this post was submitted on 26 Oct 2023
114 points (99.1% liked)

Linux

47308 readers
575 users here now

From Wikipedia, the free encyclopedia

Linux is a family of open source Unix-like operating systems based on the Linux kernel, an operating system kernel first released on September 17, 1991 by Linus Torvalds. Linux is typically packaged in a Linux distribution (or distro for short).

Distributions include the Linux kernel and supporting system software and libraries, many of which are provided by the GNU Project. Many Linux distributions use the word "Linux" in their name, but the Free Software Foundation uses the name GNU/Linux to emphasize the importance of GNU software, causing some controversy.

Rules

Related Communities

Community icon by Alpár-Etele Méder, licensed under CC BY 3.0

founded 5 years ago
MODERATORS
[email protected]
[email protected]
[email protected]
[email protected]
[email protected]
[email protected]
114
TTS voices that sound nice? (thorsten-voice.de)
submitted 10 months ago* (last edited 10 months ago) by [email protected] to c/[email protected]
27 comments fedilink hide all child comments
 

There are multiple screenreaders but their voices simply suck. KDEs espeak has way worse voices than Android eSpeak for some reason.

There is Thorsten Voice, a grest project that sounds really nice.

Not sure if he has a espeak dataset too.

Do you know any others?

Its essential for blind people, I dont think any blind person uses Linux if they have to hear these robot voices every day.

top 27 comments
sorted by: hot top controversial new old
[–] [email protected] 32 points 10 months ago* (last edited 10 months ago) (1 children)

espeak default voice backend is synthesized without using actually real voice samples. So it doesn't require downloading a huge package for each language, which is convenient in some cases, but the outcome is extremely robotic.

You can use MBROLA as backend for espeak so that it uses some voice samples and the result should be less jarring (it'd still be easy to tell it's not natural voice, but at least you'd be able to understand it better). There's a tutorial on this here: https://github.com/espeak-ng/espeak-ng/blob/master/docs/mbrola.md

Or you can try piper (https://github.com/rhasspy/piper) it's one of the most natural-sounding TTS (here are some samples).

[–] [email protected] 6 points 10 months ago

Piper is pretty good, thank you!

[–] [email protected] 16 points 10 months ago (2 children)

In addition to the very insightful reply by Ferk: Sadly most TTS development seems to be happening as online service these days. Google Neural TTS and Microsoft Azure TTS sound really great but require an online connection, an account, and possibly even paying (there's a threshold until it's free, then it costs almost nothing but almost nothing isn't free).

Btw, I don't know about the blind people you know but the ones I know use so insanle fast TTS output, the "sounds nice" aspect isn't really there in the first place. At least not to me.

[–] [email protected] 10 points 10 months ago (1 children)

The development of Piper is being driven by the Home Assistant Project. That probably makes it one of the larger OSS TTS projects. Hope may not be lost yet ;)

[–] [email protected] 1 points 10 months ago

Hope may not be lost yet ;)

And then we'll live in a TikTok TTS hellscape. 🤣

[–] [email protected] 1 points 10 months ago

Okay that is really interesting, so a TTS engine should be optimized to run very fast.

[–] [email protected] 8 points 10 months ago

Here is one of the best

https://github.com/neonbjb/tortoise-tts

[–] [email protected] 7 points 10 months ago

A few days ago I wrote down a couple of links to interesting TTS projects that I was going to look into whenever I have time, along with some brief notes.

https://github.com/coqui-ai/TTS TTS + XTTS, GPU inference? 3GB model.

https://github.com/rhasspy/piper Low resource, CPU inference. 50MB model.

https://github.com/p0p4k/vits2_pytorch GPU inference? 500MB model. https://github.com/p0p4k/vits2_pytorch/discussions/27 Someone's models for vits2

[–] [email protected] 5 points 10 months ago (1 children)

I've recently made aware of the bark project. Especially the bark-ui project. It takes a long time to run but it does work. Sometimes it makes cursed stuff too: https://social.rootaccess.org/@michaelc/111277260439738652

[–] [email protected] 2 points 10 months ago (1 children)

Wow that is crazy! 10seconds sounds like an unnecessary flex though, wouldnt like 30min/all sounds be best?

[–] [email protected] 2 points 10 months ago (1 children)

I have a tiny laptop with the literal bare minimum to get this running haha. Your probably right but the models explode your memory pretty quick.

I did get some really good audio out of this model after a while. I threw the first chapter of the hobbit at it and it seemed to be doing ok. It's better than espeak and you only need to do it once to get audiobooks out.

[–] [email protected] 2 points 10 months ago* (last edited 10 months ago) (1 children)

I have to check that! But wait is that Windows only?

[–] [email protected] 2 points 10 months ago

I got it working on Ubuntu/PopOS.

[–] [email protected] 4 points 10 months ago (3 children)

https://github.com/mozilla/DeepSpeech the best quality available that I've seen

[–] [email protected] 3 points 10 months ago (1 children)

That says it is speech to text, not text to speech

[–] [email protected] 2 points 10 months ago (1 children)

Omg is that the data from the Common voice project? Nice!

[–] [email protected] 1 points 10 months ago (1 children)

So it is, that's why some languages don't have good support yet - not enough recordings

[–] [email protected] 2 points 10 months ago

Also the last release is very old. Mozilla is weird, their pinned projects are often dead or outdated...

[–] [email protected] 1 points 10 months ago

Isn't speech to text the opposite of tts?

[–] [email protected] 2 points 10 months ago

Not sure if this is gonna be much better than the alternatives you've listed, but you can try adjusting pitch, rate, range, etc. with spd-say.

[–] [email protected] -2 points 10 months ago (2 children)

If you don't mind doing some development work, needing online connectivity, and paying for usage, AWS's Polly has some very good sounding TTS voices: https://aws.amazon.com/polly/

[–] [email protected] 3 points 10 months ago

Hm, as a backup that could be okay, but not working offline /without being a huge privacy problem...

[–] [email protected] 1 points 10 months ago

Just please don't use one of the kid voices for technical videos, a la babywogue and their GNOME development videos.

[–] [email protected] -5 points 10 months ago (2 children)

Samsung's TTS engine for android is the best I have found. I use it to listen to epub books.

[–] [email protected] 9 points 10 months ago

Not FOSS.

[–] [email protected] 1 points 10 months ago

Samsung sucks though... also softwarewise. Like, I debloated many phones and Samsung is crazy.

https://github.com/trytomakeyouprivate/Android-Tipps/tree/main/debloat

So their TTS will probly neither work offline, nor standalone