Unveil frequently asked questions

How Unveil measures loudness and tone, how scores and genre curves work, Fix My Mix, the tools, supported DAWs, plans and privacy. Can't find an answer? Email support@unveilmusic.app.

Getting Started

What file formats does Unveil support?

MP3, WAV, AIFF, FLAC, M4A, and OGG. Files up to 200 MB. For the most accurate analysis use 24-bit WAV or FLAC at your project sample rate — MP3 already-encoded artifacts will show up in the spectral measurement.

How long does analysis take?

Typically 20–45 seconds for the technical analysis plus 10–20 seconds for the AI descriptive pass. Cold-start adds another 30 seconds on the first request after a quiet period — we pre-warm the backend the moment you select a file on the upload page so this is usually paid before you click Analyze.

Do I need to create an account to use Unveil?

You can try one free analysis without an account. A free account is required to save analysis history, track score progression across mix revisions, and use Fix My Mix.

What genres does Unveil support?

Fifteen at the moment: Modern Pop, Hip-Hop, Trap, R&B / Soul, EDM, Tech House, Lo-Fi, Indie, Rock, Drill, Afrobeats, Cinematic, Jazz, Classical, and Ambient. Each carries its own loudness, dynamic range, and spectral reference curve. You can also leave the genre on "Auto-detect" and let the engine pick from tempo + spectral fingerprint.

What's the difference between Mix and Master analysis modes?

Pick "Pre-Master Mix" if you're sending the track to a mastering engineer (or about to master it yourself). The scoring shifts to favor preserved dynamics and headroom: LUFS becomes a soft check (anywhere in -20 to -14 LUFS scores 100), LRA weight doubles, and we surface a Bus Headroom metric. Pick "Mastered Track" if it's release-ready — LUFS becomes the heaviest weight (18%) and the engine grades against streaming targets.

Standards & Methodology

What standards do you implement?

Integrated Loudness is ITU-R BS.1770-4 via pyloudnorm (K-weighting filter, two-pass gating). True Peak is BS.1770-4 Annex 2 with 4× polyphase oversampling per channel — most "true peak" meters in commercial DAWs use this same method. Loudness Range is EBU R128 / Tech 3342 with both the absolute gate (-70 LUFS) and the relative gate (-20 LU below the gated mean) applied before the 95th − 10th percentile calculation.

How are the genre reference curves built?

For each genre we measure a hand-curated set of commercially mastered tracks: K-weight, STFT into 7 frequency bands, normalize to relative spectral tilt (band − mean), then compute the per-band mean and standard deviation across the set. Your mix is scored against those statistics using a z-score with sigmoidal smoothing. The source list is published — see the About page. Until a corpus is built for a given genre, that genre uses hand-tuned heuristic offsets and we say so in the analysis output.

Why z-score scoring instead of linear deviation?

Because a 1 dB deviation in the sub-bass means something very different in Trap (where sub levels are tightly controlled) vs Jazz (where bass is loose by design). Z-score scaling lets each genre's tolerance reflect its real-world variance. We also apply sigmoidal smoothing so a tight-corpus genre doesn't score harder than a loose-corpus one in ways users find unfair.

How do you compute the "worst window" timestamp on each fix?

For frequency-band issues (boomy bass, harshness, vocal masking, etc.), we compute frame-wise energy in the affected band and find the 4-second contiguous window with the highest sum. For dynamic issues (over-compression, clipping, uneven dynamics) we find the window with the most extreme RMS variance. Global issues like "thin bass" or "low headroom" span the whole track.

Why does my score change when I switch genres?

Because each genre has its own targets and reference curve. A track at -8 LUFS scores great as Hip-Hop, mediocre as Modern Pop, and poorly as Classical — that's the engine doing its job, not noise. Pick "Auto-detect" to let the engine choose based on tempo, spectral contrast, and percussive ratio.

Analysis & Scoring

How is the overall score calculated?

A weighted average of 9–13 sub-scores depending on whether you picked Mastered Track or Pre-Master Mix. The weights for each mode are published (see the methodology page). For mastered tracks the heaviest hitters are LUFS (18%), True Peak (15%), Tonal Balance (12%), and LRA (12%). For mixes, LRA jumps to 14%, Crest Factor is added at 10%, and LUFS drops to a 6% soft check.

Why do I get different scores when uploading the same track?

You shouldn't — the scoring engine is deterministic. Same file plus same genre plus same track type = same score, every time. If you see differences, check that you're uploading the byte-identical file (a DAW re-export, even with no setting changes, can introduce 0.x dB differences that move sub-scores by 1–2 points).

