Guitar Solo Transcription

Given a clip of guitar solo, we tried to transcribe not only the melody but also the playing techniques, such as bend, vibrato, hammer-on, ..., that are used in the melody. Therefore, we designed an algorithm that detects the appearance of playing techniques as well as doing note tracking.

Weakly Supervised Audio Event Detection

In this topic, we try to discover not only the sound events happened in an audio clip (clip-level information) but also the temporal positions of the detected sounds (frame-level information). Since creating frame-level annotated data can be extremely time-consuming, we proposed a model based on convolutional neural networks that relies on data with only clip-level labels (weakly supervised data) for training.