In my previous blog post, I covered a music generator Mubert, which could create melody samples based on text input from the user. As I mentioned there, this tool allowed to create bases composed from different instruments and sound generators, and all of this was produced based on text description. But one thing was lacking in the feeling of a complete musical piece, it was the addition of vocals. To complete this piece, I will present an AI tool called Kits.AI.
This tool uses a freemium business model, which means some of its functions users can access for free, and for this blog post it will be a function to create AI converted vocals. Without any charges, we can utilize the function to upload our own vocals and then generate desired outcome based on prepared vocal models. After uploading the file, we are faced with different options to convert the audio, and here one disadvantage can be already emphasized, which is the limited number of possible models. This obstacle can be resolved by training your own model, which unfortunately requires a certain amount of music knowledge as well as familiarity with the tool. For my piece, I used one of the prepared models, trying different audio inputs. Overall, I can rate my experience positively, as a basic conversion was relatively easy. It is worth noting that similarly to tool Mubert, with Kits.AI user can also specify preferences regarding recorded vocals by text input, however this option is included in paid version only. Now, with combined melody base and generated vocals, I can state that my song is finally complete.
To sum up this experience with AI tools used in the music industry, it is certainly a facilitation for artists in terms of time of producing samples and amount of generated creative ideas (De Mántaras, 2006). However, those tools must be used according to applicable laws, to not validate rights of artists on which models are based on.
References
De Mántaras, R. L. (2006). Making Music with AI: Some examples. IIIA-Artificial Intelligence Research Institute, 90–100. https://dl.acm.org/citation.cfm?id=1565089
