{"id":20303,"date":"2026-09-18T13:37:46","date_gmt":"2026-09-18T13:37:46","guid":{"rendered":"https:\/\/interspacemusic.com\/blog\/epidemic-sound-adds-conversational-search-powered-by-cyanite\/"},"modified":"2026-09-18T13:37:46","modified_gmt":"2026-09-18T13:37:46","slug":"epidemic-sound-adds-conversational-search-powered-by-cyanite","status":"publish","type":"post","link":"https:\/\/interspacemusic.com\/blog\/epidemic-sound-adds-conversational-search-powered-by-cyanite\/","title":{"rendered":"Epidemic Sound Adds Conversational Search Powered by Cyanite"},"content":{"rendered":"

Epidemic Sound<\/a><\/strong> has added conversational search<\/a> to its website, with Cyanite<\/a><\/strong>‘s music understanding and search technology operating beneath a large language model (LLM). The implementation relies on structured music metadata to make catalog search work through natural-language conversation.<\/p>\n

Cyanite spent recent months improving its tagging models, while Epidemic Sound built the conversational search experience on top of that foundation. The result allows users to query the catalog in natural language and retrieve relevant tracks.<\/p>\n

Why structured metadata matters for AI workflows<\/h2>\n

As AI moves into agentic workflows, new catalog tools will require machines to understand and find music. Structured music metadata is the layer that makes those workflows possible.<\/p>\n

Across the music industry, repetitive catalog work often depends on unwritten knowledge held by individual staff. Software has historically automated only explicitly defined processes, but language models can now mimic 80 to 90 percent of a text-based workflow when given enough context.<\/p>\n

Language models still cannot adequately understand music on their own. They need structured music metadata describing sound and lyrics to recognize patterns behind catalog decisions.<\/p>\n

In one example, a creative sync professional<\/a> at a music publisher sorts every new track into one of about 30 folders based on categories such as luxury brands, automotive campaigns, FMCG (fast-moving consumer goods), and uplifting music. Those folders reflect years of client placements rather than simple sonic similarity.<\/p>\n

An AI might identify a track as energetic, guitar-driven, and around 120 BPM (beats per minute), but that alone does not explain why it would be chosen for a specific folder. Given structured data from tracks already in those folders, the system can learn the patterns behind those decisions and suggest where a new song belongs before a human reviewer listens to it.<\/p>\n

Conversational search and catalog retrieval<\/h2>\n

At Epidemic Sound, the AI does not replace the underlying music understanding. It uses that understanding together with retrieval to make the catalog searchable through conversation.<\/p>\n

A conversational search system can interpret a request in natural language and connect it to the right tracks. A publisher can sort new uploads into an existing taxonomy, and a creative team can search for music that fits a specific brief. Different teams ask different questions, but they draw from the same catalog and pay the cost of understanding the music once.<\/p>\n

Similar deployments at Soundstripe and BeatStars<\/h2>\n

The approach is not isolated to Epidemic Sound. The following companies use Cyanite’s structured music understanding in different workflows:<\/p>\n