Most similarity models collapse melody, rhythm, and timbre into a single undifferentiated score. MERIT exposes all three as independent, interpretable signals from the same audio query.</p>\n","updatedAt":"2026-06-03T04:07:56.939Z","author":{"_id":"655431b2997379e9b0999d23","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/655431b2997379e9b0999d23/OhHCPCXfOS61Z7CSewUCU.png","fullname":"Dorien Herremans","name":"dorienh","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":12,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9134742021560669},"editors":["dorienh"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/655431b2997379e9b0999d23/OhHCPCXfOS61Z7CSewUCU.png"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2605.27346","authors":[{"_id":"6a1fa758e292c1c78ecb13ba","name":"Abhinaba Roy","hidden":false},{"_id":"6a1fa758e292c1c78ecb13bb","name":"Junyi Liang","hidden":false},{"_id":"6a1fa758e292c1c78ecb13bc","name":"Dorien Herremans","hidden":false}],"publishedAt":"2026-05-26T00:00:00.000Z","submittedOnDailyAt":"2026-06-03T00:00:00.000Z","title":"MERIT: Learning Disentangled Music Representations for Audio Similarity","submittedOnDailyBy":{"_id":"655431b2997379e9b0999d23","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/655431b2997379e9b0999d23/OhHCPCXfOS61Z7CSewUCU.png","isPro":false,"fullname":"Dorien Herremans","user":"dorienh","type":"user","name":"dorienh"},"summary":"Current music similarity models typically compute a single, monolithic score, entangling distinct musical dimensions like melody, rhythm, and timbre. This limits user control and interpretability, making it impossible to execute nuanced queries. We introduce MERIT, a framework for learning disentangled, factor-specific music representations tailored to these three core dimensions. To overcome the lack of isolated musical variations in real-world audio, we use a novel training strategy that uses conditional audio generation and source-separated stems to strongly encourage single-factor variation in training data. Our evaluations demonstrate strong factor-wise disentanglement. Each head responds strongly to its intended perceptual dimension while remaining near chance on the others, a representational property that holds across both the synthetic training domain and independent real-world audio.","upvotes":5,"discussionId":"6a1fa758e292c1c78ecb13bd","projectPage":"https://github.com/AMAAI-Lab/MERIT","githubRepo":"https://github.com/AMAAI-Lab/MERIT","githubRepoAddedBy":"user","ai_summary":"MERIT framework learns disentangled music representations for melody, rhythm, and timbre through conditional audio generation and source-separated stems, enabling nuanced musical queries.","ai_keywords":["music similarity models","disentangled representations","factor-specific music representations","conditional audio generation","source-separated stems","perceptual dimensions"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct","githubStars":20,"organization":{"_id":"655435e0db4619c8eb051bd0","name":"amaai-lab","fullname":"AMAAI Lab","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/655431b2997379e9b0999d23/zrIzOdksqPZXB0QsnGL2H.jpeg"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"655431b2997379e9b0999d23","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/655431b2997379e9b0999d23/OhHCPCXfOS61Z7CSewUCU.png","isPro":false,"fullname":"Dorien Herremans","user":"dorienh","type":"user"},{"_id":"6a1fac9798a95251fc13566a","avatarUrl":"/avatars/9c0a957c43b0c8a57b04ec6b76a3d93e.svg","isPro":false,"fullname":"Roshan Emmanuel Manoranjan","user":"redstonerosh12","type":"user"},{"_id":"665d506460135db197b2f483","avatarUrl":"/avatars/207f4c3cfe3705dd68e62b09390ada4c.svg","isPro":false,"fullname":"Geeta Puri","user":"geetapuri","type":"user"},{"_id":"6166fa64e4e97c73a0703992","avatarUrl":"/avatars/aa3006c77786dd05cdbc8cdeab9907bc.svg","isPro":false,"fullname":"Keshav Bhandari","user":"keshavbhandari","type":"user"},{"_id":"635ba0c637c6a2c12e2daef9","avatarUrl":"/avatars/9fc2932d9ace2715f540f896754ec7d2.svg","isPro":false,"fullname":"Ollie McCarthy","user":"ollieollie","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"655435e0db4619c8eb051bd0","name":"amaai-lab","fullname":"AMAAI Lab","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/655431b2997379e9b0999d23/zrIzOdksqPZXB0QsnGL2H.jpeg"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2605/2605.27346.md"}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2605.27346 in a Space README.md to link it from this page.

Add this paper to a collection to link it from this page.