TensorFlowASR

Why we included this project

For teams already working in TensorFlow, this is one of the more complete open-source options for training your own speech-to-text models. It implements a solid set of modern architectures, including Conformer, ContextNet, streaming transducers, DeepSpeech2, and Jasper, and supports both CTC and RNN-Transducer training paths, so you can pick the end-to-end setup that best matches your latency and accuracy needs. The TFLite conversion support matters if you plan to move a trained model onto phones or other on-device targets, where cutting memory and compute count. The repo comes with example configs, pretrained checkpoints, and tutorials for English and Vietnamese, which shortens the jump from reading a paper to reproducing its results. It is a research-oriented codebase, so expect to wire up your own datasets and tune models yourself rather than calling a hosted API.