google: Explained: What is Google’s Tone Transfer tool

Google’s annual developer meet, the Google I/O 2022 event has recently ended. The tech giant showcased multiple breakthroughs in new AI tech and security in this year’s event. Google’s Tone Transfer tool, is one of these advancements that was released in 2020 and also made its way to the stage during the pre-show of the Google I/O 2022 event. This tool is developed by Google’s Magenta team and it allows users to turn sounds of normal instruments or even random sounds into something different.
What is Google’s Tone Transfer?
The Tone Transfer tool is developed by Google’s Magenta team and was introduced by Google in 2020. This tool allows artists and creators to input various sounds and get something different out of it. For example, users can feed the tool with sounds like — people singing, birds chirping, or utensils banging and can convert them into sounds of musical instruments like a flute or violin.
How does it work?
The Tone Transfer tool project depends on a machine learning process called Differentiable Digital Signal Processing or DDSP. While testing the tool, the team introduce the model with a 10-minute recording of a violin to process and run through its system. Later on, the team started monitoring the progress of the program and its capability to recreate the sound. The developers found that after an hour of training, the model was able to recreate the sound similar to a synth being played. Moreover, after 10 hours of training, the tool was able to replicate the sound of a violin almost indistinguishably. The model can control several different audio signatures provided by the instrument to help it successfully recreate the sound. This project shows how far machine learning has progressed as AI gains the capability to develop and reproduce musical instruments by itself.
How to create sounds in Tone Transfer?
Google’s Magenta team has made Tone Transfer available for the public to try out for themselves. Users can either choose a sample offered by the program or can even upload their tracks. After this, the users will have to choose the instrument sound into which the track will be morphed. The available result options are — Flute, Saxophone, Trumpet and Violin.
Firstly, users have to choose a source sound and will have to click the Play icon under the sound signature to get an idea of what it sounds like. Next, they have to choose an output sound and click on Play again to hear the sound recreated and turned into a new instrument.
Users will have to select the “Add your own” option under the source sample section if they want to upload their sound sample. Users can either record a sound or upload a file from their PC if their device is compatible. Click on the Done button after the recording or uploading has finished, it will allow you to listen to the recorded or uploaded track(s). The Tone Transform tool will start processing the sample once you are happy with it and click on the Transform button.
However, make sure you are choosing a track which has only one tone playing at a time. This will help the model differentiate between various tones and create a better result. The tool also offers a Slider icon that can be used to tune the final result. Users can also adjust the mix, octave and loudness at their discretion.