Qwen 3 TTS merging and MUltispeaker finetuning

[arpit763]

Hi,
i finetuned Qwen3-TTS-12Hz-1.7B-Base on Shona language, 120hrs of data on a single speaker, its performing nice but have foreign accent.

• i want to understand how can i finetune it on multiple speaker data and then choose speaker during inference.
• i am not able to perform my stage 2 finetuning after i merge the adapter with base model.

[atulyaatul1999]

@arpit763 were you able to get this working?

[arpit763]

no, i was not able to solve any of my problem.
even 120 hrs of data is less for FT.