization of autoregressive inference: rather than receiving a continuous sequence of tokens and outputting the next token, MAGNeT operates on a set of empty and non-empty tokens, filling multiple empty positions in each iteration. Thanks to its non-causal selfattention, MAGNeT can condition its outputs on both past and future tokens, ensuring coherent generation across boundaries. This property makes MAGNeT (and similar non-autoregressive models) well-suited for loop creation, because the model can naturally attend to the loop’s start while generating its end, thereby facilitating a smoother, more seamless transition. 3.2 Rescoring MAGNeT [2] also proposes a variant that linearly interpolates the probabilities given by its own logits, with those of another audio model, such as MusicGen [1], to calculate the scores for the selection procedure. This results in a trade-off between higher quality and increased computational cost, as calculating another model’s probabilities requires running it alongside MAGNeT. Proceedings of the 26th ISMIR Conference, Daejeon, Korea, September 21-25, 2025 537 3.3 Hybrid MAGNeT As noted in [2], when presented with short audio tracks, MAGNeT produces continuations which, on average, sound better than samples created from scratch. Knowing this, we test both generating samples from scratch, and continuations of clips produced with MusicGen. 4. METHOD While our framework is, in principle, applicable to any NAR model that generates audio iteratively, we choose MAGNeT [2] as our base system because it is currently the state-of-the-art in NAR music generation. We adapt MAGNeT’s iterative inference to create a “circular” context around the central segment of tokens that will form our final loop. By replicating partial portions of this loop segment at the beginning and end of the generation window, MAGNeT can attend to the loop’s start when predicting its end, and vice versa. We refer to the central segment as the main loop tile. 4.1 Iterative overview MAGNeT generates audio tokens in several iterations. Each iteration partially fills an overall generation window of length L. We isolate a specific subrange of length cnear the center of this window to become our main loop tile. The remaining space on the left and right is filled with copies of the tile’s end or beginning, respectively, thus forming a circular context. 4.2 Inference algorithm At initialization, we start with an empty (or partially filled) window of length L. In the middle of this window, we mark out cconsecutive positions as the main loop tile. 1. Filling the Context. Before calling MAGNeT, and before each inference step, we copy: • The ending of the main tile into the left side of the window, so that the first tokens of the tile can “see” what happens at the end of it. • The beginning of the main tile into the right side of the window, so that the last tokens of the tile can “see” its start. This ensures a fully circular arrangement: the model effectively observes how the loop’s end meets its beginning. 2. MAGNeT Inference. We run MAGNeT on the entire window of length L. Because MAGNeT uses non-causal (bidirectional) attention, tokens in the main tile can be conditioned on both the left-side copy (its own end) and the right-side copy (its own start). 3. Token Selection. At the end of each iteration, only tokens within the main tile are considered for finalizing. We keep those that MAGNeT assigns the highest probability (e.g., top-kor threshold-based), marking them as fixed (i.e., no longer empty in subsequent iterations). The rest are reset to empty. main tile left padding right padding                  MAGNeT MAGNeT ... top K top K Figure 2. Diagram of our approach. The central main tile represents the final audio segment to be looped. At each inference step, only the top-ksamples are maintained and reflected in the tiles. This circular padding lets MAGNeT attend to both the start and end of the tile simultaneously, ensuring a smooth transition at the loop boundary. 