[Discussion] Spanish support & recommendations

[berengueradrian]

Hi! I had another question regarding multilingual wake word performance, specifically for Spanish.

My application is fully Spanish-speaking, so I was wondering whether there are any recommendations, docs, or best practices for training wake words optimized for Spanish phonetics/acoustics.

Some things I’d especially be interested in:

• Whether there are recommended base/pretrained models for non-English wake words
• Any known datasets that work well for Spanish wake word training/evaluation
• Whether the current pipeline is language-agnostic enough or if additional fine-tuning is usually recommended
• Any tips regarding phoneme selection or wake word structure for Spanish

Would love to hear if there’s already guidance around this or if it’s something still being explored.

Thanks again.

[pham-tuan-binh]

Right now the only supported method is with VoxCPM samples generation.

We don't have a specific backbone for other languages yet. I'm also not aware of Spanish datasets that's suitable for this. However, for words that sound similar across both spanish and english, it should work well.

The current pipeline should be agnostic enough to be usable but it really depends on your usecase whether it's reliable enough. Additional fine-tuning on your own datasets is recommended.