Hi @ken107! I highly appreciate your work on this plugin!
Some TTS engines, especially neural network-based such as Piper TTS, don't work extremely well when text is being split into many chunks. It takes a noticeable amount of time to begin speaking every chunk. Therefore, the preferred way of interacting with them is often by submitting multiple sentences and paragraphs at the same time in a single chunk, so that while the first sentence is spoken, they can generate next sentences and, once ready, append them to the speech stream
Given that this is a likely direction where text-to-speech is going, and because I'd really like this feature myself, I'd like to add a support for an option to merge texts into a single chunk
The current PR is adding an dropdown named "Merge or split text" that is saved per voice with the options to either
Split text into multiple chunks (which is a default, works the same as before), or
Merge text into one chunk
Here's the screenshots:
Now, there is alternative and possibly better solution to the problem. Namely, we can have an input field for customizing how many paragraphs can be in a single chunk, as well as an option to make it infinite (e.g. if "-1" is specified). I'm totally willing to implement that if you think that's better!
Additionally, I'm not confident that it's wise to support this customization on anything other than native TTS engines. If you'd like me to restrict it to only native TTS engines, let me know and I'm happy to give it a go (some minor pointers would be appreciated here)
Thanks again for making and supporting this extension and for taking a look at this PR!
Hi @ken107! I highly appreciate your work on this plugin!
Some TTS engines, especially neural network-based such as Piper TTS, don't work extremely well when text is being split into many chunks. It takes a noticeable amount of time to begin speaking every chunk. Therefore, the preferred way of interacting with them is often by submitting multiple sentences and paragraphs at the same time in a single chunk, so that while the first sentence is spoken, they can generate next sentences and, once ready, append them to the speech stream
Given that this is a likely direction where text-to-speech is going, and because I'd really like this feature myself, I'd like to add a support for an option to merge texts into a single chunk
The current PR is adding an dropdown named "Merge or split text" that is saved per voice with the options to either
Here's the screenshots:
Now, there is alternative and possibly better solution to the problem. Namely, we can have an input field for customizing how many paragraphs can be in a single chunk, as well as an option to make it infinite (e.g. if "-1" is specified). I'm totally willing to implement that if you think that's better!
Additionally, I'm not confident that it's wise to support this customization on anything other than native TTS engines. If you'd like me to restrict it to only native TTS engines, let me know and I'm happy to give it a go (some minor pointers would be appreciated here)
Thanks again for making and supporting this extension and for taking a look at this PR!