Valerio Velardo’s Post

Is #Suno as creative as a human musician? To answer this question, I need to introduce two concepts: exploratory creativity and transformational creativity. Bear with me. I promise it’ll be worth the read! Joe is a musician. He loves Elvis Presley’s music so much. In his music collection, you can find all of Elvis Presley’s albums and extensive memorabilia. When Joe writes music, he can’t help but write music that sounds like Elvis’s. Joe is a really good composer. His songs are terrific melodically and very well produced. But they feel like an echo of Elvis. Is Joe creative? The music is novel – there’s no Joe’s song that’s an exact copy of an Elivs’ song. Joe’s music is appreciated by many people. So, yeah, according to your typical definition of creativity (novelty + value for someone), Joe’s writing is creative. However, is it REALLY creative? The problem with Joe’s music is that it is constrained within the borders of a specific music style. It follows unwritten rules. A sort of hidden Elvis musical grammar, if you will. Joe’s music is within the creative space of Elvis Presley. So, Joe is creative in that he explores new items within Elvis’s creative space. Joe is exploratively creative. Compare Joe’s creativity to, say, Arnold Schoenberg. This Austrian composer initially wrote music in the late Romantic style, but over time, he came out of the late Romantic style. He started to break unwritten rules, only to arrive at a completely new grammar (i.e., dodecaphony), which was completely original. Schoenberg transformed the musical grammar of his time. He re-wrote the rules. The mature Schoenberg was transformationally creative. Back to the question. What’s the difference between AI and musicians’ creativity? Both humans and machines show exploratory creativity. They can create music in a precodified style that is novel within that musical grammar and has value for somebody. However, only humans show transformational creativity. To be realistic, only a small fraction of humans can come up with game-changing ideas that transform music. Be aware though that there’s no reason why machines in the future won’t be able to have transformational creativity. There is nothing special about human creativity. Why can the likes of #Udio and Suno only explore but fail to transform? Because of the architecture of their models. On a very high level, these are sophisticated memory machines that can re-combine what they have seen at training time. But they can’t go beyond what they have seen. By design. What do you think? How are machine and human creativity different? I’m currently writing about this topic for the book I’m working on, and am curious to know your take on this. #GenerativeMusic #Music #Creativity #AIMusic #MusicAI =================================== 🎹 AI music consulting -> https://lnkd.in/dTJnxune 🤖 AI music recruitment -> https://lnkd.in/dQyDMkN2 Get my content on Substack -> https://lnkd.in/d7MPrPQT

Major record labels are suing generative AI music companies #Suno and #Udio for copyright infringement. They accuse these tech companies of using their music catalogs to train generative music models without permission. For those who have eyes, this move was more likely to happen than the sun rising tomorrow. I read many people commenting on the news, saying this is a positive outcome. I disagree. First, there is a good chance this lawsuit is theatre. The majors may be suing Suno and Udio to force them into accepting copyright deals. Spotify vibes, anybody? Second, I’m skeptical that the majors are going against AI music generators to protect their artists. Rather, I think the lawsuit happened because they don’t have control over the generative models. They would rather have their own generative music models and, potentially, cut artists out. #UniversalMusic deal with #SoundLabs to create AI models of their singers’ voices seems to go in this direction. Third, the elephant in the room is that only a fraction of the training dataset used by AI music generators likely comes from the majors. The vast majority of the music may come from smaller labels and independent artists. Who’s going to protect these folks? The democratization of music-making through AI is inevitable, and it can be a good thing. For example, it could provide artists with a new income stream that wasn’t possible before. However, it must be done the right way. In my view, that means: - Open sourcing generative music AI models - Compensating artists fairly - Being transparent about music data - Removing the middleman. Artists can share their music directly with AI music developers. No need for a label. What do you think? #GenerativeMusic #AIMusic #Music #Universal #Sony #Warner #Suno #Udio #AIEthics #Music =================================== 🎹 AI music consulting -> https://lnkd.in/dTJnxune 🤖 AI music recruitment -> https://lnkd.in/dQyDMkN2 Get my content on Substack -> https://lnkd.in/d7MPrPQT 

