Spaces:
Runtime error
Runtime error
On Hokkien Varieties
#8
by
AdamNg
- opened
I was wondering if the developers have considered dealing with other varieties of Hokkien in the speech-to-speech translation system, like mixing them into one model or with other approaches.
As a speaker of Quanzhou Hokkien, I found the result generated from my accent largely biased, probably due to the difference in intonations and words. Regarding now it is a piece of cake to process English varieties, I guess it would not be too difficult to build a similar system supporting more Hokkien varieties, including decoding and generating speeches.
This model was mainly trained with Taiwanese Hokkien audio data. Details are in https://research.facebook.com/file/799432337944526/Speech-to-speech-translation-for-a-real-world-unwritten-language.pdf