4. Repeat until completion. We move on to the next inference iteration, going back to step 2, until the entirety of the main tile is filled. 5. Extract the final loop. Once the iteration limit is reached or all main-tile tokens are fixed, the algorithm stops. The central ctokens (our main loop tile) are extracted as the final result. Repeating this tile end-to-start yields a seamless loop. 4.3 Hybrid variant MAGNeT often produces higher-quality audio when continuing from a given prompt rather than generating entirely from scratch [2]. To take advantage of this, we first generate an audio segment Cwith MusicGen, empirically set to half the desired final clip length. For instance, if the final clip is intended to last 10s, we let MusicGen produce the first 5s and then provide these tokens as a partially filled main tile. This approach forces the model to generate a coherent continuation of the high-quality prompt, ensuring Proceedings of the 26th ISMIR Conference, Daejeon, Korea, September 21-25, 2025 538 the ending transitions seamlessly to the beginning. Empirically, we observe that this hybrid version surpasses samples generated without an audio prompt in terms of semantic variety, objective audio quality, and musical coherence. 4.4 Signature-aware length control A well-formed loop often sounds most musical when it aligns with full bars (e.g., 2 or 4 bars of consistent tempo). Generating loops of arbitrary length may create awkward breaks if, for instance, the tempo does not fit integer bar divisions. To mitigate this, we use the current state-of-the-art beatextraction system, beat_this [28] on the initial audio prompt C, to identify: • The average duration between beats, δ≈60/BPM. • The median number of beats per bar, µ. We use these to compute the duration of a bar that the prompt Cimplies. We want the duration of the entire loop lto be a) an exact multiple (or submultiple) of the duration of a bar; b) constrained in a time interval [α, β]. To achieve this, we repeatedly double or halve the initial candidate length luntil it fits these constraints. Algorithm 1 Beat Alignment algorithm Require: Audio clip C, min/max duration αand β, preferred number of bars n B, D ←detected beats/downbeats in beat_this(C) δ←median time elapsed between Bs▷akin to 60 BPM µ←median #beats between Ds▷bar of the clip l←nµδ ▷ duration of nbars while l < α ∨l > β do if l < α then l←2l else l←l 2 end if end while if l∈[nµδ 4,4nµδ]then ▷Too far from nbars? return l ▷ Return if sufficiently close else return ∅▷Otherwise abort (try another C) end if When used in combination, tiled generation and the beat alignment Algorithm 1produce loops that not only have smooth seam transitions but also respect musical structure. This results in clips that are more naturally loopable for applications like music production, live performance, or any setting in which tightly aligned repeating segments are required. 5. EVALUATION METRICS When assessing looped music, a clip that sounds fine in a single pass may still have an abrupt transition when it repeats. Standard metrics such as FAD [29], which measure the overall distributional similarity between generated −1.00 −0.75 −0.50 −0.25 0.00 0.25 0.50 0.75 1.00 Time (s) 0 2 4 6 8 10 Cross-Entropy MusicGen Naive MAGNeT Naive MusicGen Beat Aligned Naive MAGNeT Beat Aligned Naive MAGNeT Tiled MAGNeT Hybrid Tiled LoopGen Figure 3. Average cross entropy of MusicGen around the seam (highlighted with a dashed line) for different models/variants. and real audio using neural embeddings, may miss such artifacts. Because FAD operates over full-length clips, it can overlook localized issues such as discontinuities at the loop boundary. To address this, we propose a metric that focuses directly on the transition seam. 5.1 Seam perplexity To automatically assess the continuity of the loop around the seam, we adapt the idea of perplexity from language modeling. Let us assume that we have a well-trained music generation model (such as MusicGen) that can estimate probabilities for each token (or audio frame) in a music clip. While traditional perplexity sums over all tokens in a clip, we focus exclusively on the seam, that is, the transition point where the end meets the beginning, where loop artifacts are most likely to occur. 