RVC WebUI Voice Cloning Tips, Tricks and Experiments!

แชร์
ฝัง
  • เผยแพร่เมื่อ 16 ธ.ค. 2024

ความคิดเห็น • 176

  • @MiguelGomez-qx7qc
    @MiguelGomez-qx7qc ปีที่แล้ว +31

    Ive been using RVC for the past couple of days with better than expected results. With these tips I feel like im gonna take my projects to the next level. Thank you for all the hard work!

  • @kusog3
    @kusog3 ปีที่แล้ว +18

    RVC is damn good. Not only can it clone voice to another, it can do so in another Langauge.
    Heck, I've been using it to change the voices of some anime I watched to my favorite actors.

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว +3

      Fun, huh? 🍿

    • @dthSinthoras
      @dthSinthoras ปีที่แล้ว +3

      You are able to do other languages? I try it with german and it gets my voice pretty good, but the pronounciation of words is kinda terrible. Do we need more training data for other languages? Or do you have any additional tips for me?

    • @LegendaryLibrary-l2u
      @LegendaryLibrary-l2u ปีที่แล้ว +2

      @@dthSinthoras you need a voice in English for music in English, or a voice in German for music in German. if you take a German voice for English music, it makes an accent. I tested with several languages, you have to keep the same language for the voice and the music

    • @dissonanceparadiddle
      @dissonanceparadiddle ปีที่แล้ว +1

      woah that's so cool!!

    • @kusog3
      @kusog3 ปีที่แล้ว

      @@LegendaryLibrary-l2u I think the accent is pretty cool

  • @aa-xn5hc
    @aa-xn5hc ปีที่แล้ว +3

    You are the coolest geek ever....
    The final duet is gold

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว +1

      Thanks! Glad you liked the things 😀

  • @kait3n10
    @kait3n10 ปีที่แล้ว +7

    Great practical advice. I've personally used the free Adobe AI voice enhancer to make the extracted voices clearer, but UVR seems very promising. Will have to try that out. Thanks!

    • @__--JY-Moe--__
      @__--JY-Moe--__ ปีที่แล้ว

      Sandesu.-Hen!

    • @Charlton222
      @Charlton222 ปีที่แล้ว

      I use UVR all the time and its great.

  • @dennisliebelt3951
    @dennisliebelt3951 ปีที่แล้ว +1

    I always thought crepe will give the best results. Thanks for the research and the useful tips how to get the best results!

  • @RolanRoyce
    @RolanRoyce ปีที่แล้ว +1

    Why when I download voice models from the weights site and do an inference on a vocal track from a song it sounds like crap? It'll sound normal until a certain part comes up and then it'll sound like somebody getting strangled.

  • @blackvx
    @blackvx ปีที่แล้ว +2

    That's nerdy to the core, thanks!

  • @agiverreviga4592
    @agiverreviga4592 ปีที่แล้ว +2

    The outputs seem to be the best but the annoying quirks are that each time you want to train or retrain a model, you have to input the settings all over again. And there's no quick or convenient method to share and publish metods, so that others can use them

  • @Dante02d12
    @Dante02d12 ปีที่แล้ว +2

    This is all very interesting, but I'd like a direct answer to these questions if you don't mind :
    - does crepe or mangio-crepe make a huge difference in voice quality compared to harvest? I mean, is it really night and day, like comparing pm to harvest? I have a version of RVC where I don't have crepe, and I'd like to know if it's worth upgrading ^^.
    - I've been using UVR5 to isolate voices, and it is indeed great... except for the quality loss. What's the current best workflow with UVR5 to not lose voice quality? Basically, which models and processing methods work best to isolate a voice from most songs?
    - My voices tend to not articulate as well. I assume I'm not doing something right when I create the model. I'm using 7 minutes of voice samples extracted from a videogame, so it's professional audio quality. The voice is very recognizable, but I do feel it sometimes doesn't articulate. Is it current limitation of voice cloning tech?
    Thank you for your hindsight!

