Description

DeepSpeech 2 is an accurate speech recognition system designed for language translation. It enables users to manage speeches including noisy environments, accents and different languages, convert a probability distribution over characters into text through a decoder, organize streams of data from user requests into batches, ingest speech spectrograms and generate English text transcriptions, reinstate pipelines of hand-engineered components with neural networks, as well as optimize error rate and recognize Mandarin speech. Features include performance management, Bayesian normalization capability, image processing capability, multiple GPU support, AN4 support, audio augmentation, training graph visualization, dataset customization, and Librispeech support.