D. Audio :: Teachable Machine
AI DJ: Train Your Own Sound Recognizer! 🎧🤖
Have you ever wondered how your phone knows what song is playing, or how voice assistants understand your commands? It’s all about AI learning to recognize patterns in sound. Today, you’ll teach an AI to recognize your custom sounds!
Your Mission:
- Teach an AI to identify at least 2-3 distinct sounds.
- Experiment with how the AI learns and improve its accuracy.
- Think about how this technology could be used in music or other creative ways.
Materials You’ll Need:
- A computer or Chromebook with internet access
- A microphone (built-in is fine)
- Your creative brain and some sounds you can make! (e.g., snapping, clapping, tapping, a specific word, a simple beatbox sound)
Let’s Get Started!
Step 1: Brainstorm Your Sounds 🧠
- Think of 2 or 3 unique sounds you can easily make. These will be your “sound categories.”
- Example categories: “Snap,” “Clap,” “Whistle,” or “Hello,” “Goodbye,” “Hola,” “Adios,” “Bienvenidos”
- Goal: Choose sounds that are different enough for the AI to learn to tell them apart.
Step 2: Go to Teachable Machine 💻
- Open your web browser and go to teachablemachine.withgoogle.com.
- Click on “Get Started.”
- Select “Audio Project.”
- Choose “Standard audio models.”
Step 3: Record Your Sound Samples 🎤
You’ll see “Class 1,” “Class 2,” etc. These are your sound categories.
- Rename “Class 1” to the name of your first sound (e.g., “Snap”).
- Click the “Mic” icon under your first class.
- Record Sounds:
- Click and hold “Record 2s” and make your sound clearly into the microphone. Do this at least 10-15 times for each sound category.
- Tip: Try to make the sound similar each time, but also include slight variations if you want your AI to be more flexible. Make sure there’s not too much background noise!
- Rename “Class 2” (and “Add a class” if you have more than two sounds) and repeat the recording process for your other sound(s).
- You need at least 2 seconds of background noise samples too. Teachable Machine usually adds a “Background Noise” class automatically. Record at least 10-15 samples of silence or the typical background noise in your room.
Step 4: Train Your AI Model 🚀
- Once you have enough samples for all your sound classes (including background noise), click the “Train Model” button.
- Be patient! It might take a few moments for the AI to learn from your sounds. You’ll see a “Training…” message.
- What’s happening? The computer is analyzing all your sound examples and learning to tell the difference between your “Snap,” “Clap,” etc.
Step 5: Test Your AI! 🧪
- Once training is done, you’ll see a “Preview” section.
- Try making your sounds into the microphone. Does the AI correctly guess which sound you’re making? Look at the output bars – the AI will show how confident it is.
- Troubleshooting:
- Not working well? Maybe you need more sound samples for each class. Or perhaps your sounds are too similar. Try adding more diverse examples or making your sounds more distinct. You can go back, add more samples, and click “Train Model” again!
Step 6: Explore and Create ✨
This is where you can get creative!
- Basic Challenge: Can you get your AI to reliably recognize your chosen sounds?
- Next Level Thinking:
- How could a musician use this? Imagine triggering different drum sounds or musical effects with custom vocal sounds or hand claps.
- Could you use this to create a simple game or interactive story based on sound cues?
- Experiment: What happens if you make a sound that’s not one of your classes? How does the AI react?
Step 7: Share & Reflect 🗣️
- Ask a classmate to try and replicate your sounds and show them how your AI sound recognizer works!
- Discuss:
- What was the hardest part of training your AI?
- What was something surprising you learned?
- Can you think of three real-world applications for audio recognition technology (besides music)? (Hint: think about accessibility, safety, or even nature!)
Part 2: AI for Accessibility – Sound Solutions! 🔊❤️
What is Assistive Technology?
Assistive Technology (AT) is any device, software, or equipment that helps people with disabilities learn, work, communicate, or just live more independently. Think of things like screen readers for people who are visually impaired, or wheelchairs for those with mobility challenges.
Your AI sound model can be the starting point for creating an assistive technology!
Your New Mission:
- Brainstorm how your sound recognition AI (or a new one you train) could be used as an assistive technology.
- Design a concept for an assistive device or app that uses sound recognition to help someone.
- Think about who it would help and how.
Let’s Design for Good!
Step 1: Think About Needs 🤔
- How could recognizing specific sounds help someone? Consider different challenges people might face:
- Someone who is deaf or hard of hearing.
- Someone who has difficulty speaking.
- Someone with limited mobility who needs hands-free ways to interact with their environment.
- Someone who needs alerts for safety or health.
Step 2: Brainstorm Your Assistive AI 💡
Now, think about your Teachable Machine project.
- Option A: Adapt Your Current Model: Could the sounds you already trained (like “snap,” “clap,” or simple words) be used for an assistive purpose?
- Example: If you trained a “help” sound, how could that trigger an alert for a caregiver?
- Option B: Train a New Model for a Specific Need:
- What sounds would be crucial for your assistive idea?
- Examples for a hearing impaired person: “Doorbell,” “Smoke Alarm,” “Baby Crying,” “Kettle Whistle.”
- Examples for a non-verbal person: Distinct vocal sounds they can make that your AI could translate into simple commands (e.g., a hum for “yes,” a click for “no”).
- If you choose this option, go back to Teachable Machine and train a new audio model with these specific sounds. Remember to get enough good samples!
- What sounds would be crucial for your assistive idea?
Step 3: Design Your Assistive Technology Concept ✍️
This isn’t about building a full app right now, but about designing the idea. Answer these questions for your concept:
- What is the name of your assistive technology? (Be creative!)
- Who is it designed to help? (Be specific about the user and their needs.)
- What specific sounds will your AI recognize?
- How will it help the person when it recognizes those sounds? What action will it trigger?
- Examples:
- Flash a light on a paired device.
- Send a text message to a family member.
- Speak a pre-recorded message (e.g., “The doorbell is ringing,” or “I need assistance”).
- Vibrate a wearable device.
- Examples:
- Sketch it out! Draw a simple picture or create a storyboard showing how someone would use your assistive technology. How would it look? How would it work in a real-life scenario?
Step 4: Present Your Idea 📢
- Share your assistive technology concept with your classmates or facilitator.
- Explain:
- Who your AT is for.
- What sounds it detects.
- How it helps the user.
- Why you think it’s a useful idea.
Discussion Points:
- What are some challenges in designing AI for assistive technology? (e.g., accuracy, privacy, different user needs)
- How can AI make the world more accessible and inclusive?
- Did this change how you think about the power of simple sound recognition?
By thinking about assistive technology, you’re not just playing with AI – you’re exploring how it can make a real difference in people’s lives. This is a huge and growing field in STEM, combining technology with empathy and creative problem-solving!
