Elamur Research⌕
☰
home / papers / 2026 / CIS-BWE: Chaos-Informed Speech Bandwidth Extension

CIS-BWE: Chaos-Informed Speech Bandwidth Extension

Paper · 2026 · Tarikul Islam Tamiti, Tonmoy Das, Nursadul Mamun et al.

Summary

CIS-BWE tackles speech bandwidth extension (BWE): recovering the missing high-frequency content of narrowband speech, which supports downstream TTS and ASR. The authors argue that speech production involves deterministic chaos, and that standard GAN discriminators such as MPD and multi-resolution spectrogram critics do not model it while also being large. They propose two compact discriminators based on chaos theory:

  • MRLD compares Lyapunov exponents of real and generated waveforms over five window sizes.
  • MSDFA applies detrended fluctuation analysis to capture long-range, fractal-like correlations.

The generator is complex-valued and dual-stream, handling magnitude and phase in parallel. Its core block, ConformerNeXt, is a Conformer whose convolution module is replaced by ConvNeXt. Two Lattice blocks with learnable scalars act as gates that control how information mixes between the two streams. The phase is recovered from predicted pseudo-real and pseudo-imaginary residuals.

The model was evaluated on clean and noisy VCTK (English) and MLS (French) data. It beats the strongest baseline, AP-BWE, across nine objective and subjective metrics with roughly half the parameters. This makes it an efficient new reference point for BWE.

Key points

  • Size and compute: the generator totals 33.74M parameters versus 72.07M for AP-BWE, with about half the MACs and FLOPs (6790.86M vs 14236.65M MACs). The discriminators are reported as 40x smaller, and inference speed is similar to AP-BWE.
  • Settings and metrics: tested on 4–16, 8–16 and 16–48 kHz upsampling. Metrics included LSD, PESQ, STOI, SI-SDR, SI-SNR, NISQA-MOS, WER, MOS and a pairwise preference test.
  • Noise robustness: in noisy VCTK from -10 to 10 dB SNR, CIS-BWE outperforms AP-BWE at every level. For example, NISQA-MOS at -10 dB (4–16 kHz) is 2.73 versus 2.02.
  • Listening tests: 10 listeners took part. CIS-BWE won the pairwise preference test by 11% and had a higher overall MOS, though AP-BWE scored better on male speakers.
  • Limitations: training is slower (25 vs 17 minutes per epoch), mostly because of Lyapunov exponent estimation. The chaos features are used only in the discriminators, so they do not affect inference. The code extends the MIT-licensed AP-BWE codebase.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓George Mason UniversityFairfax · United States24 works · 1,288 citations🎓Chittagong University of Engineering & TechnologyChittagong · Bangladesh25 works · 29 citations

 Fields, topics and concepts

📚 Shelves 7The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksSignal Processing · 452 worksPhysics and Astronomy · 44 worksStatistical and Nonlinear Physics · 40 works

Cite

Cite as Tarikul Islam Tamiti, Tonmoy Das, Nursadul Mamun and Anomadarshi Barua. 2026. CIS-BWE: Chaos-Informed Speech Bandwidth Extension. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-long.1925

@inproceedings{tamiti2026cis,
  title = {{CIS-BWE: Chaos-Informed Speech Bandwidth Extension}},
  author = {Tarikul Islam Tamiti, Tonmoy Das, Nursadul Mamun and Anomadarshi Barua},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.1925},
  url = {https://doi.org/10.18653/v1/2026.acl-long.1925}
}

