194
submitted 1 week ago* (last edited 1 week ago) by boredsquirrel@slrpnk.net to c/opensource@lemmy.ml

I just found this.

Main page

This is huge!

As a german, I use thorsten medium as he simply made the best dataset.

Mixing english with german, speaking numbers, single letters, pausing without a "." but just a linebreak, all those can be essential.

And... it is nearly perfect! And all local!

This is crazy!

eSpeak can finally go to rest!

top 31 comments
sorted by: hot top controversial new old
[-] macattack@lemmy.world 8 points 6 days ago* (last edited 6 days ago)

To those late to the party, you can sample the piper-related voices here so that you're not in a crapshoot: https://rhasspy.github.io/piper-samples/

[-] boredsquirrel@slrpnk.net 2 points 6 days ago

Nice, thanks!

[-] xlash123@sh.itjust.works 4 points 6 days ago

Lmao, there's a GLaDOS voice. I'm taking that

[-] boredsquirrel@slrpnk.net 1 points 6 days ago

Whats that? Mind to share an audio file?

[-] Teppichbrand@feddit.org 3 points 6 days ago* (last edited 6 days ago)

eSpeak is a damn good conversation starter. If I fire up my OsmAnd and some german robot with a dutch accent (or is it Kölsch?!) starts giving me directions noone ever not commented on it. Downloading Thorsten Medium right now.

[-] boredsquirrel@slrpnk.net 1 points 6 days ago

Hahaha yes it is.

[-] macattack@lemmy.world 32 points 1 week ago

Might want to link to the official page for a bit more context

[-] iturnedintoanewt@lemm.ee 5 points 1 week ago* (last edited 1 week ago)

For context really, because figuring out the android install there ain't a straight thing.

I've been using these TTS for a few months now, and I'm seriously impressed with its quality, considering it's running all in local as OP said. I always find funny when it tells me to TURN LAFT ON THE NEXT ROUND... ABOUT, but that's a really minor nit picking, overall audio quality is amazing.

[-] macattack@lemmy.world 1 points 6 days ago* (last edited 6 days ago)

Just to clarify, this is standalone and doesn't need a local LLM or anything yeah? Also, any tips on tutorials for the install process?

Edit: Nevermind, I realize it's install and use

[-] iturnedintoanewt@lemm.ee 2 points 6 days ago

Yup, standalone and doesn't require anything else, no connectivity whatsoever. It runs local.

[-] BarHocker@discuss.tchncs.de 8 points 1 week ago

I there a way to install multiple languages? I, for example, would like to install both English and German so that I can switch between them depending on the language of the text.

[-] Sir_Kevin@lemmy.dbzer0.com 8 points 1 week ago

Can I install it via F-Droid?

[-] boredsquirrel@slrpnk.net 5 points 1 week ago

It is FOSS, and the build instructions are here

https://k2-fsa.github.io/sherpa/onnx/android/build-sherpa-onnx.html

Those guys are incredibly active, look at this!

https://k2-fsa.github.io/sherpa/onnx/index.html

(And no, not on F-Droid, if you look at the crazy amount of packages that makes sense)

[-] Swedneck@discuss.tchncs.de 6 points 1 week ago

that stupid huge amount of packages is honestly a pretty big UX issue, at least personally that just makes me chuckle at their optimism and close the page as every bit of energy i had vaporizes from the thought of figuring out what to download.

[-] boredsquirrel@slrpnk.net 0 points 1 week ago

Just scroll down until the page is fully loaded and use the search function.

[-] PlexSheep@infosec.pub 1 points 4 days ago

Yeah but search for what even?

[-] boredsquirrel@slrpnk.net 1 points 4 days ago

your language abbreviation like "de" or "en"

[-] Lemongrab@lemmy.one 7 points 1 week ago
[-] Rogers@lemmy.ml 6 points 1 week ago* (last edited 1 week ago)

This is amazing! If you are looking for US EN and use a phone with arm64 I can recommend sherpa-onnx-1.10.27-arm64-v8a-en-tts-vits-piper-en_US-kristin-medium and sherpa-onnx-1.10.27-arm64-v8a-en-tts-vits-piper-en_US-norman-medium.

Edit: I don't seem to be able to get the engine to show up as a tts engine but it works well within the tts app itself. Hopefully I'll find a fix I've been searching for a good tts engine for android eBook apps.

Edit 2. Fixed I mistakenly downloaded the standalone version from https://k2-fsa.github.io/sherpa/onnx/tts/apk.html instead of https://k2-fsa.github.io/sherpa/onnx/tts/apk-engine.html

[-] boredsquirrel@slrpnk.net 1 points 6 days ago

Yes they have 3 types of client apps

[-] boredsquirrel@slrpnk.net 2 points 1 week ago

These names are a joke hahaha

[-] Rogers@lemmy.ml 7 points 1 week ago

Yeah IMO they need an app that can download/manage models directly

[-] boredsquirrel@slrpnk.net 1 points 6 days ago

That would then require running code from memory, which needs to be enabled in recent GrapheneOS. But all fine, they should absolutely do that as they obviously do that currently but during compilation.

Client app + wizard to select language and model type + downloaded model

[-] semperverus@lemmy.world 2 points 1 week ago* (last edited 1 week ago)

You are clearly not the target audience then. There is a lot of extremely useful info in those model names.

[-] boredsquirrel@slrpnk.net 1 points 6 days ago

Yes of course :) I use sherpa-onnx-1.10.26-arm64-v8a-de-tts-engine-vits-piper-de_DE-thorsten-medium and there is nothing irrelevant in that name.

Still after another commenter wrote how overwhelmed they are with that INSANE download page, I felt that when you dropped those 2 names

[-] justtobbi@feddit.org 3 points 1 week ago

How do I apply these voices, so apps use it?

[-] justtobbi@feddit.org 2 points 6 days ago

Go into Settings -> Accessability -> Speech (at the very bottom; at least on Android 13) and then TTS :) hope that helps

[-] boredsquirrel@slrpnk.net 2 points 6 days ago

They have 3 types of clients, only one works as an Android TTS engine, the others are useless.

[-] justtobbi@feddit.org 2 points 6 days ago

I downloaded the thorsten medium and wondered, what the difference to thorsten high would be?

[-] Teppichbrand@feddit.org 2 points 6 days ago

You can listen to them here

[-] boredsquirrel@slrpnk.net 1 points 6 days ago

Thorsten high is silly haha. Emotional is also not meant for TTS more for research I think.

I think thorsten made the only good model in German, I really want to make my own one! Or get some famous people on board?

this post was submitted on 08 Oct 2024
194 points (98.5% liked)

Open Source

30511 readers
680 users here now

All about open source! Feel free to ask questions, and share news, and interesting stuff!

Useful Links

Rules

Related Communities

Community icon from opensource.org, but we are not affiliated with them.

founded 5 years ago
MODERATORS