5.1.1 Cross-entropy and perplexity First recall that, for a sequence X= (x1, x2, . . . , xT), the average cross-entropy H(X)is: H(X) = −1 T T X i=1 ln M(xi).(1) Intuitively, if Massigns higher probability to each token, the cross-entropy will be smaller, indicating better alignment between model and data. From cross-entropy, we derive the perplexity P(X), a standard measure of how well a model predicts a sequence: P(X) = expH(X)= exp −1 T T X i=1 ln M(xi).(2) A lower perplexity value indicates that Mfinds the sequence more predictable (or more likely). Proceedings of the 26th ISMIR Conference, Daejeon, Korea, September 21-25, 2025 539 5.1.2 Seam perplexity While global perplexity focuses on the entire clip, loop artifacts, as in Figure 3, occur specifically at the boundary where the clip wraps around. To isolate how the model perceives that transition, we compute seam perplexity on a short window around the boundary. Let us be given Ngenerated clips {X(k)}. Each X(k) has length T, and we identify a seam boundary at index b(k). We then define a window of size Wimmediately following b(k), i.e., the tokens X(k) seam =x(k) i:i∈[b(k), b(k)+W−1].(3) The average cross-entropy of the seam tokens in X(k)is: Hseam(X(k)) = −1 W b(k)+W−1 X i=b(k) ln Mx(k) i.(4) Finally, the seam perplexity is the exponential of the mean seam cross-entropy across all Nclips: Seam Perplexity = exp 1 N N X k=1 HseamX(k).(5) A low seam perplexity indicates that the seam is “easy” for a strong reference model to predict, suggesting a smooth transition. Conversely, a high value suggests abrupt discontinuities or other artifacts at the loop boundary. 6. EXPERIMENTS In the following, all generated tracks are conditioned with the same set of 100 textual prompts, and MAGNeT’s iterations are set to ⟨100,50,10,10⟩for each of the 4 codebooks from EnCodec [30] respectively. Textual prompts are generated automatically via a LLM, some of them are (e.g.): (1) “A high-energy EDM track with a powerful drop and sidechain compression” (2) “An Irish folk dance tune with energetic fiddle and bodhrán drum” 6.1 Evaluating models 6.1.1 Baselines For both MAGNeT and MusicGen, two baseline solutions are formulated: (i) Naive, a sample is generated and repeated, without further processing, and (ii) Beat-Aligned (BA) Naive, a sample is generated, ran through Algorithm 1to cut them at a musically-valid length, and repeated. 6.1.2 Our techniques From the contributions of this paper, three models are evaluated: (i) Tiled, samples are generated via the tiled generation technique described in Section 4.2, (ii) Hybrid Tiled, samples are generated with the same tiled technique, but starting from an audio prompt generated by MusicGen (ref. Section 4.3), and (iii) Beat Aligned Tiled, which uses the same technique as Hybrid Tiled, but with the additional application of the beat-alignment algorithm described in Section 4.4. This latter variant is our best performing model, and what, going forward, we call LoopGen. For each variant, both Seam Perplexity and fadtk’s [31] FAD (with embeddings from both VGGish[32] and CLAP [33] over the FMA-Pop [34] dataset) are computed. 