Spoken Language Recognition on Open-Source Datasets

Follow the full discussion on Reddit.
This was my capstone project at SMU. While researching this topic, we found that most papers in spoken language recognition used private or expensive datasets, limiting reproducibility. We also found that many papers made performance claims without comparing to other recent research. We attempted to address these problems in this capstone project. Using public sources such as Mozilla's Common Voice, we constructed an eight-language dataset and compared three neural network architectures published in the last few years.