a few days ago I released a Python library that wraps a machine learning model for spelling English words and sounding them out, even if those words aren't in the CMU Pronouncing Dictionary: https://pincelate.readthedocs.io/en/latest/ I'm still working on some of the documentation, but since I just gave a talk about it at SLSA, I thought I'd mention it here too!
this library is based on the model and code I've been using to produce all those weird phonetic stretching/tinting experiments I've been posting on here over the past few months. there's nothing especially technically interesting about the model architecture, but the library (I hope) is fun and easy to use and is specifically designed to facilitate weird creative applications
@aparrish do you have suggestions for phoneme-to-speech software? I did a little experiment to generate fake English, and the trickiest part was finding something to generate sound files from the generated phonemes… ended up using Alexa, but would prefer something local. http://www.eiman.tv/blog/posts/marklish/
@aparrish this is great! would it be possible to put a list of the phonetic feature abbreviations in the docs? I know they're in the code, but I am extremely lazy
@mewo2 that is a really good idea, I will add it to the todo list!
Hometown is adapted from Mastodon, a decentralized social network with no ads, no corporate surveillance, and ethical design.