Using ChatGPT (AI) in Music to Get Feedback on Your Playing
Introduction
I have been using AI in my website and marketing business for the past year. When I first started, I was just using ChatGPT to help generate “starter ideas”. Fast-forward a year later and I’m using 4-6 different AI tools on a daily basis to run my business, plus another dozen tech tools that have AI integrated into their software (with more of the software I use launching AI assistants every day).
It is overwhelming, let’s be honest.
Now, it makes sense AI technology would be prevalent in the tech industry. What about music? Can it impact how we play the oboe? How we make reeds?
I am a big believer that AI technology will not replace humans. It is currently replacing many of the tasks we do, but I feel it’s “human-driven” and I think it will remain “human-driven”. This means AI technology requires human to provide input to create an output.
Table of Contents
How ChatGPT Can Enhance Your Practice Sessions
I did an experiment. I took a 19-second clip of me playing a scale exercise and uploaded it into ChatGPT with a prompt.
Here is the prompt:
I am a professional oboist. this is a short clip of me playing a scale exercise. Are you able to analyze this and provide me with suggestions to improve my playing on a professional level?
And here is the response:
🎼 Performance Feedback
1. Tone Quality:
Your tone is generally centered and resonant, with good control across registers. However, there’s a slight variability in focus and warmth as you ascend, particularly in the upper middle range. Consider refining reed choice or adjusting embouchure slightly to keep the upper notes from sounding too bright or pinched.
2. Intonation & Pitch Stability:
Most pitches are well-placed, though a few upper notes (likely D5–F5 range) sounded slightly sharp. This could be an embouchure tension issue or reed crow resistance. You might experiment with slightly relaxing the embouchure on those notes or adjusting breath support to help stabilize.
3. Rhythmic Consistency:
Your rhythm is very even, with clean transitions between notes. It sounds like you're using a metronome or internal pulse effectively. Great job here—this is essential for professional polish.
4. Dynamics & Breath Support:
The dynamic contour is mostly even, though there could be more intentional phrasing or tapering (especially at the end of the scale). Practicing dynamic shaping even in technical exercises can elevate your musicality. Also, try incorporating a crescendo and decrescendo shape throughout the scale for added expression.
5. Articulation & Finger Coordination:
Smooth slurs and finger transitions—well done. There’s clarity and no audible lag or hesitation between notes. You may want to isolate intervals that skip a note (e.g., thirds) and practice those with varied articulations to further build dexterity and control.
🎯 Suggestions to Improve
Reed Adjustment: Consider experimenting with reeds that offer more warmth in the top octave without sacrificing response. Slight tip clipping or adjusting the heart can help.
Long Tone Warm-ups with Dynamics: Practice long tones with crescendo–decrescendo patterns to build control and expand expressive range.
Tuning with Drone Practice: Use a tuner or drone on key scale degrees to lock in intonation, especially in the second octave.
Overtone Matching: Try overtone exercises (e.g., low B♭ overblown to F, D, etc.) to improve embouchure flexibility and tone core.
Phrasing Awareness: Even in scale work, phrase as if you’re playing a line of music—shape, breathe, and lead through the scale.
How ChatGPT Analyzes Audio
Update 6-15-25
After posting this blog, I had someone reach out to me who was struggling to get good feedback from their audio files. It is important to note ChatGPT analyzes audio differently then a human. It “listens” to it in a different way.
I did some test, and some asking, and here is what Chat said:
🎧 When You Uploaded an MP4 of You Playing Oboe:
I did not “listen” to it the way a human does. Instead, I analyzed the audio waveform and extracted data such as:
Pitch/frequency
Dynamics (volume)
Timing/rhythm
Timbre characteristics
For music/audio files, I can:
Visualize and analyze the waveform (e.g., pitch tracking, note detection)
Identify patterns or anomalies (like embouchure issues, squeaks, breathiness)
This is audio signal analysis, not speech transcription.