    • @Jojo2
      @Jojo2 ปีที่แล้ว +1

      I'm also struggling with the last one. I personally assume it's due to such a low amount of training data? But I haven't tested it. My other theory is I'm probably not extracting the vocals of songs correctly.

  • @marcfruchtman9473
    @marcfruchtman9473 ปีที่แล้ว

    Super interesting. I really like the invert audio trick!

  • @jurandfantom
    @jurandfantom ปีที่แล้ว +3

    thank you for your hard work NR

    • @jurandfantom
      @jurandfantom ปีที่แล้ว +1

      on that note, what would be better at this stage? text2Voice with custom voice trained or text2voice with build in BARK voice and then voice2voice ?

  • @Markgen2024
    @Markgen2024 ปีที่แล้ว

    Thank you for your hard work making this video! Looking forward for a sample cover song with your model.

  • @ratside9485
    @ratside9485 ปีที่แล้ว +1

    I think that's all too nerdy for me. I'm pretty happy with the normal version. But the sometimes does not find the index file, that sucks. Right now it is much too warm in my apartment. I'm taking a break from training new models. Thanks for the tips

  • @nathanbanks2354
    @nathanbanks2354 ปีที่แล้ว +1

    I haven't listened to this song since I played Steno Arcade! (Which is free to play either for download or on Steam.) Presumably you both used it because of the license; I can't remember if it was commissioned or not.

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว +1

      Creative Commons licenses are cool - especially the ones that allow derivatives 😉

  • @RHYTE
    @RHYTE ปีที่แล้ว

    Hell yeah waited for this!

  • @voEovove
    @voEovove ปีที่แล้ว +1

    That's insane! Consider my mind blown.

  • @ris_kis
    @ris_kis ปีที่แล้ว

    Good advice about sharing, because I somehow almost managed to understand that, but you confirm my theory.

  • @fadyibrahim8649
    @fadyibrahim8649 ปีที่แล้ว +1

    The sound is incredibly amazing, but unfortunately, it's not working for me, even though the Retrieval-based-Voice-Conversion-WebUI is working perfectly.
    I keep getting an error indicating that there is an issue with Torch, even though I am certain that it is installed and present.
    I’m using windows 10
    Can you help me ?

  • @denblindedjaligator5300
    @denblindedjaligator5300 ปีที่แล้ว

    if I choose that a module should have no tone and I train it in the new version of RVC, I can still choose which tone algorithm to use. This means that it still uses RMVPE, i.e. the new version and the quality is not particularly good either. Hope it gets fixed. try to choose false in the old and in the new version.

  • @dthSinthoras
    @dthSinthoras ปีที่แล้ว +1

    I played around a lot with the Tensorboard-Graphs now. Why are you setting the smoothing this high? Because you want to see what seems to be the overall better curve? For finding the best epoch it would be better to set it to 0 right?

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว +2

      It shows the trend, which can be difficult to tell without smoothing

    • @dthSinthoras
      @dthSinthoras ปีที่แล้ว +1

      @@NerdyRodent Do you have something (a Discord, GoogleDrive, whatever) to discuss curves? I would love to deepdive into that theme :)

  • @Seany06
    @Seany06 ปีที่แล้ว +2

    Thanks for this sir, btw is the voice you use a voice model? lol
    Does anyone know of any communities where people are sharing their vocal models for RVC?

  • @denblindedjaligator5300
    @denblindedjaligator5300 ปีที่แล้ว

    using translater: When I train a voice in the new version of RVC, it looks as if it has chosen an algorithm even though I have chosen not to use pitch during the training. try training a module in version RVC0813Nvidia and then version RVC1006Nvidia. I mean when you train you can choose whether you want to use pitch or not. the sound is more natural in the old version when you don't use pitch. there is a synthesizer input module, bug how can we use this?