6.2 Hyperparameter search The most important hyperparameters for the samples’ quality we identify are classifier-free guidance (λ) and the rescoring coefficient (ω). The former controls how much a model should adhere to the conditioning information given (in our case, the textual prompt), instead of following the emerging sample. In MAGNeT’s original paper, the authors find that the best FAD is reached with λ= 10.0(linearly decreasing to λ= 1.0as the iterations pass) but, as the tiling constraint might increase the contextual information that the model can gain from the input, we verified that a lower coefficient translates into more organic generations. The rescoring coefficient ω, instead, controls the interpolation coefficient introduced in Section 3.2. When ω= 0, rescoring is not applied, when 1, only MusicGen’s probabilities are used. We test therefore our algorithm with multiple coefficients ranging from 0 to 1. A reasonable value for the cfg was chosen to be λ= 5.0; on the other hand, the rescoring was chosen through a thorough search conducted on both MAGNeT Tiled and MAGNeT Hybrid Tiled, generating 100 10-seconds samples for each model. Our results, presented in Table 1, empirically show the best rescoring to be ω= 0.5. It is worth noting that the Hybrid version of the model consistently achieves better FAD scores, but worse perplexity. The better FAD score can be clearly attributed to the initial prompt generated by MusicGen, which consistently surpasses MAGNeT’s audio quality. This hybrid combination of different models is also the reason for the increase in perplexity, since the final generation consists of a concatenation of tokens sampled from different distributions. Model Variant ωFADvggish (↓)Seam Perplexity (↓) MAGNeT Tiled 0.0 3.05 23.88 ±5.40 MAGNeT Tiled 0.25 3.22 25.17 ±5.35 MAGNeT Tiled 0.50 3.51 18.15 ±3.53 MAGNeT Tiled 0.75 3.97 24.55 ±5.43 MAGNeT Tiled 1.0 4.35 25.42 ±4.86 MAGNeT Hybrid Tiled 0.0 2.97 39.30 ±7.21 MAGNeT Hybrid Tiled 0.25 2.99 47.72 ±10.11 MAGNeT Hybrid Tiled 0.50 2.98 44.42 ±9.33 MAGNeT Hybrid Tiled 0.75 3.00 43.93 ±8.39 MAGNeT Hybrid Tiled 1.02.93 41.74 ±9.05 Table 1. Rescoring experiments (λ= 5.0) Proceedings of the 26th ISMIR Conference, Daejeon, Korea, September 21-25, 2025 540 6.3 Final results Below, we present our final results across six baselines and our three novel models, generated with the same previous 100 textual prompts, but 30 seconds long. Tiling models, using the technique described in Section 4.2, exhibit significantly lower Seam Perplexity compared to their nontiled counterparts, though at the cost of a weaker FAD score. However, LoopGen, leveraging both the Hybrid approach (Section 4.3) and Algorithm 1, achieves the best FAD score among all models. This improvement comes with a slight increase in Seam Perplexity, as previously discussed. Despite this minor trade-off in perplexity, LoopGen substantially outperforms baseline solutions, offering a more musically pleasing output due to its alignment with rhythmically meaningful cut points (Algorithm 1). This results in tracks that maintain better musical coherence compared to the standard Tiled model. Table 2presents the evaluation metrics, and the distribution of Seam Perplexity values is visualized in Figure 4. Model Variant FADvggish (↓)FADCLAP (↓)Seam Perplexity (↓) MAGNeT Vanilla 3.36 0.33 — MAGNeT Naive 3.36 0.35 1549.06 ±556.03 MAGNeT Beat Aligned Naive 3.34 0.34 153.22 ±47.69 MusicGen Vanilla 2.81 0.32 — MusicGen Naive 2.81 0.33 2512.39 ±903.16 MusicGen Beat Aligned Naive 2.86 0.33 507.07 ±163.67 MAGNeT Tiled 4.30 0.51 56.17 ±11.78 MAGNeT Hybrid Tiled 2.98 0.33 94.41 ±25.77 MAGNeT LoopGen 2.80 0.31 84.29 ±22.66 Table 2. Main experiments’ evaluation metrics. For each model, we compute the FAD with VGG-ish and CLAP embeddings using FMA-pop as a reference dataset. For reference, we also compute FAD scores for both MAGNeT and MusicGen’s standard, non-looping, generations). MusicGen Naive 2512.39 MeanModel MAGNeT Naive 1549.06 MusicGen Beat Aligned Naive 507.07 MAGNeT Beat Aligned Naive 153.22 MAGNeT Tiled 56.17 MAGNeT Hybrid Tiled 94.41 1e0 1e1 1e2 1e3 1e4 1e5 Seam Perplexity LoopGen 84.29 Figure 4.Seam Perplexity distribution of the considered models (lower is better). 