The research that is part of the Aina Project, promoted by the Generalitat de Catalunya , aims to develop a digital infrastructure for Catalan. At the same time, researchers work to achieve new improvements following the excellence in research developed by the Barcelona Supercomputing Center (BSC-CNS) . One of the specific research cases is the optimization of multidialectal speech synthesis (TTS) models. In the next article that has 3 different installments, the BSC-CNS researcher, Martí Llopart , studies how to optimize a TTS model with multiple speakers for a faster CPU inference, specifically of Matxa and Alvocat , developed by the Aina Project team.
Optimizing a multi-speaker TTS model for faster CPU inference
Part 1
Our goal was to optimize Matxa, a Catalan multispeaker and multidialectal text-to-speech (TTS) model which uses alVoCat as a vocoder. In the end, we obtained a 4.8x speedup for our model, here’s how we did it: ( see the entire article ).
Part 2
In this second part of the blog, I’ll explain in detail how we chose the previously mentioned ONNX settings for intra- and inter-operator parallelism. ( see the entire article ).
Part 3
In this final part of our blog, I’ll explain in a bit more detail some other optimization techniques that we tried, even if unsuccessfully. ( see the entire article )
Access the Matxa-TTS model
Access the Aina Kit
