After revolutionizing the universe of texts and images, the new frontier for Artificial Intelligence seems to be sound, with various tools that promise to revolutionize the world of musicincluding a Latin American experiment that created a synthetic singer from voice samples of popular singers from the region.
And while many of the advances and news about Artificial Intelligence in 2023 were focused on massive language models like Chat GPT or platforms like Dall-E, in the coming months we will surely hear more about cloning sounds, songs and voices.
These are slower advances because generating music proved to be more difficult than expectedsince it requires mastering many more components, from the tones and notes of the melodies to the voice and the arrangements.
Furthermore, it is much more difficult to find data available free of copyright and high quality to train systems.
In recent weeks, both Microsoft, Meta and Google announced digital tools that are still in beta mode but that can Create songs and sounds by simply typing descriptions (like “country broken heart song” or “80s metal rock song about traveling on the road”).
In addition, they grow on YouTube and SoundCloud apocryphal versions such as songs by Fito Páez sung by Charly García or cumbia songs with the voice of Lana del Rey.
However, these experiences seem to be only a playful first step towards their true possibilities.
Meanwhile, there is no doubt that they are putting artists on alert now the music industry in generalattentive to the violation of rights.
It is also true that this technology has the potential to democratize the pleasure of creating music without the need to know how to play an instrument or master a complicated software program.
Attentive to the potential of these tools, was born LATAM Singing Voice, “a practical experiment on voice cloning”according to its creators, the Larrosa and FUTURX teams – responsible for the first report on the use of AI in Latin America – and the artists Berenice Llorens and Pol Lain.
The model was trained with nothing less than fragments of you from Mercedes Sosa, Violeta Parra, Chavela Vargas, Tita Merello and copleras from northwest Argentina, among other sources.
That dataset allowed to create a unique expression which, according to its creators, contains the essence of each and every artist.
The first song created was Sometimes with sorrow, sometimes with hopewhose composition was also the result of an experiment with AI, with a lyric with copla meter created with Chat GPT 4 and a rhythmic, timbral and melodic base produced through audios produced by the generative samples platform Ktonal.
The best thing is that anyone can try and play with this trailer. Simply register for free at www.linkfeat.ailook for Voz Cantora Latam V4move up an audio that we have created (it can be simply with the cell phone) and listen to the results.
A new way of approaching music is born.
