Cepstra

Technology

Cepstra algorithm analyzes the top terms that describe a particular song or artist. Each artist and song can have thousands of characteristics representing them. Each one has an associated score, which describes how significant that description is for a song/artist. These tags or words are added to the model of each song and artist, which is then used for modeling what songs to recommend to a user.

With the help of sound processing technique and artificial intelligence, Cepstra Algorithm reads and understands music as if a human understands (and enjoy it! not sure if server enjoys it 😉 ). Cepstra’s Algorithms learns a massive number of important observations that humans make while listening to a song like the right pattern of modulation, the right combination of instruments playing at the right time of a song. Cepstra not only make hundreds of observation about the song also applied intelligence that a human would apply to, or create a likable song. The journey of understanding music starts from the number of instruments, voices played inside the mp3 file, modulation, and angles at which sounds are modulated, to put the information in Cepstra’s proprietary algorithm which not only learns the music patterns but also separates the voice and instruments.

The understanding starts from various frequency-based learning algorithms based on Fourier transforms and spectrograms. To learn the human likeliness of music, Cepstra’s Web crawlers read and understand the music from world wide web feed them back to our system. The system then applies unsupervised learning to structurize the information for our predictions. Analyzing RAW audio, we process the tracks through various Machine learning Models at scale, it transforms the raw sound and produces numerous characteristics like time signature, key, mode, tempo, and loudness. After being processed by our pipeline, it provides metrics to understand and define songs and music vectors. This understandingallows Cepstra to compare songs based on those key metrics. For example, someone who likes heavy metal might like songs that are more “loud.“ By combining these two models, Cepstra analyzes the similarity of different songs and artists and can recommend new songs to users every week.

Cepstra’s algorithm analyses the audio of a song to see the similarities between other songs. So, when a likeness of the new song is found to be similar to other songs user like, Cepstra can add it to a User’s playlist. All this at a large scale of processing, we recommend next likely music at O(1) complexity and speed. The algorithm is not only capable of defining the music but suggests music with milliseconds speed.