Відео

@@Entertainmentretro4 Ever since my mom bought me Puffy AmiYumi CDs when I was a kid, I wanted to learn Japanese! Getting into utau in 2009 gave me the push I needed to really get started on learning :D ! I should be a lot better since I've been studying it for over half of my life, though 😝

@@MaeBlythe That's good!

@@Taliathewitch thank you :3 !!

@@Niko_da_funni_C_L wow! I just read through the documentation for VocalSharp! It looks super cool :) if I wasn't doing Diffsinger, I'd probably take a look at it! Yeah, a LOT of people have no idea how to OTO CVVC, so you can get led astray pretty easily. The overlap (green line) is the same as in VCV. You want to put it where the preceding vowel is still stable. The red line goes between the V and the C, since the C is the beginning of the note, the same way it goes between the C and the V in CV and VCV because the vowel is the start of the note in those cases. The pink part should cover a little bit, but it's important there's always white because utau will crash if there's no white. Whatever is in the white is what will be stretched out. Whatever is in the white will be crossfaded into the next sample. What you might have been confused about is plosives like "t" and "k". In those cases, it does look like the consonant isn't included, but in reality, that silence before the actual plosive is part of the consonant! If you include the actual "t" or "k" sound, there will be an icky scratchy glitchy sound :( And never ever ever include the vowel following the consonant within the white area while doing VC. CZ told everyone to do that for VCCV and it makes realistic tuning impossible without remaking the OTO!!

@@Taliathewitch thank you so much :D !!

@@Entertainmentretro4 thank you :3 it's definitely not as difficult as it looks :D ! I definitely would be happy to label up some English for you using this method. But if you did want to learn to do it on your own, the hardest part is just figuring out how to not go insane labeling 🤪 ! It's easier than doing OTO files to me, but OTO files take much less time and effort.

@@MaeBlythe do you at least have any sort of tutorial recommendations for it?

@@Entertainmentretro4 I do have one for setting up vlabeler! ua-cam.com/video/2phh5NpSgrI/v-deo.html basically, you add a line where a new Phoneme starts :) I use Praat, though. I found code that lets me import the labels from a text file! So I make a file that has all of the phonemes, and then I make a Textgrid. I put a line everywhere I think a phoneme starts and then I run that code and use phonemes go into the labels :D !

@@Entertainmentretro4 I also have this video for Diffsinger in general :3 ua-cam.com/video/9CJHqPLdSZg/v-deo.html

@@MaeBlythe Thanks, i'll soon see if I can make it

@@chenpoid Making a Coeroink will be difficult for anyone who isn't fluent in Japanese! There first thing to must do is record one of the recording lists, but you must do it by speaking naturally. If you don't know how to the the correct intonation and timing, it will end up wonky. However, if you can speak naturally, the rest is pretty easy. You don't need to configure anything because you're using a premade list. However, the only way there seems to be to train the voices are the Google Colab. You will need 7 GB of space free in your Google Drive. There are English translations of each step available to make it even easier! You will run out of compute units pretty quick. I don't know how long it will take, but any AI project using Colab's GPU will likely take weeks due to the fact that Google actually doesn't want you to use Colab for AI training. I usually only got an hour of training a Diffsinger model before I got cut off each day.

@@MaeBlythe oh, jeez😭 my sign to learn how to speak the language then. thank you, though!

@@chenpoid you're welcome :) ! I'm not too interested in Coeroink because it's just a text to speech. I only care about singing 😎

@@the-nocebo-effect the list should come with a base oto :D !

@@ameerazam3269 here :) github.com/MLo7Ghinsan/DiffSinger_colab_notebook_MLo7

@@Taliathewitch thank you 🥰!! Oooh I wonder how Kao's original AI data set would sound with voicing instead of breathiness :D !

@@MaeBlythe omg that sounds interesting we should add more data tho ahaha

@@the-nocebo-effect That would be so awesome :D !! I would love to see what you can do with the list! Use it however you like, and feel free to edit it as much as you like :) I would love to see what you make with it!

