Amadeus Code Inc. (Head Office: Minato-ku, Tokyo; Representative Director: Jun Inoue; hereinafter “Amadeus Code”) has developed Emotion Search, a proprietary AI-powered search technology that enables users to find music related to the meaning and nuance of natural-language descriptions containing multiple emotions, such as “happy, but a little lonely” or “excited, yet anxious.”
Traditional music search typically relies on combinations of mood tags such as “Happy,” “Sad,” “Relaxing,” and “Energetic,” along with genres, instruments, BPM, intended use, and other attributes.
However, the emotions people want to express through music do not necessarily fit into a single category. They may involve multiple emotions, varying degrees of intensity, changes over time, and specific scenes or situations.
With Emotion Search, users can enter phrases such as:
“Feeling a little anxious but excited before starting something new”
“Feeling slightly sentimental when remembering the past, but not exactly sad”
“Quietly enjoying a sense of accomplishment alone after completing a major project”
The system evaluates the semantic relationship between the meaning of the entered text and the emotional information associated with each track.
The technology is built on natural-language descriptions of music and emotional information accumulated by Amadeus Code through professional musicians, combined with a newly developed AI search engine specifically designed for music discovery.
Rather than treating emotional information attached to music as simple categories or keywords, the new AI search engine interprets it as semantic information that includes relationships between emotions, their intensity, and their context.
This makes it possible to reflect nuances of complex human emotions in music search—nuances that are difficult to capture with single mood categories such as “Happy” or “Sad.”
The emotional information accumulated by Amadeus Code also includes expressions provided by native Japanese-speaking professional musicians, such as setsunai (切ない), nagorioshii (名残惜しい), and monoganashii (物悲しい)—expressions whose nuances cannot always be preserved by replacing them with a single English word.
By applying AI-powered language processing, Amadeus Code is developing multilingual search capabilities that make this Japanese-language emotional information searchable from English and other languages.
During development, Amadeus Code conducted a comparative evaluation involving approximately 100 creators under contract with the company, comparing Emotion Search with conventional mood-tag-based search. The results showed that, compared with conventional search, the average time required to reach the desired track was reduced by approximately 30%, while the rate at which users found music matching their intended image increased by approximately 45%.
Emotion Search will be made available through API and MCP via the MusicTGA-HR music dataset, and will also be progressively introduced into the royalty-free music service Evoke Music and OTOKAI, a service that uses AI music to support the maintenance of cognitive abilities.
Beyond music search for human users, the technology will also serve as infrastructure enabling generative AI and AI agents to search for and retrieve music according to the context of conversations and content.
Background: The Challenges of Music Search in the Age of Generative AI
Music streaming services and background music platforms have traditionally relied on search methods centered on track titles, artist names, genres, instruments, BPM, intended use, and similar criteria.
Search using predefined mood information such as “Happy,” “Sad,” “Relaxing,” and “Energetic” has also become common.
However, the emotions people perceive in music, and the emotions content creators want music to express, cannot always be categorized using a single word.
For example:
“Happy, but a little lonely”
“Anxious, but excited at the same time”
“Nostalgic, but without wanting to return to the past”
Real emotions often contain multiple elements and varying degrees of intensity.
This is particularly relevant in content creation for video, advertising, games, social media, and other fields. Creators may find it difficult to describe the music they are looking for using conventional search parameters such as “Pop,” “Happy,” or “Piano,” resulting in repeated searches and extensive listening before finding music that matches their intended image.
At the same time, the spread of generative AI is creating new use cases in which not only humans, but AI agents themselves select music according to the context of conversations, videos, stories, and other content.
In such an environment, music search technology needs to go beyond treating emotional information as fixed categories. It must interpret meaning and nuance in relation to the natural-language input provided by the user or AI.
To address these challenges, Amadeus Code developed Emotion Search as a technology that uses a new AI search engine to semantically interpret and search the music and emotional data the company has accumulated over time.
Key Features of Emotion Search
1. Search Using Natural-Language Descriptions of Emotions
Emotion Search does not require users to specify genres or technical music terminology.
For example, users can enter descriptions such as:
“The feeling at the end of summer vacation—happy about the memories, but a little sad that it is over”
“A mixture of excitement and nervousness, as though something big is about to begin”
“The feeling of making a quiet decision alone late at night”
Emotion Search evaluates the semantic relationship between these descriptions and the emotional information accumulated for each track, and then retrieves r