This repository implements SummaryMixing, a simpler, faster and much cheaper replacement to self-attention for automatic speech recognition (see: https://arxiv.org/abs/2307.07421). The code is ready to be used with the SpeechBrain toolkit).
This is the PR to update the code to the v1.0 of SB. With this change, we lose the capacity to replicate the results of the previous paper. This is why I created a branch that freeze the SB version that we used for this paper.
This change is necessary so that we keep using summarymixing with the latest version of speechbrain.
This is the PR to update the code to the v1.0 of SB. With this change, we lose the capacity to replicate the results of the previous paper. This is why I created a branch that freeze the SB version that we used for this paper.
This change is necessary so that we keep using summarymixing with the latest version of speechbrain.
I tested the recipe, they run.