@@MaeBlythe Thank you!

@@Niko_da_funni_C_L Utan is a discontinued Praat script by Hataori-P. However, using scripts I have figured out how to use the same process of using it to create USTs :D

@@MaeBlytheshould i try to make a bank for it? It kinda sounds like goanimate text to speech! (If you know what that is)

@@Niko_da_funni_C_L naw 😅 it's really difficult to make banks for unless you already know how to label for Diffsinger.

@@MaeBlytheyeah idk anything about diffsinger 😭

@@Niko_da_funni_C_L Diffsinger is a lot of fun, but it takes a lot of work!

@@the-nocebo-effect just Praat and a script to export to UST :D ! I also use the utau plugin utalis for pitch. I made a video and a blog post about how to do it :3 ua-cam.com/video/Auax3wJovs0/v-deo.html

@@felixetcie thank you :D !!

@@nobody-qv9cn thank you so much :D !

@@MaeBlythe Of course! You're very welcome! ^_^

Thank you so much :D ! I added the UST to the description :) it's CVVC :D

​@@MaeBlythe you're welcome and thank you so much!!!

Thank you :D !! I can't take all the credit, though. It's voice copy :D

Thank you so much :D !

You're welcome :D !

Yeah :D ! If you do decide to make an utau bank, I hope you have fun :)

Thank you so much :) ! She is my utau. Just like me, she is a girl :3

@@MaeBlythe Yw it does sound like you a bit I was unfamiliar with it because I know you use a lot of lost utaus

@@nathankid76 yeah :D !

A lot of people start out doing just that :) ! A lot of USTs are very beautifully tuned, so it isn't necessary to learn to tune on you own until you want to make a UST on your own.

@@MaeBlythe Yeah and I have MMD too I made two models posted them on deviantart I’m proud of that

@@kuraikanjoutau that's awesome :D ! Making MMD models is hard, so you should be proud :3

@@MaeBlythe I got the hang of it I saw tutorials all I did was make textures and downloaded parts for it

@@kuraikanjoutau yeah :) ! Making textures is fun :D

Yeah :D ! I'm trying out doing it on my new phone, so it's a lot easier than on my old phone :3

@@MaeBlythe ohoho niceee!!

As far as I know, there isn't an instrumental version. I used AI to remove the vocals :)

Making a Diffsinger bank is less complex than it sounds, but it is very time consuming if you want to do it well. The biggest component is learning to label data. You need to tell the computer where every phoneme exists. It could be tempting to use a tool like Whisper to automate this, but it takes a LOT of time to use AI to train the model, so it's better to spend the time to get it right first! You do not need any specific hardware to train models. You can use a website called Google Collab :)

I'm sorry, I don't use that. My only social media is Twitter (annamaeblythe)

When you select an audio object in the object viewer, the pitch button will show up in the object viewer :)

I made a new video on my channel that explains it better :D !

@@MaeBlythe Thank you!! I'll go watch it when I'm on PC 😊

I can't use synth V, but maybe you could use synth V to make a voice sing the English lyrics I make up and then you can use RVC and then Kao could sing the song :D

@@MaeBlythe yeah def i would need a midi but def down for it

@@Taliathewitch yay :D ! I would yeet over the midi with the lyrics input :3 thank you :D

@@MaeBlythe so hyped!!!

Thank you :D ! I would have never had the confidence to record this high without your encouragement :3 !

I never finished the UST, I am sorry 😔

😎

Yeah :D ! It's funny how different culture was back in the day. Today, people would make fun of anyone who liked Iku because of her history. Back then, no one got made fun of. We just thought she was great!

Thank you so much :D !!

:D !

It's a very time consuming process! The first thing you need to do is accurately label every single phoneme in at least 10 minutes of audio. I use Praat, but most people use VLabeler.

@@MaeBlythe do you plan on making a tutorial?

@@berryweary I have a video showing how to label Japanese if I remember correctly :) ! But all of the up to date information is in the PixPrucer Google docs.