  • @ismail1976
    @ismail1976 5 หลายเดือนก่อน

    How can i combine Harvest +Diffgrad and Adam , wenn i just use the RVC Interface from mangio or Applio ? So in the Json file in "Learning Rate i have these entry : "learning_rate": 1e-4 . Why is it different to yours ? How should i change it ?

  • @ismail1976
    @ismail1976 5 หลายเดือนก่อน

    Could you please react of the new Converting Mode : rmvpe ? Harvest is better ? Or the New One ? And Please tell how to set up the Optimizers

  • @TerminatorSAW2k
    @TerminatorSAW2k ปีที่แล้ว +1

    I need a ObS Studio Live Tutorial with this voice cloning tool, please :) .

  • @denblindedjaligator5300
    @denblindedjaligator5300 10 หลายเดือนก่อน

    could we make a shared dropbox? i will like to know if you trained one of my dataset on the false settings on a batchsize of 40 will the autotune effect go away? my dataset is 1 Hour long? and train it for 1000 epochs thanks. how mutch can i clock my GPU in the EVGA Precision X1 my graphic card is 400

  • @denblindedjaligator5300
    @denblindedjaligator5300 ปีที่แล้ว

    I'm just trying to explain something more clearly via google translator: where you have to choose during the training whether you want Whether the model has pitch guidance (required for singing, optional for speech):

    threaten

    false. When you choose false and you have to do a Model Inference, it has chosen one of those rmvpe. that is, if I load Ai_Hoshino_TTS.pth it doesn't ignore RMVPE, but the old version does.

  • @denblindedjaligator5300
    @denblindedjaligator5300 10 หลายเดือนก่อน

    can i send you my dataset it is 1 Hour 7 min. and 6 sec. the index file can ony get up to 600.2 mb. could you train it on 1000 epochs, 48 khz and pitch set to false? I would like to hear the difference between a batch size p¨26 and a batch size p¨40 we could make a dropbox folder.

  • @macronomicus
    @macronomicus ปีที่แล้ว +1

    This app is so much fun!

  • @timeTegus
    @timeTegus ปีที่แล้ว

    when i use Ultimate Vocal Remover the vr models are very bad every voice sounds like a mikimosue and the seperation is very bad. maby i installed someting wrong idk. but the mdx models sound good. but they seem to runn on my cpu insted of my gpu

  • @toph_beifong
    @toph_beifong ปีที่แล้ว +1

    Ay this is awesome! After getting the hang of so-vits, this is looking real promising.
    Just wondering though, is there a easy way to use a model for TTS or would I need to create a separate model specifically for that?

  • @calvin54
    @calvin54 ปีที่แล้ว

    And changing the learning time, what am i suppose to change that to? can this be done if a model is already training or would i need to retrain again with these changes made

  • @H4AZ
    @H4AZ ปีที่แล้ว

    hey, how can i use the harvest-diffgrad model to train? I can't find them in google colab

  • @ismail1976
    @ismail1976 5 หลายเดือนก่อน

    Dont understand why you done the audio in 40k. Please explain . Does it change somethink in the long run ?

  • @JohnSundayBigChin
    @JohnSundayBigChin ปีที่แล้ว

    Hi Nerdy, thanks for the tips!. Is it necessary to update the RVC that I installed a couple of weeks ago?
    I ask because I only have pm, harvest and crepe in my RVC.

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว +1

      Personally I git pull every time, but whatever version I’d best got you!

