Episode Details

How does your brain isolate a referee's whistle from thousands of screaming fans? This episode explores Colin Cherry's groundbreaking "Cocktail Party Problem" from cognitive science and asks why machines struggle with the same challenge—especially when building computer vision systems for volleyball. Hosts discuss the neuroscience of selective hearing and what it means for audio isolation in sports technology. 00:00 - Opening: The paradox of the volleyball whistle 02:30 - Colin Cherry's Cocktail Party Problem (1953) 05:00 - The dichotic listening experiment and human perception 08:00 - Why machines fail at noise isolation 11:00 - Connection to the volleyball CV project 14:30 - Practical audio tech and machine learning solutions --- Sources & further reading: • Colin Cherry — "Some Experiments on the Recognition of Speech, with One and with Two Ears" (1953) — (published in Journal of the Acoustical Society of America, Vol. 25, No. 5) • Alexandre Défossez, Nicolas Usunier, Léon Bottou, Francis Bach (Meta AI) — "Hybrid Spectrogram and Waveform Source Separation": https://arxiv.org/pdf/2111.03600 • Hasan et al. — "Detecting Referee Whistle Sounds in Soccer Videos Using Machine Learning Models": https://publications.isods.org/index.php/jdsai/article/view/33 • Norbert Wiener — Extrapolation, Interpolation, and Smoothing of Stationary Time Series (1949) — (MIT Press) • Nemisindo / Sound & Design — "The Final Whistle Blows" (referee whistle acoustics): https://soundand.design/the-final-whistle-blows-615299703aaa • Number Analytics — "Ultimate Guide: Wiener Filtering Audio Signal Processing": https://www.numberanalytics.com/blog/ultimate-guide-wiener-filtering-audio-signal-processing • Stöter, Liutkus, Ito (sigsep) — "Open-Unmix: A Reference Implementation for Music Source Separation": https://github.com/sigsep/open-unmix-pytorch • TensorFlow — "Sound Classification with YAMNet": https://www.tensorflow.org/hub/tutorials/yamnet • TensorFlow — "Transfer Learning with YAMNet for Environmental Sound Classification": https://www.tensorflow.org/tutorials/audio/transfer_learning_audio • UCSD Music — "Harmonic Product Spectrum": http://musicweb.ucsd.edu/~trsmyth/analysis/Harmonic_Product_Spectrum.html • Mix Online — "1966: Ray Dolby, Dolby A-Type Noise Reduction": https://www.mixonline.com/technology/1966-ray-dolby-dolby-type-noise-reduction-383598 • National Museum of American History — "Dolby's Noise Reduction System": https://americanhistory.si.edu/explore/exhibitions/americas-listening/online/dolby-noise-reduction • Thuresson — "Human Whistle Detection and Frequency Estimation" (thesis): https://www.diva-portal.org/smash/get/diva2:836227/FULLTEXT01.pdf • Phys.org — "How loud is too loud when it comes to sports whistles?" (2019): https://phys.org/news/2019-05-loud-sports.html • Zambon & Faccinetto — "Automated Detection of Sport Highlights from Audio and Video Sources" (Jan 2025): https://arxiv.org/abs/2501.16100 • Facebook Research — Demucs GitHub repository: https://github.com/facebookresearch/demucs • J. Stephen Lansing — Perfect Order: Recognizing Complexity in Bali (2006) — (Princeton University Press) • Universal lens check: "Every map is an argument" applies loosely — a spectrogram is literally a map of audio, and choosing which frequencies to keep is an argument about what matters — but it doesn't do enough work here to be worth calling out as a lens. None of the three lenses fit tightly enough to justify invoking. Skip them. This podcast episode was fully generated by AI — research, script, voices, and production. Built with Claude, Piper TTS, and automated pipeline tooling.