Tomoki Hayshi, "Duration-controlled LSTM for Polyphonic Sound Event Detection"

2017/02/05 19:33 に 後藤太一 が投稿

In this study, we aim at sound event detection.
To deal with problems such as cannot represent actual duration of each sound event and decrease performance under noisy condition,
we proposed new method that combine Sound Activity Detection Network with BLSTM-HSMM.
Result of experimental comparison, proposed method shoe higher performance than exiting techniques.
Henceforth, apply to real & big recording dataset.