  • @denblindedjaligator5300
    @denblindedjaligator5300 7 หลายเดือนก่อน

    if i make a module that doesn't respect pitch how can i get it to change pitch when i transbone it i have a daft punk vocoder module and when i transpose it will work the module even if it has no pitch

    • @NerdyRodent
      @NerdyRodent  7 หลายเดือนก่อน

      Try including more pitch variety in your training data

    • @denblindedjaligator5300
      @denblindedjaligator5300 7 หลายเดือนก่อน

      @@NerdyRodent can i send you the daftpunk vocoder modul

  • @MistahJ100
    @MistahJ100 ปีที่แล้ว

    I really need some help, my software decided to stop working, i made a few cover songs and they turned out cool, but now for some reason when i click convert on model inference it just stops and will not process. i had another app that converted the sounds with my models RVC-GUI-pkg
    And it just does not work anymore and gives me a blank mp3 file.... Why would this happen? is there some time limit for these apps??? i see nothing on the internet about this i changed nothing with my directories as to why this would occur. PLEASE help i want to make some more covers i dont know why its not letting me

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว

      You may need to git pull again to get the latest version (assuming you did a normal install, not a zip file)

  • @denblindedjaligator5300
    @denblindedjaligator5300 6 หลายเดือนก่อน

    how can people train without v1 or v2'i have a model where it says none

  • @choppergirl
    @choppergirl ปีที่แล้ว

    I have 128gb system ram, but only an 8gb GPU I bought yesterday on Ebay.
    Ist his going to work for me with only a 3070 8gb card?

  • @limpopo171
    @limpopo171 ปีที่แล้ว

    That is impressive quality I have to say. I wonder does it work for text to speech? or is it only for singing?

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว +1

      You can do tts then convert that to a new voice. If it’s not singing, probably best to keep singing out of your dataset

  • @workflowinmind
    @workflowinmind ปีที่แล้ว +3

    Impressive!

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว +2

      Glad you liked the things!

  • @dthSinthoras
    @dthSinthoras ปีที่แล้ว

    Hi there, I am playing around with Tortoise for text to speach and RVC for speach to speach, but with both I have a pretty similar problem: They sound like me, but they can pronounce german sentences.
    Are they just not trained for that? Are there other base-models for other languages? Or is my training data of ~45 minutes of speaking still not enough?

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว +2

      you may need to ensure that you have a balanced range of phonemes in your dataset

    • @dthSinthoras
      @dthSinthoras ปีที่แล้ว

      @@NerdyRodent So.. it should be possible yes? :)

    • @YumeLabs
      @YumeLabs ปีที่แล้ว +2

      Quality, not quantity I say. I have an english voice speaking vietnamese, indonesian, japanese, french, russian, cantonese, mandarin, and dutch, so of course it can handle many known languages. It can even pronounce Middle English! I just wish there was a middle english text to speech so we could hear Chaucer style parodies of current events... But german should be very easy for it. Try singing a little maybe, or varying the enunciation of common words.

  • @calvin54
    @calvin54 ปีที่แล้ว

    what does changing the log interval from 200 to 100 do or benefit? also if im using v2 models then i would need to change the 48k_v2 json config file ?

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว

      It just logs more often. Who doesn’t love extra datapoints? 😉

  • @toasteroven6761
    @toasteroven6761 ปีที่แล้ว +1

    What does the X-axis on the graphs represent?

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว +1

      Steps

    • @toasteroven6761
      @toasteroven6761 ปีที่แล้ว +1

      @@NerdyRodent Thanks, one more question, does the file format used for the training (wav, mp3, flac) affect the resulting model's output accuracy and quality in a noticeable way or not? If so, what export setting do you recommend in audacity?

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว

      @@toasteroven6761 go for wav

    • @toasteroven6761
      @toasteroven6761 ปีที่แล้ว

      ​@@NerdyRodent Welp, looks like I'm going to have to wait over 4 hours just to upload my 50 minute wav file to colab----rural internet...
      So mp3 must really tank the quality then...

  • @QHawk7
    @QHawk7 ปีที่แล้ว

    *I need to translate srt file (subtitle) from English To Arabic , please , any gooood solution? with the best quality human like translation , not google translate or anything like that?*

  • @denblindedjaligator5300
    @denblindedjaligator5300 7 หลายเดือนก่อน

    and what batch size should i train on?