6.4 Human evaluation Using the previous setups, we prepare a set of: 100 10-seconds clips from LoopGen (ours), and 100 10seconds clips from MAGNeT Hybrid Naive (without Tilinggeneration, baseline). We select the latter model because it is the most similar to ours, without any of the modifications introduced in this paper. This ensures a fair comparison, with the primary expected difference being seamlessness. The clips are chosen to be 10 seconds long for ease of listening. With this set of samples, we conduct a blind listening experiment with a group of users. Each volunteer listens to up to 30 randomly selected clips (15 from our model, 15 from the baseline) and rates the perceptibility of the seam on a Likert scale (1 = Evident cut, 5 = Imperceptible cut). In total, we collect 506 data points sfrom 18 12345 Human-assigned score 0.0 0.1 0.2 0.3 0.4 Density Model Baseline LoopGen Figure 5. Distribution of perceptibility ratings, comparing LoopGen with the baseline. Lines are model’s mean. listening sessions. Computing each user’s average rating of each model; we run a paired t-test such that H0≡µLoopGen =µbaseline (6) This yields t(17) = 12.21, p < 10−9, providing overwhelming evidence against H0. Furthermore, the effect size is large (d= 2.88), confirming a very strong evidence that our technique substantially reduces the perceptibility of the seam, as can also be seen in Figure 5. 7. CONCLUSIONS With this research, we have introduced a novel inferenceonly approach for generating loopable music, leveraging a simple “circular” padding scheme within MAGNeT’s nonautoregressive framework to ensure seamless boundaries. Our experiments demonstrated clear gains in loop continuity, validated both by a new perplexity-based seam metric and by human listening tests. The whole procedure does not require additional training or specialized loop datasets. By aligning loop length to musical beats, the generated audio segments more naturally fit common compositional structures, further improving their usability in practice. Overall, this work underscores the potential of lightweight, inference-time solutions for enhancing generative music models. Proceedings of the 26th ISMIR Conference, Daejeon, Korea, September 21-25, 2025 541 8. ACKNOWLEDGMENTS This work is supported by Sapienza University of Rome via the Seed of ERC grant “MINT.AI”, cup B83C25001040001. Furthermore, we thank all of the participants in the human evaluation test. 9. REFERENCES [1] J. Copet, F. Kreuk, I. Gat, T. Remez, D. Kant, G. Synnaeve, Y. Adi, and A. Défossez, “Simple and controllable music generation,” Advances in Neural Information Processing Systems, vol. 36, 2024. [2] A. Ziv, I. Gat, G. Le Lan, T. Remez, F. Kreuk, J. Copet, A. Défossez, G. Synnaeve, and Y. Adi, “Masked audio generation using a single non-autoregressive transformer,” in The Twelfth International Conference on Learning Representations, 2024. [3] P. Dhariwal, H. Jun, C. Payne, J. W. Kim, A. Radford, and I. Sutskever, “Jukebox: A generative model for music,” arXiv preprint arXiv:2005.00341, 2020. [4] H. Liu, Z. Chen, Y. Yuan, X. Mei, X. Liu, D. Mandic, W. Wang, and M. D. Plumbley, “AudioLDM: Text-toaudio generation with latent diffusion models,” Proceedings of the International Conference on Machine Learning, pp. 21 450–21 474, 2023. [5] F. Schneider, O. Kamal, Z. Jin, and B. Schölkopf, “Moûsai: Text-to-music generation with long-context latent diffusion,” 2023. [Online]. Available: https: //arxiv.org/abs/2301.11757 [6] S. Forsgren and H. Martiros, “Riffusion - stable diffusion for real-time music generation,” 2022. [Online]. Available: https://riffusion.com/about [7] O. Goren, E. Nachmani, and L. Wolf, “A-muze-net: Music generation by composing the harmony based on the generated melody,” 2021. [Online]. Available: https://arxiv.org/abs/2111.12986 [8] C.