This repository provides scripts for constructing ASR audio models (Kaldi nnet3 chain, wav2letter++), language models (KenLM in ARPA format), and grapheme-to-phoneme models (sequitur g2p) from speech data sources like VoxForge. It processes speech and text corpora to train models that decode audio to text. Pre-trained models and a Docker-based STT demo service are also provided.