Tamas GPlease, why does Espeak delete the "H" in phrases like "not have that" - I'm not OK with that. I'm not having any of that. Espeak, this is on you, not TGSpeechBox.
@Tamasg You can take espeak out of your processing chain and then do whatever you want. Every other synth has its own word to phoneme rules. I can't think of another one that doesn't. It might be difficult if you want to support all those languages, but it should give you a more flexible system in the end.
@tspivey yeah, the language support is the big one, no way to build out so many without extensive rules, although honestly English is the killer and there's already stress dictionary lookup happening, which is really one step away from just outright straight text to phonemes themselves. It's something actively turning around in my head, just not a good comprehensive G2P that doesn't just use ARPANET, which is very archaic. So yeah, huge project, bit by bit :D
@Tamasg what do you mean? I'm using espeak myself, and I hear the H fine. I'm also using american english, maybe that's why?
@esoteric_programmer so with "not have" it kind of, blends the "H" straight with the end of the Not, with a flap, rather than separating the two like Eloquence. Something I guess my ear picked up on. I don't think it's entirely wrong, some English speakers do connect the two heavily like that, but it makes it also less clear when you speed it up.
@Tamasg hmm, interesting, but also, what's tg speech box?
@esoteric_programmer oh, formant synth I forked off Speechplayer code, now approved to be MIT by NVAccess which is great. https://github.com/tgeczy/TGSpeechBox
@Tamasg awesome! does it work only on windows? I assume it's c++ or something, right?
@esoteric_programmer C++ for the DSP and Frontend, Linux renderer, SAPI engine yeah. Swift and Kotlin for iOS and Android which has been a heap of fun to pick up. The Linux renderer is meant to be flexible with the phonemizer it can pass STDIN/STDOUT with, but uses Espeak and a generic speech dispatcher config. I always build 7 targets now, so it's gotten quite wieldy
@Tamasg do you have some audio samples and maybe linux packages? also, do you know of this new speech format on linux called spiel? with that, you can package your app as a flatpak and people can install that as a speech engine, without having to muck around with config files and stuff
@esoteric_programmer flatpacks are a really cool concept on Linux, even if they can reinstall some dependencies multiple times if the app bundles the lib, but the isolation for easier removal is really worth it. Will definitely look into that, then CI could just build it for me remotely.
@Tamasg about reinstalling dependencies, if you package it with a runtime other apps also use, the set of extra dependencies you pull in is minimal, if not almost 0
@esoteric_programmer I actually explored the whole architecture today, the D-Bus interface, the reference
espeak provider, Flatpak packaging. it's really promising and exactly what would make TGSpeechBox a one-click install for Orca users.
right now the Linux renderer is phonemizer-agnostic (IPA in, PCM out), so you can hook up any phonemizer you want.
that's a strength but it means more setup for users, especially if your distro doesn't ship espeak-ng.
a Spiel provider would bundle espeak statically into one binary, text in, audio out. you lose a bit of mix-and-match
flexibility but gain "install and it works." it's on the roadmap for v3.0 but needs all the Linux plumbing built from
scratch (D-Bus, Flatpak manifest, CI). definitely happening though!
linux tarballs on the releases page work today if you're comfortable with manual setup. They also include an install.sh that tries to copy things in for you, one easier task is making that script a bit more Espeak aware so if it's not on a distro it prompts to install it.
espeak provider, Flatpak packaging. it's really promising and exactly what would make TGSpeechBox a one-click install for Orca users.
right now the Linux renderer is phonemizer-agnostic (IPA in, PCM out), so you can hook up any phonemizer you want.
that's a strength but it means more setup for users, especially if your distro doesn't ship espeak-ng.
a Spiel provider would bundle espeak statically into one binary, text in, audio out. you lose a bit of mix-and-match
flexibility but gain "install and it works." it's on the roadmap for v3.0 but needs all the Linux plumbing built from
scratch (D-Bus, Flatpak manifest, CI). definitely happening though!
linux tarballs on the releases page work today if you're comfortable with manual setup. They also include an install.sh that tries to copy things in for you, one easier task is making that script a bit more Espeak aware so if it's not on a distro it prompts to install it.