Description

DeepSpeech is an open-sourced embedded (offline, on-device) speech-to-text engine designed for application developers. It enables users to execute the command to download the model files in the current directory, re-export or downgrade the model from the checkpoint using a newer version of the code, ingest speech spectrograms and generate English text transcriptions, as well as transform the character probabilities into textual transcripts with the beam search algorithm. Features include pre-trained English model support, NVIDIA GPU support, augmentation management, transcription management, UTF-8 support, and domain management.