MixScope | Audio Mix Analyzer

The Overview

The goal of the project was to create a tool that goes beyond basic audio metrics and delivers actionable, engineer-oriented feedback. MixScope focuses on providing insights that are actually useful in a professional mixing context. The application processes uploaded audio files, extracts advanced audio features, and translates this data into structured, technical feedback using an LLM. This enables the client to quickly identify weaknesses in areas such as loudness, dynamics, stereo image, and spectral balance—either as a standalone analysis or in comparison with a reference track.

The Approach

To build the application, I began by defining the core analysis objectives together with the client and the audio engineer, identifying which audio features the client expected to extract from the analysis. Based on these requirements, I designed the initial MVP architecture.
I developed the backend in Python, using NumPy, Librosa, and other audio-processing libraries to extract audio features from uploaded tracks. From the beginning, I focused on keeping the processing pipeline efficient by limiting computation to essential and meaningful features only. Throughout this phase, I worked in close collaboration with the audio engineer, continuously validating the extracted data and refining the analysis pipeline to ensure both technical accuracy and practical usefulness for mix evaluation.
Once reliable audio features were available, I integrated an LLM to convert the extracted data into structured, professional feedback. In this phase as well, I worked closely with the audio engineer to design and refine a prompt that enforced technical language, clear structure, and actionable processing recommendations, such as EQ, compression, and stereo adjustments.
In parallel, I collaborated closely with the graphic designer to improve the UI/UX, aligning the visual design and interaction patterns with the client’s expectations. This process ensured that the information was presented in a clear, intuitive, and efficient way.
After validating both analytical accuracy and usability, I focused on performance optimization, refining the audio analysis pipeline and improving overall system responsiveness to meet the client’s requirements for speed and reliability.

The Outcome

To address performance issues, I redesigned the audio analysis pipeline to minimize redundant processing and limit computations to essential features only. I also implemented asynchronous processing, allowing multiple analysis tasks to run in parallel and improving overall system responsiveness. These optimizations reduced the average analysis time by approximately ~30%, while being accurate.
On the feedback side, refining the prompt in close collaboration with the audio engineer greatly improved the quality, consistency, and technical reliability of the generated responses. The final system provides detailed insights into stereo image, loudness, dynamics, and spectral balance, along with concrete suggestions on how to improve a mix.
MixScope is now a robust tool designed to simplify the daily workflow of music producers and audio engineers by delivering fast, reliable, and professional mix feedback.

Contact me for more information about the App and code.

Future Steps

Future improvements: Potential future improvements for the project include:

• Training or fine-tuning an LLM on a broader collection of professionally mixed and mastered tracks, in order to provide more realistic and context-aware feedback.
• Implementing genre-specific audio analysis, allowing the system to adapt feedback based on musical style.
• Introducing user profiles to store and track past analyses.
• Adding visual representations of the frequency spectrum, improving clarity and interpretability of the analysis.