-Z. A. Huang, A. Vaswani, J. Uszkoreit, N. Shazeer, I. Simon, C. Hawthorne, A. Dai, M. Hoffman, M. Dinculescu, and D. Eck, “Music transformer: Generating music with long-term structure,” 2019. [Online]. Available: https://arxiv.org/abs/1809.04281 [9] A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. u. Kaiser, and I. Polosukhin, “Attention is all you need,” in Advances in Neural Information Processing Systems, vol. 30, 2017. [10] A. Agostinelli, T. I. Denk, Z. Borsos, J. Engel, M. Verzetti, A. Caillon, Q. Huang, A. Jansen, A. Roberts, M. Tagliasacchi, M. Sharifi, N. Zeghidour, and C. Frank, “Musiclm: Generating music from text,” 2023. [Online]. Available: https://arxiv.org/abs/2301. 11325 [11] S. Vasquez, S. Vasquez, M. Lewis, M. Lewis, M. Lewis, M. Lewis, M. Lewis, and M. Lewis, “Melnet: A generative model for audio in the frequency domain,” arXiv: Audio and Speech Processing, 2019. [12] J. Gardner, S. Durand, D. Stoller, and R. Bittner, “Llark: A multimodal instruction-following language model for music,” Proc. of the International Conference on Machine Learning (ICML), 2024. [13] J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” in Advances in Neural Information Processing Systems, vol. 33, 2020, pp. 6840–6851. [14] J. Nistal, M. Pasini, C. Aouameur, M. Grachten, and S. Lattner, “Diff-a-riff: Musical accompaniment cocreation via latent diffusion models,” 2024. [Online]. Available: https://arxiv.org/abs/2406.08384 [15] Z. Evans, C. Carr, J. Taylor, S. H. Hawley, and J. Pons, “Fast timing-conditioned latent audio diffusion,” 2024. [Online]. Available: https://arxiv.org/abs/2402.04825 [16] G. Mariani, I. Tallini, E. Postolache, M. Mancusi, L. Cosmo, and E. Rodolà, “Multi-source diffusion models for simultaneous music generation and separation,” in The Twelfth International Conference on Learning Representations, 2024. [Online]. Available: https://openreview.net/forum?id=h922Qhkmx1 [17] J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “BERT: Pre-training of deep bidirectional transformers for language understanding,” in Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers). Association for Computational Linguistics, 2019, pp. 4171–4186. [18] J. Li, T. Tang, W. X. Zhao, J.-Y. Nie, and J.-R. Wen, “Elmer: A non-autoregressive pre-trained language model for efficient and effective text generation,” 2022. [19] H. F. F. Garcia, P. Seetharaman, R. Kumar, and B. Pardo, “Vampnet: Music generation via masked acoustic token modeling,” in Ismir 2023 Hybrid Conference, 2023. [20] Z. Borsos, M. Sharifi, D. Vincent, E. Kharitonov, N. Zeghidour, and M. Tagliasacchi, “Soundstorm: Efficient parallel audio generation,” 2023. [21] P. Chandna, A. Ramires, X. Serra, and E. Gómez, “Loopnet: Musical loop synthesis conditioned on intuitive musical parameters,” in ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2021, pp. 3395– 3399. [22] G.-Y. Chen and V.-W. Soo, “Controllable music loops generation with midi and text via multi-stage cross attention and instrument-aware reinforcement learning,” in Proceedings of the 32nd ACM International Conference on Multimedia, 2024, p. 6851–6859. Proceedings of the 26th ISMIR Conference, Daejeon, Korea, September 21-25, 2025 542 [23] S. Han, H. R. Ihm, M. Lee, and W. Lim, “Symbolic music loop generation with neural discrete representations,” in International Society for Music Information Retrieval Conference, 2022. [Online]. Available: https://api.semanticscholar.org/CorpusID: 251493133 [24] Z. Novack, J. McAuley, T. Berg-Kirkpatrick, and N. J. Bryan, “Ditto: Diffusion inference-time t-optimization for music generation,” 2024. [25] A. Frühstück, I. Alhashim, and P. Wonka, “Tilegan: synthesis of large-scale non-homogeneous textures,” ACM Trans. Graph., vol. 38, pp. 58:1–58:11, 2019. [26] C. Rodríguez-Pardo and E. Garces, “Seamlessgan: Self-supervised synthesis of tileable texture maps,” IEEE Trans. Vis. Comput. Graph., vol. 29, pp. 2914– 2925, 2023. [27] O. Madar and O. Fried, “Tiled diffusion,” 2024. [Online]. Available: https://arxiv.org/abs/2412.15185 [28] F. Foscarin, J. Schlüter, and G. Widmer, “Beat this! accurate beat tracking without dbn