Title: Research @ AudioShake

Research scientists from AudioShake: Ondřej Cifka, Luke Miner, Hendrik Schreiber, Fabian-Robert Stöter, Oliver Valery, Cheng-I Wang

Abstract: 

At AudioShake, we use machine learning to open up audio recordings into stems and transcriptions. This makes them more interactive, accessible and useful.

Our team of scientists at AudioShake come from a variety of academic backgrounds and are all still active members of the MIR and signal processing communities. This active participation enables us to stay at the forefront of developments in the field and to continually improve our methods.

We provide an in-depth insight into the cutting-edge research and methodologies used by our team. As a group of researchers, we specialise in the research and optimisation of audio separation and transcription models for speech, music and generic audio domains.

We begin by highlighting our work on high-quality stereo stem separation, which has reached the state of the art. We then look at specialised dialogue-related tasks such as DME and music removal. We present our approach to multi-speaker and multi-singer vocal separation before moving on to our other main research area: transcription. We introduce the task of formatting aware multi-lingual automatic lyric transcription and demo our state-of-the-art transcription models.