add a pause in a sentence

[fakirsu]

Hello,

In my workflow, I need to add silences to the generated TTS audio (at several points, in the middle of the spoken text).

I've tried many things, adding symbols like \n, \n\n, ..., (...), -, _, etc., at the desired silence points.

But it makes absolutely no difference. The TTS is still spoken without any silence, or even a breath, at the intended locations.

Do you have any ideas?

PS: Of course, it's possible to add silences afterward, using an alignment method, ASR, etc., but that's not what I'm looking for because these methods (even with sophisticated volume analysis techniques, etc.) lack precision.

Similarly, it's possible to chunk the spoken text, but that's not what I want either (because the execution time is too long if I have to chunk it, even with heavy batching).

In short, I'd really like to be able to tell the model, "Take a short break right here."

By the way, kudos to the developers !!! I've tested almost every text-to-speech (TTS) model, and this one is exceptional, with reasonable inference times.

[wulaoshi]

I have the same requirement. Have you found a solution yet?

[Finrandojin]

You can try using "pause after [word]" type of instruct but it is not 100% percent reliable.

[mcrumble1961-boop]

I found a solution to my problem with inconsistent pauses using Qwen-TTS. If I place a sentence in quotes and add spaces where I want to pause.

Ex. "Hi! I'm so-and-so. Welcome to my page. "

The number of spaces seems to determine the length of the pause.