References

55 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Wallace Abreu and Luiz Wagner Pereira Biscainho. 2024. Aeromamba: An efficient architecture for audio super-resolution using generative adversarial networks and state space models. arXiv preprint arXiv:2411.07364. find
  2. Bajibabu Bollepalli, Lauri Juvela, and Paavo Alku. 2019. Generative adversarial network-based glottal waveform model for statistical parametric speech synthesis. arXiv preprint arXiv:1903.05955. — [arXiv:1903.05955](https://arxiv.org/abs/1903.05955) find
  3. Yubing Cao, Yongming Li, Liejun Wang, and Yinfeng Yu. 2024. Vnet: A gan-based multi-tier discriminator network for speech synthesis vocoders. In 2024 IEEE International Conference on Systems, Man, and Cybernetics (SMC), pages 4384–4389. IEEE. learning bandwidth Donahue, Julian McAuley, and Miller Pu… — [arXiv:1802.04208](https://arxiv.org/abs/1802.04208) find
  4. A. Erell and M. Weintraub. 1990. Estimation using log-spectral-distance criterion for noise-robust speech recognition. In International Conference on Acoustics, Speech, and Signal Processing, pages 853–856 vol.2. find
  5. W Tecumseh Fitch. 2025. Applying nonlinear dynamics to the voice: a historical perspective. Philosophical Transactions B, 380(1923):20240024. find
  6. David Gaddy and Dan Klein. 2021. An improved model for voicing silent speech. arXiv preprint arXiv:2106.01933. find
  7. Timo Gerkmann, Martin Krawczyk, and Robert Rehr. 2012. Phase estimation in speech enhancement—unimportant, important, or impossible? In 2012 IEEE 27th Convention of Electrical and Electronics Engineers in Israel, pages 1–5. IEEE. find
  8. Anmol Gulati, James Qin, Chung-Cheng Chiu, Niki Parmar, Yu Zhang, Jiahui Yu, Wei Han, Shibo Wang, Zhengdong Zhang, Yonghui Wu, and Ruoming Pang. 2020. Conformer: Convolution-augmented transformer for speech recognition. In Interspeech 2020, pages 5036–5040.
  9. Awni Hannun, Carl Case, Jared Casper, Bryan Catanzaro, Greg Diamos, Erich Elsen, Ryan Prenger, Sanjeev Satheesh, Shubho Sengupta, Adam Coates, and 1 others. 2014. Deep speech: Scaling up end-to-end speech recognition. arxiv. arXiv preprint arXiv:1412.5567.
  10. Julien Hauret, Thomas Joubaud, Véronique Zimpfer, and Éric Bavu. 2023. Eben: Extreme bandwidth extension network applied to speech signals captured with noise-resilient body-conduction microphones. In ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICAS… find
  11. David Haws and Xiaodong Cui. 2019. Cyclegan bandwidth extension acoustic modeling for automatic speech recognition. In ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 6780–6784. IEEE. find
  12. Hanspeter Herzel, David A. Berry, Ingo R. Titze, and M. Saleh. 1994. Analysis of vocal disorders with methods from nonlinear dynamics. Chaos: An Interdisciplinary Journal of Nonlinear Science, 4(3):341– 348. find
  13. Chun-Wei Ho, Pin-Jui Ku, Hao Yen, Sabato Marco Siniscalchi, Yu Tsao, and Chin-Hui Lee. 2025. An investigation on combining geometry and consistency constraints into phase estimation for speech enhancement. arXiv preprint arXiv:2507.02192. — [arXiv:2507.02192](https://arxiv.org/abs/2507.02192) find
  14. Pengfei Hu, Hui Zhuang, Panneer Selvam San-thalingam, Riccardo Spolaor, Parth Pathak, Guoming Zhang, and Xiuzhen Cheng. 2022. Accear: Accelerometer acoustic eavesdropping with unconstrained vocabulary. In 2022 IEEE Symposium on Security and Privacy (SP), pages 1757–1773. IEEE. find
  15. Won Jang, Dan Lim, Jaesam Yoon, Bongwan Kim, and Juntae Kim. 2021. Univnet: A neural vocoder with multi-resolution spectrogram discriminators for high-fidelity waveform generation. arXiv preprint arXiv:2106.07889. find
  16. Jack J Jiang, Yu Zhang, and Jennifer Stern. 2001. Modeling of chaotic vibrations in symmetric vocal folds. The Journal of the Acoustical Society of America, 110(4):2120–2128. find
  17. Kehuang Li and Chin-Hui Lee. 2015. A deep neural network approach to speech bandwidth expansion. In 2015 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 4395– 4399. IEEE. find
  18. Max Little, Patrick Mcsharry, Stephen Roberts, Declan Costello, and Irene Moroz. 2007. Exploiting nonlinear recurrence and fractal scaling properties for voice disorder detection. Nature Precedings, pages 1–1. find