I'm really looking forward to giving a talk during the workshop organised by the AI Song Contest. I'll provide a quick presentation about generative AI music tools. If you're into #AIMusic consider signing for the workshop! #AI #GenAI #GenerativeMusic

It's the final day of the 2nd edition of the Generative Music AI workshop in #Barcelona, I organised in collaboration with Music Technology Group. What a ride! Yesterday, we had a fantastic presentation by Xavier Serra on the music AI landscape. This morning, I led a debate on how AI is going to shape the future of music. So many interesting insights! Most importantly, yesterday teams put on an amazing concert using the generative music AI technology they developed during the workshop. The performances were incredible (you can get a sneak peek in the video below). The blend of tech and musicianship was over the roof, considering teams had 4 days to prepare everything. Would you like to see the concert published on The Sound of AI YouTube channel? This evening, we'll close the event with presentations from the teams. They'll explain what they built and, most importantly, how they did it. I'll keep you posted! #GenerativeMusic #GenAI #AIMusic #GenerativeMusicAIWorkshop #AI ================================== 🎹 AI music consulting -> https://lnkd.in/dTJnxune 🤖 AI music recruitment -> https://lnkd.in/dQyDMkN2 Get my content on Substack -> https://lnkd.in/d7MPrPQT

3rd day of the Generative Music AI Workshop in #Barcelona. with Music Technology Group. Today, Roser Batlle Roca is leading an amazing session on #GenerativeMusic and ethics. Participants are actively discussing important ethical questions such as: - Is it fair to train generative music models on copyrighted data? - Is musicians' consent fundamental for training? - Should AI-generated music be copyrighted? - Can AI-generated music be new and original? Positions are quite strong, sometimes contrasting. I'm secretly taking notes/inspiration for my book on #AIMusic ;) Beyond this session, teams are working hard on their projects. Some already have an initial proof of concept! #MusicAI #AI #Ethics #AIEthics #MusicTech #GenerativeAI ================================== 🎹 AI music consulting -> https://lnkd.in/dTJnxune 🤖 AI music recruitment -> https://lnkd.in/dQyDMkN2 Get my content on Substack -> https://lnkd.in/d7MPrPQT

The 2nd Generative Music AI Workshop has started 🚀 Yesterday was fantastic. Participants from all over the world have gathered in #Barcelona for this event. They come from diverse backgrounds, including AI researchers, musicians, software developers, musicologists, and cognitive scientists. Yesterday, they: 1. Networked extensively 2. Self-organized into teams 3. Came up with workshop ideas related to #GenerativeMusic Over the next four days, they'll work intensely to build an app, performance, or product where humans and machines collaborate to create music. For me, it’s wonderful to be back in Barcelona for another collaboration between The Sound of AI and Music Technology Group. This place is starting to feel like home. My highlight of the day was the Speed Dating session. We divided participants into two lines and had them chat for one and a half minutes, before moving on to the next person. It was a super noisy environment, but super fun (check out the pictures below!). Above all, it’s just amazing to work with a community of people who live and breathe music, computing, and AI. I’ll keep you posted on how this continues! #GenMusAIWorkshop #GenerativeMusic #MusicAI #AIMusic #AIMusic #MIR =================================== 🎹 AI music consulting -> https://lnkd.in/dTJnxune 🤖 AI music recruitment -> https://lnkd.in/dQyDMkN2 Get my content on Substack -> https://lnkd.in/d7MPrPQT

📢 HIRING - #Speech ML Job Offer At The Sound of AI, we’re hiring a Speech ML Research Engineer (2 positions) for Play. Check the link to the job posts in the comments 👇 for more info and for applying. Please share this job post with your network to help reach interested candidates  🙏 #JobPost #Speech #VoiceCloning #AudioTech #SpeechSynthesis #TTS #MachineLearning =================================== 🎹 AI music consulting -> https://lnkd.in/dTJnxune 🤖 AI music recruitment -> https://lnkd.in/dQyDMkN2 Get my content on Substack -> https://lnkd.in/d7MPrPQT