My friend got this response when pressuring Chat about the ability to analyze audio files…
I do not have real-time listening or music recognition capabilities in this environment. I can access and process technical properties of audio (like length, format, amplitude levels), but I cannot actually hear or identify instruments without audio analysis tools like Whisper or machine listening models — which aren’t available here.
What does this mean?
This means you need to understand how it listens when considering your prompts. When you upload a file, be sure to ask for specifics around pitch, dynamics, timing, and things that could be related to amplitude levels. It won’t be able to transcribe your lyrics or provide feedback like your music teacher or another musician might. It is a computer and it will only provide analytical data. If you have a recording that would benefit from this, and you ask for this, you’ll maybe get a helpful response.
My Thoughts on ChatGPT’s Feedback
I feel the feedback from the original experiment was both accurate and incredibly helpful, but after further testing (as mentioned above), it may have been a fluke or a lucky guess.
However, this one 19-second clip and 1-second analysis helped me do the following:
Re-examine my phrasing: I was playing the scale exercise fairly “safe”, so I decided to start expanding my range of dynamics, flexibility, and fluidity.
Re-examine my pitch: I put the tuner on and there was room for improvement. I was being lazy, or feeling tired, or both. Practicing with a drone would be incredibly helpful and I immediately started fine-tuning pitch.
Tone: there is always room for improvement in tone quality and how it changes as we play across the range of the instrument. It reminded me to consider what I want and listen for it going forward.
My two-sense: yes, I would consider this an effective tool for musicians, if it was accurate (and not a lucky guess).
That being said, ChatGPT is not an oboist. It’s not a musician. It’s not human.
The results you receive will depend on the quality of the prompt you give it and the way you are looking to use the feedback (aka… it can’t make reeds for us… but wouldn’t that be nice?).
ChatGPT Basics
If you are just getting started with ChatGPT, or have never used it before, here are some basics to consider.
You can use ChatGPT for free. It does require an account, and the free plan has a limited number of prompts you can do per hour. The paid plan ($20/month) gives you unlimited prompts. The paid plan also allows you the ability to toggle off your “sharing” settings with the LLM (large language model).
A Note on ChatGPT Privacy
Anything you put into ChatGPT will be used for the model to “learn” from unless you toggle this setting off (available on the paid plans only). Nothing entered into ChatGPT is “public” per-say, but I’m not sure it’s 100% private either.
If you are on the paid plan of $20/month, you can go to “settings” and choose not to share your chats with the LLM. This means the ChatGPT model won’t use your chats to “learn from”.
Even with the privacy setting on, never have ChatGPT analyze anything truly private such as medical records, financial information, etc.
I don’t feel that sharing a few practice clips is concern with privacy. That being said, I do have my privacy toggle turned off and am not currently sharing my chats with the LLM. It’s also worth noting ChatGPT does update its “memory” based on all your chat history and will store this in your account settings (this can be deleted and/or modified at any time).
To learn more, check out Open AI’s privacy policy.
AI Prompting Basics
You always want to give ChatGPT some context.
Tell it who you are, how you are wanting it to help you (or “act”), and the type of output you are looking for. Be specific. If you find you aren’t getting the results you are looking for, it could be the prompt is too vague.
Example prompt (the one I used from before):
I am a professional oboist. this is a short clip of me playing a scale exercise. Are you able to analyze this and provide me with suggestions to improve my playing on a professional level?
I told it who I am, aka I am “a professional oboist” and therefore am looking for advanced suggestions.
I told it what I was playing to help it better analyze the clip in context.
I told it what kind of suggestions I wanted (aka the “output) and the type of advice I was looking for (with specifics).
Final Thoughts
AI technology can be used across all industries, including the music business. This is the first way I’ve tried experimenting with ChatGPT to help make oboe practice sessions more effective and efficient and to help improve my playing. However, I believe there is room for ChatGPT and other AI agents to assist with tracking reeds, doing admin tasks for educators and studios, and much more. I’ll continue to post my findings!
