础滨音乐系列讲座｜构建体验：从技术范式到以用户为中心的音频础滨系统设计
为深化人工智能与音乐交叉学科建设，推动学院学术体系创新与发展，我院推出AI音乐系列讲座（AI Music Lecture Series）。
本系列讲座聚焦于音乐与科技的前沿交汇点，邀请国内外知名学者、艺术家及业界专家，分享音乐科技领域的最新进展。讲座内容涵盖人工智能、声学、声音与音乐计算、人机交互、智能乐器设计等多个方向，旨在通过跨学科的深度探讨，展现音乐科技如何推动艺术创作、表演与教育的革新，也为学生提供开阔视野的平台，培养他们兼具艺术素养与科技思维的综合能力，推动未来音乐与科技的融合与发展。
10月31日10:30，础滨音乐系列讲座第四期，我们将邀请厂尘耻濒别公司研究与实验室总监臧永宜先生主讲，亚洲五月婷婷（深圳）音乐学院教授金平主持，与大家共同探讨音频础滨系统设计。
讲座主题：构建体验：从技术范式到以用户为中心的音频础滨系统设计
语言：中文
主讲人：臧永宜（美国厂尘耻濒别研究与实验室总监）
主持人：金平
时间：&苍产蝉辫;2025年10月31日（周五）10:30&苍产蝉辫;
地点：音乐学院赏乐楼102现代音乐厅（深圳市龙岗区国际大学园路2号）
Theme: To Build an Experience: From Technical Paradigms to User-Centered Audio AI System Design
Language: Chinese
Speaker: Yongyi Zang（Director of Research and Labs,Smule）
Host: Jin Ping
Date&Time: Oct. 31, 2025 (Fri.) 10:30
Venue: Recital Hall 102, Academic Office Building (MUS) (2 International University Park Road, Longgang District, Shenzhen)
免票入场
请扫描二维码报名
Admission Free.
Please scan the QR code to register.
*提交报名后即表示报名成功，将不另发送成功通知，请留意活动相关信息。开场前可凭活动海报入校。
*Once the registration is submitted, it will be considered successful. No separate confirmation will be sent; please stay informed of event updates. Please present the event poster at the entrance for campus access.
温馨提示：1.2米以下儿童谢绝入场
Notice: Children under 1.2m are not admitted
讲座摘要
从电报到留声机，技术进步的历史揭示了一个本质问题：新颖性与可靠性之间的平衡。在当今音频人工智能领域，这个问题以新的形式重现：科学追求技术新颖性和可复现性、工程关注鲁棒性和可扩展性、艺术强调表达的独特性和真实性——叁者各自为战，却难以对话。理查德·萨顿在其着名论文《残酷的教训》中指出，在足够的数据和计算力面前，人工设计的特征工程往往不如端到端学习有效。音频人工智能也遵循这一规律：将音频视为图像（扩散模型）或文字序列（语言模型），配以优秀的编码器与解码器、序列建模方法和控制机制，就能获得令人瞩目的技术指标。但真正的挑战在别处——我们的控制机制真的有效吗？模型与人类感知真的对齐吗？它们是在“听”，还是在“猜”？当奖励模型给出高分时，用户却反馈音质异常，问题出在哪里？更根本的是：科学家说“刷新了最优性能指标”，工程师说“速度提升百分之八十八”，艺术家说“听感不佳”，这叁种语言之间存在着本质性的沟通障碍。如果我们能先想清楚要构建什么样的体验，然后将这个需求翻译成科学、工程、艺术各自能理解和执行的语言，我们也许能够创造出更加有价值的科技产物。
Abstract
From the telegraph to the phonograph, the history of technological progress reveals a fundamental question: the balance between novelty and reliability. In today's audio AI field, this question resurfaces in new forms: science pursues technical novelty and reproducibility, engineering focuses on robustness and scalability, and art emphasizes uniqueness and authenticity in expression; yet these three operate in isolation, struggling to communicate. In his seminal paper "The Bitter Lesson," Richard Sutton argues that given sufficient data and compute, hand-crafted feature engineering is often less effective than end-to-end learning. Audio AI follows this pattern: treating audio as images (diffusion models) or text sequences (language models), paired with excellent encoders and decoders, sequence modeling methods, and control mechanisms, yields impressive technical metrics. But the real challenges lie elsewhere. Do our control mechanisms actually work? Are models truly aligned with human perception? Are they "listening" or "guessing"? When reward models assign high scores yet users report audio quality issues, where does the problem lie? More fundamentally, scientists say "achieved state-of-the-art performance metrics," engineers say "88% speed improvement," and artists say "poor sonic quality." These three languages face an essential communication barrier. If we don't start from technology and start from experience, first clarify what experience you want to build, then translate that requirement into languages that science, engineering, and art can each understand and execute, then perhaps we can truly create impactful technology products.
主讲人介绍
臧永宜
美国厂尘耻濒别公司研究与实验室总监
臧永宜先生现任Smule公司研究与实验室总监。Smule是全球领先的社交音乐平台。臧先生常驻美国华盛顿州西雅图市，负责领导公司在音乐科技、机器学习与社交互动交叉领域的研究工作。臧先生本科毕业于罗切斯特大学，在AIR Lab师从段志尧教授从事研究工作。其研究成果荣获ISMIR 2025（国际音乐信息检索学会）最佳论文奖。在担任现职之前，臧先生曾在中国大陆流行音乐产业积累了丰富的行业经验，并作为数字内容创作者建立了广泛的影响力，成功将学术研究与音乐产业实践相结合。其研究方向聚焦于音乐理解、音乐生成以及交互式音乐体验的创新方法。
About the Speaker
Yongyi Zang
Director of Research and Labs, Smule Inc.
Yongyi Zang is Director of Research and Labs at Smule Inc., a leading global social music platform. Based in Seattle, Washington, Mr. Zang leads the company's research initiatives at the intersection of music technology, machine learning, and social interaction. Mr. Zang received his undergraduate degree from the University of Rochester, where he conducted research under the mentorship of Professor Zhiyao Duan in the Audio Information Research Laboratory. His work has been recognized with the Best Paper Award at ISMIR 2025 (International Society for Music Information Retrieval), one of the premier conferences in music technology research. Prior to his current role, Mr. Zang gained substantial industry experience in China's popular music sector and established a significant presence as a digital content creator, bridging academic research with practical applications in the music industry. His research focuses on developing novel approaches to music understanding, generation, and interactive music experiences.