postprocessing,” in Proceedings of the 25th International Society for Music Information Retrieval Conference (ISMIR), San Francisco, CA, United States, Nov. 2024. [29] K. Kilgour, M. Zuluaga, D. Roblek, and M. Sharifi, “Fréchet audio distance: A metric for evaluating music enhancement algorithms,” 2019. [Online]. Available: https://arxiv.org/abs/1812.08466 [30] A. Défossez, J. Copet, G. Synnaeve, and Y. Adi, “High fidelity neural audio compression,” arXiv preprint arXiv:2210.13438, 2022. [31] A. Gui, H. Gamper, S. Braun, and D. Emmanouilidou, “Adapting frechet audio distance for generative music evaluation,” in Proc. IEEE ICASSP 2024, 2024. [Online]. Available: https://arxiv.org/abs/2311.01616 [32] S. Hershey, S. Chaudhuri, D. P. W. Ellis, J. F. Gemmeke, A. Jansen, R. C. Moore, M. Plakal, D. Platt, R. A. Saurous, B. Seybold, M. Slaney, R. J. Weiss, and K. Wilson, “Cnn architectures for large-scale audio classification,” in 2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2017, pp. 131–135. [33] Y. Wu*, K. Chen*, T. Zhang*, Y. Hui*, T. BergKirkpatrick, and S. Dubnov, “Large-scale contrastive language-audio pretraining with feature fusion and keyword-to-caption augmentation,” in IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP, 2023. [34] M. Defferrard, K. Benzi, P. Vandergheynst, and X. Bresson, “FMA: A dataset for music analysis,” in 18th International Society for Music Information Retrieval Conference (ISMIR), 2017. [Online]. Available: https://arxiv.org/abs/1612.01840 Proceedings of the 26th ISMIR Conference, Daejeon, Korea, September 21-25, 2025 543 A. SEAM PERPLEXITY’S ERROR MARGINS In various tables, we present the values of our Seam Perplexity as center ±standard error. Because perplexity is the exponentiation of average cross-entropy, it is impossible to actually compute error margins directly. To obtain these values, we start from Equation (4) and compute for each dataset of samples X={X1, . . . , XN}the mean cross-entropy: µX=1 N N X k=1 HseamX(k)(7) and standard deviation σX=v u u t 1 N−1 N X k=1 µX−HseamX(k)2.(8) We then compute the 95% confidence intervals for the cross-entropy hµX−1.96 σX √N, µX+ 1.96 σX √Ni,(9) and transform them into exponential space hl= exp µX−1.96 σX √N, r = exp µX+ 1.96 σX √Ni.(10) Finally, we calculate the provided values as center =1 2(l+r),standard error =1 2(r−l).(11) This approach differs from the common method of showing a value with error margins, where the error is modeled as Gaussian, and the center value is assumed to be the empirical mean of the measured quantity. In this case, however, since the perplexity operation itself is computed as the exponentiation of its mean, it would be impossible to calculate a symmetric Gaussian error margin directly (not without running calculations on multiple folds of the data). B. 10 SECONDS EXPERIMENTS During development, we also explored the same final experiments seen in the main article (Table 2) with the 10 seconds variant of MAGNeT. The results of these experiments are detailed in Table 3and visualized in Figure 6. Notably, the Seam Perplexity exhibits a significant change with this modification. While it is unclear whether this change is solely attributable to the different models, the shorter track length, or a combination thereof, we empirically observed no discernible perceptual difference in the seamlessness of the 10-second and 30-second samples. Model Variant FADvggish(↓)FADCLAP(↓)Seam Perplexity (↓) MAGNeT Vanilla 3.05 0.39 — MAGNeT Naive 3.02 0.31 310.21 ±98.33 MAGNeT Beat Aligned Naive 3.03 0.35 202.43 ±67.23 MusicGen Vanilla 3.28 0.41 — MusicGen Naive 3.21 0.34 529.79 ±167.87 MusicGen Beat Aligned Naive 3.24 0.31 302.88 ±79.91 MAGNeT Tiled 3.51 0.40 18.15 ±3.53 MAGNeT Hybrid Tiled 2.98 0.33 44.42 ±9.33 MAGNeT LoopGen 2.95 0.33 60.85 ±15.24 Table 3. 10 seconds versions of main experiments’ evaluation Proceedings of the 26th ISMIR Conference, Daejeon, Korea, September 21-25, 2025 544