  19. Haohe Liu, Woosung Choi, Xubo Liu, Qiuqiang Kong, Qiao Tian, and DeLiang Wang. 2022a. Neural vocoder is all you need for speech super-resolution. In Interspeech. find
  20. Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, and Saining Xie. 2022b. A convnet for the 2020s. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11976–11986. find
  21. Ye-Xin Lu, Yang Ai, Hui-Peng Du, and Zhen-Hua Ling. 2024a. Towards high-quality and efficient speech bandwidth extension with parallel amplitude and phase prediction. IEEE/ACM Transactions on Audio, Speech, and Language Processing. find
  22. Ye-Xin Lu, Yang Ai, and Zhen-Hua Ling. 2025. Explicit estimation of magnitude and phase spectra in parallel for high-quality speech enhancement. Neural Networks, page 107562. find
  23. Ye-Xin Lu, Yang Ai, Zheng-Yan Sheng, and Zhen-Hua Ling. 2024b. Multi-stage speech bandwidth extension with flexible sampling rate control. arXiv preprint arXiv:2406.02250. find
  24. Ji-Hoon Kim, Sang-Hoon Lee, Ji-Hyun Lee, and Seong-Whan Lee. 2021. Fre-gan: Adversarial frequency-consistent audio synthesis. arXiv preprint arXiv:2106.02297. — [arXiv:2106.02297](https://arxiv.org/abs/2106.02297) find
  25. Xiaotong Luo, Yuan Xie, Yulun Zhang, Yanyun Qu, Cui-hua Li, and Yun Fu. 2020. LatticeNet: Towards Lightweight Image Super-Resolution with Lattice Block. In Andrea Vedaldi, Horst Bischof, Thomas Brox, and Jan-Michael Frahm, editors, Computer Vision – ECCV 2020 (Lecture Notes in Computer Science, vol… find
  26. Jungil Kong, Jaehyeon Kim, and Jaekyoung Bae. 2020. Hifi-gan: Generative adversarial networks for efficient and high fidelity speech synthesis. Advances in neural information processing systems, 33:17022– 17033. find
  27. Yi Luo and Nima Mesgarani. 2018. Tasnet: time-domain audio separation network for real-time, single-channel speech separation. In 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 696–700. IEEE. find
  28. Kundan Kumar, Rithesh Kumar, Thibault De Boissiere, Lucas Gestin, Wei Zhen Teoh, Jose Sotelo, Alexandre De Brebisson, Yoshua Bengio, and Aaron C Courville. 2019. Melgan: Generative adversarial networks for conditional waveform synthesis. Advances in neural information processing systems, 32. find
  29. Julia K MacCallum, Li Cai, Liang Zhou, Yu Zhang, and Jack J Jiang. 2009. Acoustic analysis of aperiodic voice: perturbation and nonlinear dynamic properties in esophageal phonation. Journal of Voice, 23(3):283–290. find
  30. Jonathan Le Roux, Scott Wisdom, Hakan Erdogan, and John R Hershey. 2019. Sdr–half-baked or well done? In ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 626–630. IEEE. find
  31. Moshe Mandel, Or Tal, and Yossi Adi. 2023. Aero: Audio super resolution in the spectral domain. In ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 1–5. IEEE. find
  32. F Martinez, Antonio Guillamón, Javier Conte Alcaraz, and MC Alcaraz. 2002. Detection of chaotic behaviour in speech signals using the largest lyapunov exponent. In 2002 14th International Conference on Digital Signal Processing Proceedings. find
  33. DSP 2002 (Cat. No. 02TH8628), volume 1, pages 317–320. IEEE. find
  34. Yueyuan Sui, Minghui Zhao, Junxi Xia, Xiaofan Jiang, and Stephen Xia. 2024. Tramba: A hybrid transformer and mamba architecture for practical audio and bone conduction speech super resolution and enhancement on mobile and wearable platforms. Proceedings of the ACM on Interactive, Mobile, Wearable an… find
  35. Nicholas A May and Ronald C Scherer. 2023. The effects of vocal tract constrictions on aerodynamic measures in a synthetic vocal fold model. The Journal of the Acoustical Society of America, 154(5):3310–3320. find
  36. Cees H Taal, Richard C Hendriks, Richard Heusdens, and Jesper Jensen. 2011. An algorithm for intelligibility prediction of time–frequency weighted noisy speech. IEEE Transactions on audio, speech, and language processing, 19(7):2125–2136. find
  37. BANBOOK Michael. 1999. Speech characterization and synthesis by nonlinear methods. IEEE Trans. Speech, Audio Processing. find
