Tailored Design of Audio-Visual Speech Recognition Models using Branchformers

Official model checkpoints for the paper "Tailored Design of Audio-Visual Speech Recognition Models using Branchformers". Checkpoints for our audio-only, video-only, and audio-visual models are available along with their corresponding model configuration files. The same for the LMs used during beam-search inference.

Source code to evaluate our models, fine-tune them, and train new ones for your database of interest can be found in our official GitHub repository.