Can I only use deepvoice3 to learn dataset embeddings?

[george-roussos]

Basically what the title says. I am experimenting with multi-speaker adaptation to unseen speakers and my setup has 4 cores. I was wondering if I could use the pipeline to only learn the speaker embeddings in the dimension space and not train any TTS model, thus only leveraging the GPUs for the embeddings task, so to save time and possibly get better performance.