• Innovation through disruptive and scalable technology .
  • Cutting-edge AI research .
  • Accelerating innovations in research and service .
  • We strive for (and achieve) excellence! .
  • “SotA” (State-of-the-Art) .
  • Visual demo of research and service innovation
  • Human. Machine. Experience Together .

Data Augmentation and Model Optimization for Piano Transcription


ISMIR 2019


* MIREX Challenge (Ranked 1st)

We present two data augmentation methods that are suitable for an automatic piano transcription. After deep learningbased approaches are applied to the diverse music information retrieval problems, the performance of automatic piano transcription is also improved. However, a lack of dataset with various music genres causes the difficulty in model generalization. To solve this problem, we analyzed the two major piano transcription dataset and suggest data augmentation methods in both symbolic and audio domain based on Onsetsand-Frames architecture. Also, to maximize the performance, we tried the meta learner system to find out the best hyperparameters, and used model ensemble.

More Detail: Click Here