Audio Data Augmentation Is All You Need

แชร์
ฝัง
  • เผยแพร่เมื่อ 8 พ.ย. 2024

ความคิดเห็น • 40

  • @CourseWare-xg2wq
    @CourseWare-xg2wq 8 หลายเดือนก่อน

    This changed my every view on data augmentation 😮, thnx much Valerio

  • @jeb4148
    @jeb4148 2 ปีที่แล้ว +1

    Thanks for all the content!! Just finished up the audio signal processing series, excited to jump into this one

  • @nmirza2013
    @nmirza2013 ปีที่แล้ว

    I can only say Stay blessed for such a good work

  • @SHADABALAM2002
    @SHADABALAM2002 2 ปีที่แล้ว

    Much needed and awaited topic

  • @lorenzo87052
    @lorenzo87052 2 ปีที่แล้ว

    Already waiting next video of this topic! Thank you Valerio!

  • @AbhinayKhoparzi
    @AbhinayKhoparzi 2 ปีที่แล้ว

    This is so insightful. Eagerly waiting for more in the series.

  • @kalkidangazahegn5038
    @kalkidangazahegn5038 2 ปีที่แล้ว

    Thanks a lot Valerio, can I apply this audio data augmentation technique to heart sound signal?

  • @AbdullahOlcay-wh4or
    @AbdullahOlcay-wh4or ปีที่แล้ว

    Thanks very much Valerio, but why using a coupled code (online augmentation) makes it disadvantageous or advantageous (in offline augmentation). I think this depends on how flexibly one can use augmentation and model together without overfitting and with getting a high accuracy.

  • @AdrianFernandezFazio
    @AdrianFernandezFazio 2 ปีที่แล้ว

    Amazing vídeo! Greetings from Argentina!

  • @syllacamara5504
    @syllacamara5504 2 ปีที่แล้ว

    Thanks a lot for your great work.
    Will you do a video on speech recognition with CTC loss function ?
    Best regards

  • @erikgoron5928
    @erikgoron5928 2 ปีที่แล้ว +1

    Great video, thanks Valerio!
    Question :
    Do you have any tips on how to improve generalizability asides from data augmentation for audio tasks?
    I’m currently working on a speech pathology classifier, it works great on single datasets but when I use the trained models on unseen data from different datasets (same task) it doesn’t perform well anymore. Meaning it would not be a good model in production..
    Any help is appreciate it.

    • @ValerioVelardoTheSoundofAI
      @ValerioVelardoTheSoundofAI  2 ปีที่แล้ว

      I addressed this topic in a Linkedin post this wekk. Here's the link: www.linkedin.com/posts/valeriovelardo_ml-data-ai-activity-6886601744583860225-ymSw

  • @heychamp2054
    @heychamp2054 2 ปีที่แล้ว

    Thanks, Valerio for this amazing content!
    Question: how can we use onsets for finding similarities between two audio files?
    I am currently working on a project. Please help me out.
    Any help will be appreciated.

  • @ignaziogulino
    @ignaziogulino ปีที่แล้ว

    Complimenti Valerio!

  • @doyourealise
    @doyourealise 2 ปีที่แล้ว

    thank you so much! because of you this kind of videos exist :)

  • @SHADABALAM2002
    @SHADABALAM2002 2 ปีที่แล้ว

    Hello Valerio... How can we ensure that semantics are not lost after augmentation.

  • @Kyrios_X
    @Kyrios_X 2 ปีที่แล้ว

    Hi Valerio! Nice topic, can't wait for your implementation. I am training a CNN with a 7k word audio dataset. Could I reduce underfitting by using data augmentation ?

    • @ValerioVelardoTheSoundofAI
      @ValerioVelardoTheSoundofAI  2 ปีที่แล้ว +1

      I don't think augmenting data will resolve underfitting. For that, increasing model complexity should help.

    • @Kyrios_X
      @Kyrios_X 2 ปีที่แล้ว

      @@ValerioVelardoTheSoundofAI I am using your CNN implementation for the music genre classification. Underfitting tends to decrease after increasing epochs number. Do you think I must increase model complexity or I should find the ideal number of epochs?

    • @ValerioVelardoTheSoundofAI
      @ValerioVelardoTheSoundofAI  2 ปีที่แล้ว

      @@Kyrios_X first increase # of epochs. If you still underfit, try increasing complexity.

  • @ivanstepanovftw
    @ivanstepanovftw 6 หลายเดือนก่อน

    Audio in this video augmented too? :)

  • @maryamkhoshkhoo9030
    @maryamkhoshkhoo9030 2 ปีที่แล้ว

    Do you have classes to participate for deep learning eeg in python?

  • @BRAGAARS
    @BRAGAARS 2 ปีที่แล้ว

    Hello Valerio, just wondering, did you ever work on audio event detection? Been currently trying to build DNNs that detect a specific sound in audio files through image representation such as spectrograms and can't find a lot of things so I am wondering you ever templated doing something similar.

    • @ValerioVelardoTheSoundofAI
      @ValerioVelardoTheSoundofAI  2 ปีที่แล้ว +1

      I've worked on audio event detection. If the data is good, both spectrograms and Mel spectrograms should be valuable representations to use. If I had to pick one, I'd suggest the latter.

    • @Jononor
      @Jononor 2 ปีที่แล้ว +1

      Sound Event Detection using Machine Learning (EuroPython 2021)
      th-cam.com/video/JrhsFfCOL-s/w-d-xo.html

  • @madmike0304
    @madmike0304 ปีที่แล้ว

    what about audio data augmentation for music generative models?

  • @debabratagogoi9038
    @debabratagogoi9038 2 ปีที่แล้ว

    Thank you for the nice explanation

  • @mahmodaldahol
    @mahmodaldahol 2 ปีที่แล้ว

    this is exciting topic
    thank you

  • @MLDawn
    @MLDawn 2 ปีที่แล้ว

    really nice. well done man!

  • @Sawaedo
    @Sawaedo 2 ปีที่แล้ว

    If there would be a conversion from real_data to augmented_data, how much augmented_data do we need to compensate a decrease in real_data?

    • @ValerioVelardoTheSoundofAI
      @ValerioVelardoTheSoundofAI  2 ปีที่แล้ว +2

      You should keep the original data in the dataset and add the augmented samples to it. As a general rule I would produce 2-4X the original data. More than that you start having too much redundancy.

  • @Bindassmanus
    @Bindassmanus 2 ปีที่แล้ว

    Sir Can you help for detection of vowel like region in speech signal i have my mtech dissertation project on it

    • @ValerioVelardoTheSoundofAI
      @ValerioVelardoTheSoundofAI  2 ปีที่แล้ว

      I suggest you to ask a question with specific details of the problems you're encountering in The Sound of AI Slack.

    • @Bindassmanus
      @Bindassmanus 2 ปีที่แล้ว

      @@ValerioVelardoTheSoundofAI ok thanks for replying

  • @sutarorem8297
    @sutarorem8297 2 ปีที่แล้ว

    I love you so much