pied
pied copied to clipboard
Elleo
It skips letters, does not read the end, does not read at all in Firefox - Polish language, on Arch Linux
I installed TextToSpeech Pied on my Arch Linux, and selected the Polish voice. The voice is available via the `spd-say "Hello world!" command, but it skips some letters and won't read longer texts in full. In Firefox Reading Mode is available, but it doesn't even start reading at all. How to fix it?
Hi @look997,
This sounds like it might be a bug with the polish voice model or the phonemes being selected for use with it. I'll try to find some time to investigate this when I can (I'm a bit lacking in free time at the moment I'm afraid, so it might be a while before I can get to this)
@Elleo Additional info. I checked out Piper in Piper Voices (experimental) in the Read Aloud: A Text to Speech Voice Reader extension. There, reading in the Polish voice works much better, although it is still not perfect, it loses some characters in places.
This doesn't use spd-say, just this https://piper.ttstool.com/ which opens in the background.
In the Read Aloud: A Text to Speech Voice Reader extension, you can also use the voice installed in the system directly, through Standard Voices Piper, then it still works as bad as I described in the first entry, just like when I use spd-say.
Oh, and some info that in Firefox Reading Mode, the voice however reads in this mode if I restart Firefox, after each voice change in Pied. But it still has the same problems with voice quality, as when using spd-say.
I've encountered the same (or a similar) problem with Mandarin. The the Chinese TTS model (Huayan) can speak something with spd-say -- it's not English or Chinese, but some audio comes out -- but no output comes out when the input is in Chinese. Again, the model works perfectly fine with the Read Aloud plugin on Firefox.
Same as @atkirtland here, Chinese characters in the input are completely skipped by Huayan for some reason, can't figure out why
Any news on this? can confirm that spd-say ignores ąóżźóńłćń symbols. There is not model problem, we can test the model here https://huggingface.co/spaces/hossamdaoud/text-to-speech-2025 (just select polish and voicemodel, i checked with csukuangfj/vits-piper-pl_PL-gosia-medium) @Elleo Could you pls suggest what can cause an issue?