  • @frattuncbas
    @frattuncbas ปีที่แล้ว

    How can i continue my pre-trained model later? I want to train it 20000 epochs for most realistic quality but i need to run day by day my same pre trained model..Is it possible on Easygui Colab? I have Pro colab.

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว

      Sure, you can do that on colab though 20,000 epochs is a bit high…

    • @frattuncbas
      @frattuncbas ปีที่แล้ว

      @@NerdyRodent How can i?

  • @H4zuZazu
    @H4zuZazu ปีที่แล้ว

    Is there a TTS tool that can use the self trained models?

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว

      As this is voice to voice, and tts generates a voice, pretty much any tts generated voice will do!

  • @ris_kis
    @ris_kis ปีที่แล้ว

    Hello and thanks for your video. I suffer a little with getting voice for training and training time (don't have it)

  • @shabadouwilou
    @shabadouwilou ปีที่แล้ว

    is there an ai to change the lyrics of a song ? like in goes a singing voice and out comes same voice with the lyrics changed

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว +1

      Probably best done by humans at the moment!

  • @denblindedjaligator5300
    @denblindedjaligator5300 ปีที่แล้ว

    what equipment do you have? I mean how much have you paid for your computer, can you use a 12 GB graphics card, or do you have to have a 16 GB to be on the safe side.

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว

      The more VRAM the better, though it all depends what one does with one’s computer

    • @denblindedjaligator5300
      @denblindedjaligator5300 ปีที่แล้ว

      So i can train the most on 16gb?

  • @LegendaryLibrary-l2u
    @LegendaryLibrary-l2u ปีที่แล้ว

    Is it possible to use the files without having the index? Because many people have shared without this file and it seems to work?

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว +1

      Yup, it just sounds a lot better with the index file too in my experience

    • @LegendaryLibrary-l2u
      @LegendaryLibrary-l2u ปีที่แล้ว

      @@NerdyRodent Ok thanks. I tried several times to copy voices, but without the index it never worked. I also tried to use RVC-beta which launches quickly and has a simplified interface. But you have to use zip files, with inside you need the "index" file. Is there a way to create this "index" file?

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว

      @@LegendaryLibrary-l2u not without the original dataset

  • @exidion54
    @exidion54 ปีที่แล้ว

    Sorry if this is a stupid question, but how can I resume the training at a checkpoint, if possible?

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว

      Yes, you can just resume

    • @exidion54
      @exidion54 ปีที่แล้ว

      @@NerdyRodentShould I just change the pretrained models in step 3 with the checkpoint model to do that?

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว

      @@exidion54 simply increase the number of epochs and press train again

  • @mahmood392
    @mahmood392 ปีที่แล้ว

    Was wondering the batch size u used for this?

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว

      As much as your GPU can handle 😉

    • @mahmood392
      @mahmood392 ปีที่แล้ว

      @@NerdyRodent I have 24gb 3090, but idk why it’s training soo slow.. I set it to a batch of 40, I left for many many hours it just hit 40epochs.. very weird on epoch 34 it took 1 hour and 44 min to do 1 epoch.. i wasn't home to notice when that happened.. now I'm on epoch epoch 44 took 2 hours and 45min to finish.. and another on took 58 min to finish.. i am very confused

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว

      @@mahmood392 it could be that you have a two or three hour long dataset. The bigger your dataset, the longer it will take. Usually 30 minutes is absolutely fine.

    • @mahmood392
      @mahmood392 ปีที่แล้ว

      it was 10min dataset.. something was broken because some epochs took 3 hours and then the epoch before it or after it was just 2min.... anyways i ran it again with a batch of 28.. it finished in 16min, 5seconds per epoch@@NerdyRodent

  • @denblindedjaligator5300
    @denblindedjaligator5300 ปีที่แล้ว

    Can I easily train on the 40 80 or do I have to have a 40 90

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว

      Any modern Nvidia GPU will do!

