Issue: Speaker Token Not Locking Voice Consistently in Veena TTS
Hi team,
I'm using the Veena TTS model and specifying the speaker token (e.g., ) at the start of each prompt. However, the generated voice seems inconsistent — it often switches to a different speaker mid-generation or produces varying tones. It looks like the model doesn't reliably lock onto the intended speaker voice, Is there a way to enforce consistent speaker conditioning throughout the output?
Thanks in advance!
Yes facing this issue a lot. Tried to use other voices but looks like its voice is automatically changing to default voice. Problem is no matter how many times you try, sentences which are converted in default voice will not convert TTS to selected voice.
I am facing the same problem.
one observation though the voice maitri is consistent and rest are not