In their first foray into the music production world, Bytedance - the company behind social media platform TikTok - have announced the development of a curious new plugin.
Mawf uses machine learning to 'morph' incoming audio signals into emulations of real instruments in your DAW. The plugin's ML synthesis engine can also run on MIDI input alone. This means you can use Mawf as an effect to colour an existing sound, or use it as a virtual instrument by itself. In its beta version, Mawf offers models of three instruments: saxophone, trumpet and the khlui, a Thai flute.
The developers behind Mawf used machine learning to analyse recordings of professional musicians playing instruments. The ML models extracted expressive changes in the instrument's sound that were linked to variations in pitch and amplitude. Mawf then uses these trained models to approximate the sound of these instruments based on input provided by the user. They're keen to distinguish this from physical modelling synthesis, which requires "specialised equations" for each instrument modelled. Mawf needs only a solo recording of the target instrument in order to imitate it.
Though the range of modelled instruments is a little small, Mawf does feature some interesting additions. There's an in-built compressor, chorus, and reverb effects, and a number of Control Modes that adjust how Mawf's synth engine is triggered, allowing the user to control the pitch of the processed audio through MIDI.
In a statement on their website, the developers commented on the current limitations of the technology: "Like the first ever analogue synthesiser, expect some funky bleeps and bloops from Mawf. ML for audio synthesis is a new technology no one has really perfected yet."
Mawf can be downloaded for free by users outside of the U.S., but beta testing is limited to the first 500 sign-ups, so we suggest moving fast if you'd like to snag a copy.