535.535. Speech Recognition Basics

Speech recognition systems use acoustic features such as MFCCs to classify spoken words. In this problem, you will implement a basic speech recognition system using a neural network.

**Example:** Train a neural network to recognize spoken digits using a dataset of audio recordings.

**Constraints:** Use the Keras library to build the neural network, and implement a custom loss function to optimize the model.

**Note:** You can use the librosa library to load the audio recordings, but you should implement the speech recognition system from scratch.