    • @denblindedjaligator5300
      @denblindedjaligator5300 ปีที่แล้ว

      @@NerdyRodent i mean the performance on¨the 2 altso 4080 ore 4090

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว

      4090 is better

  • @ForgottenT
    @ForgottenT ปีที่แล้ว

    Does anyone know if it's possible to use trained voices with text to speech somehow?

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว

      Yup! TTS-> voice -> voice

  • @Heraplayswith
    @Heraplayswith ปีที่แล้ว

    Can i do Text to speech with this? How can I do that? I really need TTS :)

  • @ericanderson5139
    @ericanderson5139 ปีที่แล้ว

    Can it change voice on live for videos making or calls

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว

      Yup, it’s just about fast enough for real time

    • @ericanderson5139
      @ericanderson5139 ปีที่แล้ว

      @@NerdyRodent How do you get the real-time up I can't see it please

  • @douglasteixeiradeabreu
    @douglasteixeiradeabreu ปีที่แล้ว +2

    Eu não entendi muito bem o que isto faz, mas senti exatamente a dimensão da experiência. Você fez uma grande demostração, estou acioso para acompanhar o desenrolar disso, mais demostrações serão bem-vindas 😮!

  • @hosniabouzahra
    @hosniabouzahra ปีที่แล้ว

    Hi Nerdy, thanks for the tips, would you please share a zip file of (Mangio-RVC-Fork) same as the (RVC-beta 7z) so i run it directly in my windows. Thank you

  • @camspider9887
    @camspider9887 ปีที่แล้ว

    can u tell us where to change that seed?

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว

      It’s in the config file for your training, as shown

  • @hungluu8404
    @hungluu8404 ปีที่แล้ว

    how to change the optimizer?

  • @denblindedjaligator5300
    @denblindedjaligator5300 10 หลายเดือนก่อน

    what batchsize do you train on

    • @NerdyRodent
      @NerdyRodent  10 หลายเดือนก่อน

      40

    • @denblindedjaligator5300
      @denblindedjaligator5300 10 หลายเดือนก่อน

      @@NerdyRodent so if I can only train with 24 then I'm not getting a good quality

    • @NerdyRodent
      @NerdyRodent  10 หลายเดือนก่อน

      @@denblindedjaligator5300 24 is fine 👍🏽

  • @dthSinthoras
    @dthSinthoras ปีที่แล้ว

    I try to reproduce your optimizations, but I get this Error: AttributeError: module 'torch.optim' has no attribute 'DiffGrad'
    How to fix that?

    • @dthSinthoras
      @dthSinthoras ปีที่แล้ว

      AdamW and RAdam are working, I just dont get DiffGrad to run as optimizer.

    • @dthSinthoras
      @dthSinthoras ปีที่แล้ว

      The Output with RAdam is just silence for me :O

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว +1

      DiffGrad isn’t a default optimiser, it’s part of pytorch optimisers

    • @dthSinthoras
      @dthSinthoras ปีที่แล้ว

      @@NerdyRodent I have questions regarding this, but they always disappear, so this comment just wants to see, if youtube doesnt like my questions, or if everything disappears...

    • @dthSinthoras
      @dthSinthoras ปีที่แล้ว

      @@NerdyRodent Ok, my testanswer seems to stay, so TH-cam probably dont like code snippets here? I tried to show what I tried, because I still get the same error. I will try to show what I tried without using something that looks like code...
      I did a pip install torch_optimizer and I replaced AdamW with DiffGrad.
      What else is needed?
      Ohh and thank you for the other answer, that seemed to be at least a good hint!

  • @b1ll1on_ai
    @b1ll1on_ai ปีที่แล้ว

    hello Nerdy MASTER! why Index first, Training second ? thnks

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว

      Index trains fast 😉

  • @elarcadenoah9000
    @elarcadenoah9000 ปีที่แล้ว

    teach us how to use bark and install

  • @lazerusmfh
    @lazerusmfh ปีที่แล้ว +1

    Freaking awesome

  • @GATUK1773R
    @GATUK1773R ปีที่แล้ว

    DAMM, WHY NO MAKE A WINDOWS NODE CODE A SCREEN A BUTTONS ,STAR,COPY,CHANGE,OK,CONVERT?

