Audio-Keyword-Spotting-in-Chinese

Combine Recurrent Neural Network and Convolutional layer on audio data. The target keyword is "雅婷姊".

---

Baseline Model

I. One convolutional layer

II. Two GRU layers

III. Several dropout and batch normalization layer

IV. Fully-connected layer with sigmoid

Proposed Model

I. One convolutional layer

II. Two GRU layers and modified first into bi-directional and residual

III. Several dropout and batch normalization layer

IV. Fully-connected layer with sigmoid