Introducing Wav2latter – Towards Data Science

Speech recognition systems has been one of the most developed areas in the deep learning ecosystem. The current generation of speech recognition models rely mostly on recurrent neural networks(RNNs) for acoustic and language modeling and on computationally-expensive artifacts such as feature extraction pipelines for knowledge building. While RNN-based techniques have proven to be effective in speech recognition tasks, they require large volumes of training data and computing power often resulting prohibited for most organizations. Recently, Facebook AI Research(FAIR) team published a research paper proposing a new speech recognition technique based solely on convolutional neural networks(CNNs). The FAIR team went beyond research and open sourced the Wav2letter, a high performance speech recognition toolkit based on the fully-convolutional method. The use of CNNs in speech recognition systems is not new in the deep learning space but has been mostly constrained to specific tasks and if often combined with RNNs into more complete architecture.