Want an #AI job? Avoid these mistakes when applying. I’ve been hiring AI audio/music engineers for tech companies for a few years now. The majority of applications I receive are low quality. I’m not talking about candidates' professional or academic track records—I’m referring to the quality of the applications themselves. I assume you are a strong candidate. That’s necessary but not sufficient to get noticed. To stand out, your application must be well-crafted. Most candidates fail at this. Do it right, and you’ll dramatically increase your chance of getting your dream job. Avoid these mistakes at all costs: 1. Generic cover letter. Your cover letter is my first glimpse into who you are and why you’re a fit for the role. Too many are copy-pasted with only the company name changed. I can smell those every time. Show me who you are, your goals, and why this position excites you. Be specific and provide concrete examples of how your experience and skills match the requirements in the job post. 2. Generic CV. Tailor your CV specifically for the AI (music) role you’re applying for. Highlight relevant experience and transferable skills, even if your past roles were different. 3. Not following application instructions. If I request specific information (e.g., LinkedIn, GitHub), please provide it. Failing to do so signals a lack of attention to detail, and I’ll skip over your application. 4. Asking for information already in the job posting. If the job description says the job is remote, don’t message me asking if it’s remote. This shows you haven’t read the posting carefully. Of course, questions are welcome. But make sure they haven’t already been answered. 5. Overly long cover letters and CVs. Keep your CV to 2 pages and your cover letter to 1 page. Focus on the most important information. I skim cover letters first, then CVs. If they catch my interest, I’ll read them in full. Present your most critical information upfront. 6. LinkedIn and GiHub wastelands. An inactive LinkedIn or GitHub profile is a no go. Once I go past your CV and cover letter, I check these platforms to get a sense of your professional and technical background. Ensure your LinkedIn profile is up-to-date with relevant experiences, skills, and endorsements. Similarly, your GitHub should showcase your best projects, demonstrating your coding abilities and interests. In my view, it all comes down to quality over quantity. Instead of sending 100 generic applications, focus on 10 where you are a great fit. Tailor your application. This is especially crucial for highly specialized roles like AI Music Engineer. I never follow up on a generic application. I hope this helps. #JobAdvice #AIJobs #AIMusic #MusicTech #AudioTech #Speech ===================== 🎹 AI music consulting -> https://lnkd.in/dTJnxune 🤖 AI music recruitment -> https://lnkd.in/dQyDMkN2 Get my content on Substack -> https://lnkd.in/d7MPrPQT 

Wanna learn Deep Learning (for Audio) with Python? I published a full course on The Sound of AI YouTube channel. It’s FREE. Hundreds of thousands of people have taken it so far. It's also the first course I've ever published on The Sound of AI. So, I'm super fond of it ;) The course will teach you all you need to get started in the world of deep learning, with a particular focus on audio and music processing. It’s organised in 19 videos that cover the intuition, theory, math and code implementation. You even get to implement a neural network from scratch - backpropagation and all the rest included- without any DL framework! Here’s a list of topics: 1. Overview 2. AI, machine learning and deep learning 3. Implementing an artificial neuron from scratch 4. Vector and matrix operations 5. Computation in neural networks 6. Implementing a neural network from scratch in Python 7. Training a neural network: Backward propagation and gradient descent 8. Implementing backpropagation and gradient descent from scratch 9. How to implement a (simple) neural network with TensorFlow 2 10. Understanding audio data for deep learning 11. Preprocessing audio data for Deep Learning 12. Music genre classification: Preparing the dataset 13. Implementing a neural network for music genre classification 14. Solving overfitting in neural networks 15. Convolutional Neural Networks Explained Easily 16. How to Implement a CNN for Music Genre Classification 17. Recurrent Neural Networks Explained Easily 18. Long Short Term Memory (LSTM) Networks Explained Easily 19. How to Implement an RNN-LSTM Network for Music Genre Classification I’ve also shared the slides and the code, so you can easily follow along. Check the link to the course in the comments 👇 Have fun!