Optional Prefix Audio

[MarisKay]

Not sure how this has to work , but if the prefix is in seductive slow tone, the generated part is nothing like it. Shouldnt it pick up the pace, intonation and feel?

[darkacorn]

did you have the prefix also transcribed as prefix for the text ?

[MarisKay]

umm, i guess not, how should i do that? Please give an example or point to where can i read the "how to" . Or wait, should i just type what the prefix says, without any special markings or something? I think i tried it once too, but the voice was not the one that is in the source voice, prefix voice was different person's and output was like my prefix voice and then - totally different tempo and feel - my source voice's generated rest part of the text. Result felt like two audio pieces from different persons were just randomly sticked together.

[darkacorn]

you have audio prefix in wav.. and the transcription of that has to be the prefix on your regular text too otherwise the prefix wont condition much

[coezbek]

This is also explained in #14 and there is an implementation in #148

When using the prefix audio (not reference audio) in the gradio interface, you also need to put in the text of the prefix in the text box.

[MarisKay]

thanks, will try, so it should pick up the tone and feel of the prefix audio, no matter that it is spoken by some other voice, not the one in reference?

[xdevfaheem]

umm, i guess not, how should i do that? Please give an example or point to where can i read the "how to" . Or wait, should i just type what the prefix says, without any special markings or something? I think i tried it once too, but the voice was not the one that is in the source voice, prefix voice was different person's and output was like my prefix voice and then - totally different tempo and feel - my source voice's generated rest part of the text. Result felt like two audio pieces from different persons were just randomly sticked together.

i have commented throughout the code #148 check 'em out

[MarisKay]

okay, i added a prefix audio, transcribed it in front of my speech text and generated. There is absolutely no link between the emotion, speed and feel of prefix audio and what comes after in generated results. None. Like voices from two different life situations simply sticked together one after another. Something is not working or the implementation is that weak so it cant deliver its initial intention.

[petermg]

okay, i added a prefix audio, transcribed it in front of my speech text and generated. There is absolutely no link between the emotion, speed and feel of prefix audio and what comes after in generated results. None. Like voices from two different life situations simply sticked together one after another. Something is not working or the implementation is that weak so it cant deliver its initial intention.

I agree. I just tried this and all I am getting generated is the exact same audio I put in for the prefix. I don't think it's supposed to work that way. It seems to be broken.

[coezbek]

Can you share your code of what you tried?

[MarisKay]

I dont have any code, i used included in git gradio interface. There is no code. All done through UI