Applio speaker embedding

[icy bough]

I am using Applio to create a clone of my voice for singing, and although the original file is very good, it has given me a voice that does not pronounce words well. Which speaker embedding should I use to get a good result? Thank you.

[dapper crane]

• if not sure, use the default contentvec
• depend on the language as it is mostly optimized for English
• make sure the model dataset and the inference sample articulate well

[icy bough]

Thanks. If I wabt to use a dataset with two linguage? Is it better make two different model? One for every linguage?

[dapper crane]

I wouldn't say a single bilingual dataset really bad, you gotta try and see

[icy bough]

I'm doing some tests. The Italian accent isn't very good; it sounds like an Englishman singing in Italian. Which speaker embedding could I use for Italian?

[oblique meadow]

How long is your dataset?

[icy bough]

7 min

[oblique meadow]

Does it only consists of you singing or also speaking?

[icy bough]

singing in italian and english and speaking in italian and english. But the sing parts are longer.

[oblique meadow]

If you really want the output to sound Italian then I'd recommend to only use Italian in your dataset and use the index to improve accent

[icy bough]

Ok. What do you mean with use the index? I use replay to create cover. Don't know if this software use the index file too...

[oblique meadow]

I've never heard of that software

[oblique meadow]

Why don't you use Applio to make covers

[icy bough]

I'm learning how to use it now and I'm starting to make my model voice.

[dapper crane]

you may want it longer to improve

[icy bough]

they are 7 min of HQ sound anyway.