Glad to share with you this sample of the second iteration of OptiSpeech-TTS. In this iteration I fixed some issues related to punctuation pauses and pitch intonation. Needless to say that the model generates quality speech while being very fast and efficient. Thanks again to @pneumasolutions for their continued support.
awesome! except, is it only me, or the voice has a certain pronounced hissing issue? also, there's a lot of noise there, similar to what one would get with flite or festival voices. O yeah, it didn't pronounce the a in accessibility at the near beginning of the recording, which makes me wonder, how will pronunciation issues like these get fixed in this AI tts paradigm? Using more traditional methods, rules could be written to the phonemiser to compensate, like the ibm tts community dictionary for example, but with this, are you gonna have to retrain the entire model, or a significant part of it?
@esoteric_programmer Regarding the audible artifacts, this is half way in training. Most of them will be gone when training reaches step 1000000. Regarding pronunciation, phonemic rules can be written to enhance pronunciation same as with other TTS systems. Also, this particular model is very controllable. You can elongate phonemes, change the pitch and add stress at the phoneme level if you want.
@esoteric_programmer It does use that. But I'm working on a text preprocessing layer that can accept user dictionaries and apply changes at the phonemization and inference stages.
@esoteric_programmer Yes. This is a custom architecture. The baseline is JETS, but I augmented it with several components from the latest research in the field. Keep in mind that this has only 13M parameters, while Piper has 23M parameters. And OptiSpeech is at least an order of magnitude faster than Piper.
It looks like Bliss OS (Android for PC hardware) was used to help set a world record this week. The DIY Perks video shows some of the OS details. guinnessworldrecords.com/news/…
Wow, that's crazy: the world's largest iPhone, 2 meters high – got into Guiness (no, not the beer, the world record). And the craziest part? It runs a customized Android 🤪
Such a good video, although slightly disappointed by the android theming... I was excited for the details on how they managed to get a standard display signal out of a hilariously small iphone board
@daniel theming: well, it should look like an iPhone 🤷♂️ But yeah, it was all pretty impressive. When they plugged in that massive USB cable I got a laugh-flash 🙈
François Legault has had 6+ years to recruit, graduate, and place more teachers... instead, he practices the politics of fear, division, and hate w/ language and fabric. mastodon.world/@somecanuckchic… #assnat #polQC #QCpoli
Good news: #IzzyOnDroid is now also present at #OpenCollective 🥳 Thanks to the help of @electrikjesus we go with BlissLabs as our Fiscal Host. Like IoD, Bliss focuses on Android, and on extending the life of Android devices, helping futher community innovation and helping with the world's e-waste problems – which we found to be a good match!
Thanks Jon & BlissLabs accepting to be our Fiscal Host – and we look forward to a fruitful companionship!
I joined the club! I got a pair of Ray-Ban Meta smart glasses this weekend and am very impressed with them so far. I went with the Wayfarer Shiny Black style with clear lenses.
Do you happen to have any musical keyboard instrument? I'm curious to know if it's possible to use a smart glass to read a keyboard screen and program it at will, e.g. choose and edit tones and styles.
I just learnt that Kitboga and co made a piece of software called 'Seraph Secure' which can help stop scammers accessing your computer. It's free, but extra options do cost if you need those. seraphsecure.com/
A quick story/rant.. my cousin got my mother a new coffee maker. I don't expect my mom to use the WiFi functionality, but I'm still concerned about why we would need a whole coffee maker to have a small computer in it... I understand the versatility of not having a hub, and that hubs are also a single point of access for any malicious actors, but I'm sure a hub for multiple devices instead of a thing for a coffee maker would be updated much more frequently, and supported for longer. If nothing else, maybe I can pull it into my janky Home Assistant setup somehow.
Even if it doesn't end up being a gateway for snooping on home network, it is quite silly, in my opinion. Now, I must go check to see if anyone's hacked at these yet lol...
Macht doch bitte mal mit allen Euren Freund*innen, Eurer Familie, Menschen an Eurem Arbeitsplatz folgendes, ganz einfaches Experiment:
Fragt, ob die Person im Schnitt mehr als 1000 Euro am Tag an Zins- und Dividendengewinnen auf ihrem Konto gutgeschrieben bekommt, oder entsprechende Aktiengewinne einstreicht.
Wenn das nicht der Fall ist, habt Ihr eine weitere Person gefunden, die von einer Vermögenssteuer profitieren würde und keinesfalls selbst von der Steuer (als zahlende*r) betroffen wäre. Dann ist es nur noch ein kleiner Schritt, die Person davon zu überzeugen, die Europäische Initiative zur Einführung einer Vermögenssteuer mitzuzeichnen:
Da braucht es nämlich noch ein paar Stimmen ... ein paar viele sogar. Im Gegensatz zu den zahlreichen Klickibunti-Petitionen, die keinerlei Verbindlichkeit haben, ist das hier eine, die in einem Gesetzgebungsverfahren münden könnte, wenn genügend Menschen sich anschließen.
Mein Gast (ein erwachsener Mann) mag Katzen und Einhörner. Zu seinem Pech bin ich doof.
Er wirds überleben, und schlimmstenfalls hab ich noch langweilige andere.