  38. Tarikul Islam Tamiti and Anomadarshi Barua. 2025. A practical approach to power saving in hearables using sub-nyquist sampling with bandwidth extension. arXiv preprint arXiv:2506.22321. — [arXiv:2506.22321](https://arxiv.org/abs/2506.22321) find
  39. Gabriel Mittag, Babak Naderi, Assmaa Chehadi, and Sebastian Möller. 2021. Nisqa: A deep cnn-self-attention model for multidimensional speech quality prediction with crowdsourced datasets. arXiv preprint arXiv:2104.09494. find
  40. V.I. Oseledec. 1968. A multiplicative ergodic theorem: Lyapunov characteristic numbers for dynamical systems. Trans. Moscow Math. Soc., 19:197–231. find
  41. David Pearce and Hans-Günter Hirsch. 2000. The aurora experimental framework for the performance evaluation of speech recognition systems under noisy conditions. In Proceedings of the 6th International Conference on Spoken Language Processing (ICSLP 2000), pages 29–32. ISCA Archive. find
  42. C.-K. Peng, S. V. Buldyrev, S. Havlin, M. Simons, H. E. Stanley, and A. L. Goldberger. 1994. Mosaic organization of dna nucleotides. Physical Review E, 49(2):1685–1689. find
  43. Vineel Pratap, Qiantong Xu, Anuroop Sriram, Gabriel Synnaeve, and Ronan Collobert. 2020. Mls: A large-scale multilingual dataset for speech research. arXiv preprint arXiv:2012.03411.
  44. Drew Rendall. 2025. Nonlinear phenomena in animal vocalizations: do they reflect alternative functional modes of voice control,‘leaked’cues to quality or condition, or both? Philosophical Transactions B, 380(1923):20240010. find
  45. Antony W Rix, John G Beerends, Michael P Hollier, and Andries P Hekstra. 2001. Perceptual evaluation of speech quality (pesq)-a new method for speech quality assessment of telephone networks and codecs. In 2001 IEEE international conference on acoustics, speech, and signal processing. Proceedings (C… find
  46. Leyuan Sheng and Evgeniy N Pavlovskiy. 2019. Reducing over-smoothness in speech synthesis using generative adversarial networks. In 2019 International Multi-Conference on Engineering, Computer and Information Sciences (SIBIRCON), pages 0972– 0974. IEEE. find
  47. Tarikul Islam Tamiti, Biraj Joshi, Rida Hasan, Rashedul Hasan, Taieba Athay, Nursad Mamun, and Anomadarshi Barua. 2025. A high-fidelity speech super resolution network using a complex global attention module with spectro-temporal loss. arXiv preprint arXiv:2507.00229. — [arXiv:2507.00229](https://arxiv.org/abs/2507.00229) find
  48. Chao Tao and Jack J Jiang. 2008. Chaotic component obscured by strong periodicity in voice production system. Physical Review E—Statistical, Nonlinear, and Soft Matter Physics, 77(6):061922. find
  49. Qiao Tian, Yi Chen, Zewang Zhang, Heng Lu, Linghui Chen, Lei Xie, and Shan Liu. 2020. Tfgan: Time and frequency domain based generative adversarial network for high-fidelity speech synthesis. arXiv preprint arXiv:2011.12206. — [arXiv:2011.12206](https://arxiv.org/abs/2011.12206) find
  50. Ingo R Titze. 2008. Nonlinear source–filter coupling in phonation: Theory. The Journal of the Acoustical Society of America, 123(5):2733–2749. find
  51. Alan Wolf, J. B. Swift, Harry L. Swinney, and John A. Vastano. 1985. Determining lyapunov exponents from a time series. Physica D: Nonlinear Phenomena, 16(3):285–317. find
  52. Junichi Yamagishi, Christophe Veaux, Kirsten MacDonald, and 1 others. 2019. Cstr vctk corpus: English multi-speaker corpus for cstr voice cloning toolkit (version 0.92). University of Edinburgh. The Centre for Speech Technology Research (CSTR), pages 271–350. find
  53. Geng Yang, Shan Yang, Kai Liu, Peng Fang, Wei Chen, and Lei Xie. 2021. Multi-band melgan: Faster waveform generation for high-quality text-to-speech. In 2021 IEEE Spoken Language Technology Workshop (SLT), pages 492–498. IEEE. find
  54. Dacheng Yin, Chong Luo, Zhiwei Xiong, and Wenjun Zeng. 2020. Phasen: A phase-and-harmonics-aware speech enhancement network. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 34, pages 9458–9465. find
  55. Zhaoyan Zhang. 2023. The influence of source-filter interaction on the voice source in a three-dimensional computational model of voice production. The Journal of the Acoustical Society of America, 154(4):2462–2475. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/tamiti-2026-cis-bwe-chaos-informed-speech/