Hey there, I was wondering if there are better alternatives than Voice, which is what I use

  • Deemo@lemmy.fmhy.ml
    link
    fedilink
    English
    arrow-up
    3
    ·
    1 year ago

    So kinda heads up my work flow is kinda janky.

    For eleven labs:

    While this may sound stupidly crude I just copy text directly into elven labs website. While eleven labs sounds much much much better than wavenet they are very expensive (I use a few burner accounts). Currently I really only use them for J-novel prepubs parts. I would not consider/recomend converting a whole book with them.

    For wavenet:

    Disclamer their is probably a less janky way of doing this.

    Prereqs:

    Android phone with aiitts (replaces text to speech engine with wavenet) and Librera Reader FD https://www.milmike.com/aitts-google-wavenet-voices-for-android-tts https://f-droid.org/en/packages/com.foobnix.pro.pdf.reader/

    A google cloud account with wavenet api key

    1. I convert the book/text I want into a pdf (using tools like calibre)

    2. Set up ai tts (import your api key and pick your voice I’m curently using US Wavenet F)

    1. I open the pdf in Libera reader and use its Record TTS

    1. Open the folder in Smart audio book and enjoy listening

    Side notes:

    1. You get about 1 million characters free with wavenet (about two light novels roughly)

    2. Make sure to test the play button in Libera reader first if you don’t hear audio you may need to check if your plugin is installed properly (aitts). Alternatively if it isn’t working its possible you may need to create a pdf with larger text (tldr the plugin can send to many queries rapidly if text is to small causing wavenet to reject them)

    3. If your doing a whole light novel (or book conversion) do it in chunks of 50-100 pages. After each block conversion check if the files are a few kilobytes. If you see something like 413 bytes its more than likely that file is corrupt due to wavenet not liking the text (ex to small or font, etc), you may have to tinker with the pdf a bit.

    Future hopes:

    I in the future I kinda hope to figure out a better way to handle wavenet. I have tried some programs in github but It was kinda hit /miss and got busy. Possibly might look into coqui ai (its a self hosted equivalent to wavenet) down the line:

    https://github.com/coqui-ai/TTS

    • Sisyphus
      link
      fedilink
      English
      arrow-up
      2
      ·
      1 year ago

      Hey, that’s awesome that you shared so much. I’ll dig into it and see what I can use it for. Those voices are really impressive. I listen to audiobooks enough that I could never afford to use this method regularly, but it’s very cool.