SoundsLike – AI-based Tagging and Search for Large Music Catalogs

SoundsLike offers a comprehensive set of standard categories, including:

• Speech Detector: Speech, NoSpeech
• Music Detector: Music, NoMusic
• Genre: Classical, Country, Electronica, Jazz, Latin, Pop, Rap, Rock, Soul, Speech
• Emotion: Anxious, Depressed, Exuberant, Content
• Valence: High, Low
• Arousal: High, Low
• Perceived Tempo: Fast, MidFast, Mid, MidSlow, Slow
• Beats: per minute e.g. 120 bpm
• Key: C to B
• Color: Bright, Dark
• Texture: Hard, Soft
• Instrumental Density: Full, Sparse
• Distortion: Clean, Gained, Overdrive, XXL
• Dynamic: Changing, Continuous
• Percussive: Nonpercussive, Percussive
• Synthetic: Acoustic, Synthetic
• Loop-specific Genres: Bass_music, Breaks, Cinematic, Disco, Hard_dance, Hip_hop, House, Idm, Pop, Blues, Country, Downtempo, Drum_and_bass, Funk, Reggae, Soul_and_rnb, Synthwave, Rock, Techno, Trance, Trap, World
• Loop-specific Instruments: Bass, Cymbal, Drum, Effects, Guitar, Keyboard, Mallet, Mixed, Percussion, String, Synth, Vocal, Wind_and_brass

SoundsLike provides the flexibility to use our standard metadata categories as a starting point and customize them based on your specific application – or use your very own category system as a basis. You can easily configure and compile categories to meet your unique requirements.

Absolutely! SoundsLike's AI technology can analyze your music archive and adapt to your specific metadata and category system. Together with our experts, you can develop a custom-fit category system optimized for your electronic or classical music archive.

Optionally, the AI model can also be trained using your existing music files and metadata to automatically annotate and assign new content.

The classification time for a 4-minute song with the default configuration takes around 8-10 seconds on an ordinary PC. The classification process supports multithreading/parallelization, resulting in reduced classification times on 8-/16-core CPUs (e.g., annotating 1000 songs in approximately 10 minutes on a 16-core CPU). Custom configurations may lead to different runtimes. Deep learning-based classification variants can be accelerated by GPUs.

1000 songs on 16-core CPU are annotated in about 10 minutes.

A desktop computer can perform Similarity Search in a database with 1 million songs in approximately 2 seconds. Recommendations can be precalculated regularly for an existing archive, ensuring immediate retrieval. Even when new titles are added, the similarity relations of the existing data remain stored in the cache and are updated in the background based on the new titles, providing near real-time access to recommendation lists.

The number of similar music titles output for a search query can be individually set in the configuration. You can specify whether you want the most similar title only or a list of several hundred titles sorted in descending order of similarity to the source title.

The results of Similarity Search can be further filtered using the annotated metadata from SoundsLike Annotation. These filters can be combined as desired, allowing you to refine your search results based on specific criteria. For example, you can search for jazz music with a slow tempo and vocals.