Issue link: https://digital.copcomm.com/i/1081428
www.postmagazine.com 15 POST JAN/FEB 2019 TECHNOLOGY algorithms with large, carefully-selected databases. AI can then reference this information to learn, identify and sep- arate speech or other specific types of audio content from complex mono or multi-channel mixes. The ability to train and improve AI helps us to separate these elements at the highest quality imaginable." Can you discuss some of the biggest or most common audio challenges that AI has solved so far? "Audionamix AI can separate various types of content from complex mixes. In daily post work, AI helps to preserve speech in a recording and separate it from complex evolving background noise. This can help in a post production context, but it can also empower other speech-related solutions, e.g. speech recognition algorithms and speech to text conversion. On the music side, Audio Source Separation AI can help to rebal- ance songs, fixing mixes and lowering the volume of the vocal in a song so it won't compete with dialogue or voiceover in a larger session. This is important, especial- ly in mixing songs for commercials, TV and film production. Lastly, the ability to create stems from stereo or mono mixes means that old recordings can be revised and that the universe of available samples for production is now unlimited." What acted as the catalyst for Audionamix becoming interested in AI? "The idea to use AI in our work came from our researchers based in Paris. They decided to move towards machine learning AI as it was a very efficient solution which could improve upon the probabilistic approach we were using before. At first, we felt it would be good to use this machine learning as a component of the full sep- aration process, exclusively for automatic voice activity detection, so that it could help us to identify where vocals exist in a song. This kind of analysis was needed to proceed with an accurate vocal separation. As we saw its potential we began working with it more and more, and now machine learn- ing AI is an integral part of our latest, and most ad- vanced, separation algorithms. What do you see coming down the road from Audionamix that will be making use of AI? "We are continuously improving the quality of our AI-based products, and are thrilled by the results of the latest algorithms. Our goal is to make the results continually more consistent and independent from the complexity of the original source. We are taking another big step forward with fully-au- tomatic, AI-based vocal separation and instrumental creation, which we are planning to offer to our users very soon. We are also working on enhanc- ing all of our algorithms with realtime processing. Future challenges include targeting complex problems, including multi-speaker speech separation." Some feel as though AI is taking the creativity out of roles, specifically the editing process, by overly automating it. What are your thoughts? "I would argue that AI acts as a shortcut to creative processes, by saving time and getting to the target easier. AI is not able to fully replace the human creative element. As a composer, I believe that the human factor will always be important. Even if in the future human creativity will be largely empowered by AI, an artist will still be making artistic decisions by acting more as a creative director. But before we get there, AI will act as a hardworking assistant, replacing time consuming and tedious editing pro- cesses, saving more time and energy for new projects by offering faster workflows. The future looks very interesting and we are planning to stay avant-garde by offering audio solu- tions empowered by AI." "IN POST WORK, AI PRESERVES SPEECH IN A RECORDING & SEPARATES IT FROM COMPLEX BACKGROUND NOISE." Maciej Zielinski