speech-recognition

• Currently data is stored at data/audio and data/label. Maybe can follow LibriSpeech directory structure.
• AudioDataset - read data from directory and convert to spectogram. Variable length.
• dataloader_audio.collate_fn pad sequence. Yet to check compatiblity with nn.
• utils.TextProcess to clean text. Currently include to lower case, remove punctuations, numbers(including years) to words.