Mozilla DeepSpeech

An open-source speech-to-text engine based on Baidu’s Deep Speech research and implemented via TensorFlow.

DeepSpeech delivers a production-ready STT framework that transforms audio into text using a streamlined end-to-end architecture. By bypassing traditional phonetic-to-word mappings in favor of a single recurrent neural network (RNN), the engine achieves high accuracy across diverse acoustic environments. It supports real-time transcription on hardware ranging from high-end NVIDIA GPUs to low-power Raspberry Pi 4 devices. Developers can leverage pre-trained English models or train custom datasets to handle specific vocabularies and accents (e.g., medical terminology or regional dialects).

https://github.com/mozilla/DeepSpeech