Browser-native audio decomposer
Inspect AI music with math-first precision.
Analyze and inspect frequency regions in your track. This decomposer runs real STFT spectrogram analysis in your browser, built for AI music workflows but friendly to any song you upload.
1. Upload a track
Free, local-only processing, so no uploads or logins required!
1. Analyze upload
Run the inspection pass locally to generate math diagnostics, spectrogram views, and composition charts.
2. Math inspection
Quick numeric diagnostics from waveform sampling (RMS, crest factor, DC offset). Great for spotting AI artifacts.
| Metric | Value |
|---|---|
| Core | |
| Duration | -- |
| Sample rate | -- |
| Channels | -- |
| Peak / RMS | -- |
| Crest factor | -- |
| Zero crossing rate | -- |
| DC offset | -- |
| Estimated BPM | -- |
| Spectral | |
| Style profile | -- |
| Spectral flatness | -- |
| Spectral centroid | -- |
| Spectral rolloff (95%) | -- |
| High-band energy ratio | -- |
| Spectral flux | -- |
| Transient sharpness | -- |
| Texture | |
| Texture confidence | -- |
| Texture uncertainty | -- |
AI texture score (heuristic)
--
Higher values suggest noisier, brighter, or more artifact-prone texture.
3. Spectrogram map
An STFT-based time-frequency map of your track. Use it to spot dense bands, transient spikes, and AI artifacts.
Resolution
Focus band
4. Composition views
Multiple perspectives on the same audio: energy over time, frequency balance, and brightness movement.
5. Analysis focus
This page is built for inspection, not export. Use the visual masks + charts to diagnose AI artifacts and understand structure without altering the file.
What the charts mean
- • Energy timeline highlights loudness shifts and section changes.
- • Frequency balance shows where the mix lives (sub, mid, air).
- • Brightness drift tracks how tonal color evolves over time.
AI inspection focus
- • Look for persistent high‑band haze or metallic spikes.
- • Compare crest factor dips (over‑compression) to known references.
- • Watch for abrupt texture shifts without musical intent.
AI music knowledge bank
Quick, grounded facts about what you’re seeing — plus how to level up the math when you want true spectral analysis.
What the STFT spectrogram shows
- • The audio is split into overlapping frames and windowed with a Hann curve.
- • Each frame runs an FFT to measure real frequency magnitudes.
- • Magnitudes are converted to dB, normalized, and plotted over time.
- • Frequency bins are remapped to keep low-end and high-end inspection readable.
- • The emphasis slider boosts upper-band contrast without injecting synthetic noise.
Why crest factor matters
- • Crest factor = peak / RMS (dynamic contrast).
- • Over‑limited mixes often show low crest (flat dynamics).
- • AI renders sometimes exaggerate transients → higher crest spikes.
AI texture score formula
- • STFT features used: flatness (F), centroid (C), rolloff (R), high-band ratio (H), flux (X), transient sharpness (T).
- • Waveform terms: zero-crossing rate (Z), crest anomaly (K), and DC offset (D).
- • First, a style profile is inferred (Warm, Balanced, Percussive, Textured), then each term is normalized to that style.
- • Score = clamp(0.17F + 0.12C + 0.10R + 0.18H + 0.14X + 0.11T + 0.08Z + 0.07K + 0.03D).
- • Confidence uses score boundary distance, component consistency, and analyzed duration; uncertainty = 1 - confidence.
AI artifact checklist
- • Persistent hiss between 8–14kHz.
- • Metallic “chirps” around 2–4kHz clusters.
- • Smearing of transients (blurred drum hits).
- • Sudden tonal shifts without arrangement change.