My track sounds great to me but scored low. Why?

Unveil measures technical translation, not musical quality. A low score points to measurable problems — frequency masking where instruments fight for the same band, sustained negative phase correlation, low-end mono incompatibility, or excessive loudness that flattens transients. Open the Issues tab to see exactly what tripped it and the time window where it's worst.

What's the Versions tab for?

When you upload a revised mix of the same project, the engine carries the previous analysis forward and renders a per-metric comparison: which CLAUDE.md metric improved, by how many points, and the raw value change. Useful for proving to yourself that the EQ cut you made actually fixed the boomy bass instead of just moving it.

What does the Visualizations tab show?

Seven charts rendered server-side from the full track: waveform, spectrogram (frequency over time), frequency spectrum (energy distribution), chromagram (harmonic content), stereo field, vectorscope/goniometer (phase correlation), and dynamic range over time. They lazy-load when you open the tab so analyses without anyone opening them don't waste CPU.

Fix My Mix

What does Fix My Mix actually do?

It takes the same analysis you just read and feeds it into a biomimetic M/S mastering chain. Each frequency conflict the engine detected becomes a targeted EQ move: boomy bass becomes a bell cut at the exact Hz, harshness becomes a de-essing peak, sub imbalance becomes a high-pass. Then a true-peak limiter brings the master to your streaming target LUFS. Stems are used for vocal masking analysis only — the original stereo audio is what actually gets processed.

Is the 30-second preview really free?

Yes. The preview is the conversion lever — reading a report can't compete with hearing the fix on your own track. It generates a 30-second slice of the loudest section, runs the full mastering chain, and caches the output so re-visiting an analysis doesn't cost anything to render.

What's the difference between the three Fix My Mix tiers?

Unmask ($15) applies mastering only — vocal-clarity bells and LUFS targeting. Best for tracks that are already balanced. Fix + Master ($29) is the recommended default: every analyzer-detected conflict gets corrected, plus the biomimetic masking chain, plus the master. Surgical ($59) adds per-stem corrective EQ before the mixdown — for mixes with multiple severe issues that need root-cause fixing.

When does the Fix My Mix panel not appear?

The panel hides itself when none of your conflicts are correctable on a master bus. Flag-only issues (over-compression, uneven dynamics, clipping, low headroom) can't be safely "fixed" downstream — they're mix-stage problems. If those are the only issues found, the panel doesn't render because there's nothing for it to do.

Features & Tools

How does the MIDI transcription (Midify) work?

Unveil separates your track into stems (drums, bass, vocals, other) using Demucs on our cloud servers, then converts each stem to MIDI using Spotify's Basic Pitch. The MIDI is cleaned and quantized using music21 for musically intelligent results — including key detection and enharmonic spelling.

Is Lyrics Transcription accurate?

Accuracy depends on vocal clarity and production style. Clean lead vocals with minimal background elements typically transcribe well. Heavy autotune, ad-libs over dense production, or non-English content may reduce accuracy.

What is the Aural Preview?

Aural Preview lets you hear your track with the suggested EQ corrections applied in real-time. It builds a chain of EQ filters based on the specific issues detected in your mix, so you can A/B the original vs corrected version before touching anything in your DAW. Different from Fix My Mix: Aural Preview is browser-side EQ-only; Fix My Mix is full server-side biomimetic mastering.

What DAWs are supported for fix instructions?

Ableton Live, Logic Pro, FL Studio, Pro Tools, Cubase, REAPER, Studio One, and GarageBand. Each fix carries DAW-specific step-by-step instructions naming the actual stock plugins for that DAW.

Account & Access

What is included in the free plan?

During beta, everything: unlimited analyses, full scoring with genre-aware targets, frequency conflict detection with timestamps, DAW-specific fix guides, mix version history, Aural Preview, Visualizations, and the 30-second Fix My Mix preview. Paid tiers kick in for the full-length Fix My Mix mastered downloads.

When does Stripe payment go live?

Soon. Until then, the Fix My Mix CTA shows a "payment integration in progress" message for non-owner accounts.

Privacy & Data

Do you store my audio files?

Yes — encrypted, in your private Firebase Storage path. Free-tier uploads auto-expire after 24 hours; paid-tier uploads persist as long as your account is active. Analysis results (scores, metrics, fix suggestions, conflict timestamps) are stored in Firestore and are persistent.

Who can see my analyses?

Only you by default. You can choose to share specific analysis scorecards via your public profile. Your full analysis history is always private to your account.

Can I delete my data?

Yes. You can delete individual analyses from your history, or contact support@unveilmusic.app to delete your entire account and all associated data including stored audio.