PS: Those in my timeline who do know him *eyes two certain cats*: he has NOT seen it yet, so please no spoilers in the Signal group or so, just in case
@Bubu ich glaube meine Bubble ist da die falsche, die meisten fänden es witzig oder sogar toll, die meisten Leute da draussen würden schreiend wegrennen
Uuuuiii, finally: It's online! Machiav3lli & I were speakers at DroidCon Berlin earlier this year (in July), talking about "Open Source Community Building and Maintenance". Now the video is online:
Having been sick and tired of the Java-dependent and outdated RSSOwlnix and the buggy Luna RSS, I finally switched to Fluent Reader for Windows. I like its interface, and it doesn't suffer from the issues affecting Luna or RSSOwlnix. It offers RSS feeds and articles in an accessible web interface, and can even grab the full text of articles from sources such as Wired. JAWS reads check boxes in the app's Settings/ Sources tab better compared with NVDA, but it's not a showstopper. If you know a better or a more capable RSS reader for Windows, please do let us know. github.com/yang991178/fluent-r…
thanks for the recommendation, but what I'd actually like is an RSS to e-mail service. I know Blogtrottr, but it doesn't allow me to set the item author to be at the e-mail sender.
@archos snad to rozhýbu 😁 spíš se bojím toho že to je to pooperační...ale doktoři tvrdili že po 3 měsících by to mělo vydržet jakoukoliv zátěž co se kola týče.Tak jim budu věřit 😉
With Firefox having AI added in the recent update. Here's how you can disable it.
Open about:config in your browser.
Accept the Warning it gives.
Search browser.ml and blank all values and set false where necessary as shown in the screenshot, anything that requires a numerical string can be set as 0 .
Once you restart you should no longer see the Grey-ed out checkbox checked, and the AI chatbot disabled from ever functioning.
I did another of these #ElevenLabs speech-to-speech things over music, this time Super Mario Bros 'Running about' music, with my daughter Alice as the voice. What a crazy onomatopoeia moment.
Upon request from @lulu_bear, how about Overworld from Super Mario World as voiced by Stephen Fry? Warning, again... Put your drink down now. No seriously, put it right down. Your computer keyboard/mouse/screen will thank you. #ElevenLabs
I wish GitHub would just simply ban the use of “Act on your behalf” - there is NEVER a good reason to give a third party service FULL access to your account. This is always a very bad idea.
🎙️ Looking for new podcasts to discover? Check out index.castopod.org ! 🌟 Find a wide range of shows from creators all across Mastodon. 🦣Perfect for anyone wanting to follow, listen, and support independent voices. 📢🎧 #Podcast
Kolumne: Wladimir Putin führt längst Krieg gegen den Westen, mit Sabotage, Spionage, Bestechung und Propaganda - auch bei uns. Es wird Zeit, dass Politik, Behörden und Medien sich dieser Realität endlich vollumfänglich stellen – und entsprechend handeln.
1. Die Doppelgänger-Kampagne ist seit mindestens 8 Monaten bekannt - wieso hat die FBI-Aktion so lang gedauert? Und was machen deutsche Behörden eigentlich? 2. Warum sitzen in deutschen Talkshows weiterhin Russlandpropagandisten?
Are you sick and tired of websites, especially technical ones that have complicated commands or code snippets, preventing you from copy/paste? Here's a fix that website owners HATE!
Open Firefox (because you shouldn't be using anything else), and in the Awesomebar, type in "about:config" and press enter. Accept the warning. Search for the option "dom.event.clipboardevents.enabled" and set that sucker to FALSE.
Hi! I'm developing an app to help blind people like me to tag objects with NFC tags, especially, to organize our closet, describing clothes and making outfits to dress elegantly 😊. But there is a react-native component that has accessibility issues (accessibility properties are not propagated to the native component). could someone solve this Github issue? My ObjectiveC knowledge is null. thanks in advance! github.com/react-native-picker…
Nick Giannak III
in reply to Musharraf • • •Musharraf
in reply to Nick Giannak III • • •Yes the dataset has some alignment issues. I hope this will be fixed with more training.
Nick Giannak III
in reply to Musharraf • • •the esoteric programmer
in reply to Musharraf • • •Musharraf
in reply to the esoteric programmer • • •Regarding pronunciation, phonemic rules can be written to enhance pronunciation same as with other TTS systems. Also, this particular model is very controllable. You can elongate phonemes, change the pitch and add stress at the phoneme level if you want.
the esoteric programmer
in reply to Musharraf • • •Musharraf
in reply to the esoteric programmer • • •It does use that. But I'm working on a text preprocessing layer that can accept user dictionaries and apply changes at the phonemization and inference stages.
the esoteric programmer
in reply to Musharraf • • •Musharraf
in reply to the esoteric programmer • • •Keep in mind that this has only 13M parameters, while Piper has 23M parameters. And OptiSpeech is at least an order of magnitude faster than Piper.
the esoteric programmer
in reply to Musharraf • • •Musharraf
in reply to the esoteric programmer • • •pypi.org/project/ospeech/
Devin Prater :blind:
in reply to Musharraf • • •