so on my linux desktop I use scripts with xsel and cepstral swift to easily speak selected text (I do similar on my android) but the problem I run into is getting a decent tts on the librem 5. The options I have found that can generate voice dynamicly are robotic voices from things like festival or espeak-ng . you can generate a good wav file from kokoro-tts but that takes time and is not just quickly reading back selected text. So with that said I wondered if anyone has any other ideas on what to try for dynamic reading of text that would at least be as good as what my android phone [grapheneos] does (I use @voice aloud reader )
On Librem 5 I use xclip with espeak in my scripts to tts
#!/bin/bash
xclip -o | espeak
espeak let’s me listen text in greek, spanish, hindi, danish, italian …
It’s fine for me.
I am not a fan or LLM’s bad called AI.
One option that gives pretty good results is piper. I haven’t checked how long it takes on the Librem 5 though.
yes this is the current rub.. there is good sound or there is fast enough but I have not found one that is easy to hear yet not slow (peper sounds great but is very slow)
Would be a solution to offload piper to a more capable external device and stream the results back to the Librem 5? I certainly know having the Librem 5 in a Wireguard VPN with your own servers works pretty reliable for all kinds of stuff. So if you had another device with enough performance to process piper faster, that could do it remotely.