  • @denblindedjaligator5300
    @denblindedjaligator5300 ปีที่แล้ว

    What nvidia card is supported?

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว

      You’ll get the best performance with something like a 4090. Nice and fast with plenty of VRAM!

    • @denblindedjaligator5300
      @denblindedjaligator5300 ปีที่แล้ว

      @@NerdyRodent what about 4080 it is god? and 16 gb of ram

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว

      @@denblindedjaligator5300 16gb VRAM can do lots of things for sure

    • @denblindedjaligator5300
      @denblindedjaligator5300 ปีที่แล้ว

      @@NerdyRodent what is the lirning rate and the seed? Could you make a video about pipper tts

  • @__--JY-Moe--__
    @__--JY-Moe--__ ปีที่แล้ว

    Ittai dare ga watashi no doa o nokku shite iru nodeshou ka? Sā, mō koko ni wa konaide kudasai yoru osoi no ga miemasen ka? Totemo tsukarete ite kibun ga yoku arimasen watashi ga nozomu no wa hitori ni naru koto dake chikadzukanaide, watashinoie ni shin'nyū shinaide kudasai soto de burabura shite itara saikōdesu haitte konaide, nigete kakureru dakedesu imanara dare ni narerudeshou ka? Imanara dare ni narerudeshou ka? Imanara dare ni narerudeshou ka? Imanara dare ni narerudeshou ka?

  • @009badboy
    @009badboy ปีที่แล้ว

    Hi Nerdy Rodent.. do you have a discord comunity?

  • @flonixcorn
    @flonixcorn ปีที่แล้ว

    Great Videoi

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว

      Thanks! Glad you liked it

    • @flonixcorn
      @flonixcorn ปีที่แล้ว

      @@NerdyRodent im getting a pretty robotic voice, with a lot of random high pitch "artifacts" and quick tips you can give me?

  • @banzai316
    @banzai316 ปีที่แล้ว +1

    Tips, Tricks and Bookmarks

  • @RedDragonGecko
    @RedDragonGecko ปีที่แล้ว

    "Download the weights" Weights? What weights? Where?

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว +1

      From the link to the weights on the github page ;)

  • @PurpleRhymesWithOrange
    @PurpleRhymesWithOrange ปีที่แล้ว +1

    Pretty impressive that it can make a rat talk with a British accent!

  • @ismail1976
    @ismail1976 5 หลายเดือนก่อน

    Why you trained 40k and dont 48k ?

  • @ArtificallyIntelligentAi
    @ArtificallyIntelligentAi ปีที่แล้ว

    Wow!

  • @QHawk7
    @QHawk7 ปีที่แล้ว

    *Open Source & Free: Tenacity , instead of Audacity*

  • @patricksweetman3285
    @patricksweetman3285 ปีที่แล้ว

    Wow

  • @dontmindmejustwatching
    @dontmindmejustwatching ปีที่แล้ว

    best

  • @timeTegus
    @timeTegus ปีที่แล้ว

    very nerdy :)

  • @Ravisidharthan
    @Ravisidharthan 8 หลายเดือนก่อน

    Why my trained sound not sounds not as intended? What can go wrong?

    • @NerdyRodent
      @NerdyRodent  8 หลายเดือนก่อน +1

      One thing that could go wrong is to have a very poor quality data set

  • @imagesas
    @imagesas ปีที่แล้ว

    torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 2.00 MiB (GPU 0; 4.00 GiB total capacity; 3.03 GiB already allocated; 0 bytes free; 3.49 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

    • @NerdyRodent
      @NerdyRodent  ปีที่แล้ว +1

      You’re better off using colab if you’ve got 4GB VRAM. Even games today will struggle on just 8GB!