The Norwegian word “påsto” when clicked, the text to speech seems to read some underlining code or something instead of the word. Something like “vertikal strek p å sto vertikal strek”. I tried editing the word in the sentence, it doesn’t fix it.
I have noticed this for a couple of words before but can’t remember which ones, if I encounter them I’ll post those here as well.
As a side note to this, I know in sentence mode you can play the audio of the sentence and it auto finds it from the uploaded audio. If it is at all possible to have this as a text to speech option that would be great, so clicking a word will play the word as it is in the uploaded audio file, but maybe that is too complicated?
You might be sick of me by now, but I think I can give you a better answer than anyone else on this forum.
The text to speech issue is a known problem of those AI voice generators. Sometimes it just says something completely unrelated or garbled. Unsure if LingQ can do anything about it other than try workarounds for specific words.
LingQ uses Aeneas for automatically synchronising audio and text. The output does not contain word level timestamps so it probably won’t happen. There are other options but they require a much larger computer to run efficiently.
Whisper generated transcripts and YouTube autogenerated subtitles have precise word level timing but LingQ doesn’t take advantage of these yet. That would be the best bet for the native audio for words
In this case it’s not complete gibberish, it’s basically reading out something like “| p å sto |” instead of “påsto”, but if there’s nothing they can do then it is what it is I guess. Maybe someone at lingq can chime in.
Fair enough. I thought it might not be possible. Thanks for the info though.
Sorry. We have fixed the TTS for that word now. There are some recurring TTS issues in Norwegian for this voice. We are going to be working on resetting all words so this problem goes away. Unfortunately, the problematic words get cached and remain until a process is run to clear this cached version. Once cleared this problem should not reoccur.
Ye I realised the issue is probably related to |word| being sent as text to speech as a partial workaround for those AI voices. Interesting if you can find a 100% solution.
The word timestamps issue was from this section of OPs post where he asked whether the individual word could use native audio instead of TTS. Seemed like a pipe dream outside of Whisper and YouTube autogen imports.
Thank you! It’s rare enough to not be the end of the world, but glad you will be able to fix it in time. I do wish there were better “free” TTS options available though.
