.SUSS

The 4 steps for analyzing music with neural networks include: 1. Collecting data 2. Preprocessing audio data 3. Training the neural network 4. Testing and evaluating the network

Regarding our example of genre classification, recent research has shown that the accuracy of a CNN architecture (82%) can surpass human accuracy (70%), which is quite impressive. Depending on the specific task, accuracy can be even higher.

Let’s say music is a code.

music is systematically structured, categorised, it follows a strict “grammar”. It is not mysterious, but enigmatic. Music is auditory code. A code that needs to be deciphered and translated. And we can process this code by technological means, like any other sign system. Unlike other codes, however, the code of music is not stable and predictable, but surprising and diverse.

One example is the digital scene “aurora” from the series Sound Data Sculpture Sketches. The creation process starts with a set of dots that move on a sphere. Over time their path is traced to form tubes, this produces an organic appearance. A representation of the underlying song’s frequencies is texture-mapped onto the geometry of the tubes and used to generate colour gradients that react to music. From this interpretative, digitally mediated translation of the original song, a dreamy audio-sculpture is created. By interpreting the musical parameters, this artwork goes further than a mere technical analysis. It thereby contemplates the poetry and beauty of the sonic language, seeking to visually formulate an accurate translation.

The whole theme of “translation” points to the fact, that music is socially formalised and follows symbolic structures. Music is deeply connected to our human experience because it works like a language, because it translates into emotion and bodily reactions. The notion that music is tangible and rests upon patterns that we can calculate and process with digital technologies is not as weird or scary as it might seems. Music is a code – and that is a beautiful thing.

1. Objective catalog language (tagging): the entity of keywords and tags often described as taxonomy or tag anthology (quantity, classes and wording). „Which tags do I use. 2. Subjective catalog language (understanding of tagging): the understanding of tags and their connection to certain sound qualities. „When do I assign a certain tag?“

Uplifting: Fresh, invigorating, and mood-boosting Energetic: Danceable, lively, and highly-active Relaxing: Kick back, chill out, and vibe Dark: Gritty, hard, unpredictable, and emotional Happy: Cheerful, bubbly, and carefree Tense: Edgy, angsty, with low or deep basslines Calm: Easy, smooth, and steady Melancholic: Moody, gloomy, and in your feelings More information about each track is also available such as voice presence, energy level, and energy dynamics: Voice Presence: the amount of singing voice throughout the full duration of the track, displayed as low, medium, or high Energy Level – the intensity of an analyzed track, displayed as low, medium, high, or variable Energy Dynamics – the progression of the Energy Level throughout the track, displayed as low, high, or variable

SWR Labs

1. AI Music Creation 2. Search & Recommendation 3. Auto-tagging 4. AI Mastering

Companies such as Cyanite, Musiio, Musimap, or FeedForward are able to extract descriptive metadata from the audio file.

Proprietary data is the data that remains in the hands of the company and rarely gets disclosed. Some data used by recommender systems is proprietary data. For example, a song’s similarity score is proprietary data that can be based on performance data. This type of data also includes insights from ad campaigns, sales, merch, ticketing, and etc.

Cyanite’s moods are presented in a comparison chart where overarching mood and the least present mood can be easily identified.

You can select up to 10 keywords from the Augmented Keywords set in the search bar. The chosen keywords will appear on the right with a weight bar and a cursor which you can place at any point between -1 to 1. For example, the keyword search input sparkling: 0.5, sad: -1, rock: 1, dreamy: 1, would refer to a search for a rock track that is dreamy, slightly sparkling, and not at all sad.

The Keyword Search by Weights:   Allows for increased relevance of search results  Provides better control over the outcome of the search Ensures a level of granularity and ability to create slight or total differentiation between tracks Takes away the stress of scrolling and checking through hundreds of search results Saves time on finding the right fitting songs for various use cases

Sonic Similarity Search

Similarity search algorithms usually work to compute the distance between songs in the space of different audio features. Calculating similarity distance allows to input a query that will then return tracks with the shortest similarity distances which means they are the most similar ones.

1. Finding similar songs using audio references for sync and music briefs

Using own songs and finding similar songs on Spotify to detect blind spots in a catalog

Finding samples across the catalog

n sample catalogs, Similarity Search algorithms can be used to find similar samples. Instead of spending hours digging through sample libraries, the user can select a reference sample or the first sample they like and get a list of similar-sounding samples.

Similarity Search can reduce search time up to 86% and offload some labor-intensive tasks from users as well as music catalog owners.

Two other companies that specialize in MIR (music information retrieval) are AIMS and Cyanite, who use AI music tagging – adding descriptive metadata identifiers to songs – and sonic similarity – search functionality that enhances the discoverability of material with matching audio characteristics – to extract value from catalogs. The use case scenarios here are multiple. Music rights holders can deploy the technology to tag vast collections, maximizing their assets and monetizing not just the hits but every song. It also promises to be an incredibly powerful tool for sync and music licensing, particularly during a time when demand is exploding across the entertainment sector and beyond.

AI Music Creation / Search & Recommendation / Auto-tagging / AI Mastering

BPM / Key / Mood / Main Genre / Sub Genre / Voice / Emotional Profile / Instruments / Energy Level / Musical Era

The results they produced feel detailed, accurate and informative, bringing tangible sonic representation from the language of machine code.

Text Search will provide a list of suitable tracks. The search requests may be as complex as “A city lost to time, its buildings half buried by the jungle that has claimed it” or as simple as “Walking down a dusty highway”.

Cyanite helps music companies turn their catalogs into their own personal Spotify – providing music libraries with the usability, transparency and functionality users have become accustomed to.