AVH Studios — Tutorial

1 Getting Started

Creating an account

When you first open the app, you'll see a small bar in the top-right corner with Log In and Sign Up Free buttons. Click Sign Up Free, fill in your name, email, and a password (at least 6 characters), and click Create Account. This is free — you don't need a plan yet just to explore the interface.

Choosing a plan

Right below the header you'll find the Plans & Pricing section with three options. The plans do genuinely different jobs, so it's worth understanding the split before you pick:

• Free — $0/month. A social account: a public profile, the community’s New Music feed, and the marketplace where you can buy songs & beats. Free has no studio access — humanizing, instrumental re-performance, re-performance and voice all need Starter or Pro.
• Starter — $14.99/month. Direct AI Vocal Humanization. The app humanizes the AI vocal itself — nudging pitch jitter, shimmer, HNR and vibrato into measured human-normal ranges and running the automatic vocal chain — and rebuilds the instrumental with real instruments, so the final mix passes AI detection for the whole song. Includes 5 file downloads per month. Advanced Settings are visible so you can see exactly what Pro unlocks, but the controls are locked.
• Pro — $24.99/month. Everything in Starter, plus your own voice: a Personal Trained Voice (a private model of your real voice that sings the lead) and the “Nuclear Option” sing-along (sing along live) — blended into the AI song — plus unlimited downloads and full hands-on access to every Advanced Setting.

Click Choose Starter or Choose Pro to activate your plan. You can upgrade at any time — your current plan is marked clearly on its pricing card, and the page tells you how many downloads you have left each month.

Why three tiers? Free is a social account — browse the community and buy songs & beats. The studio plans make the whole song pass: Direct AI Vocal Humanization rewrites the vocals and the instrumental is rebuilt with real instruments. Pro is for when you want your own voice in the song — the Personal Trained Voice or a live sing-along — blended into the AI track.

Free accounts can browse the community and buy songs & beats. The studios — humanizing, instrumental re-performance, re-performance and voice — need Starter or Pro. You can look around the studio first; a plan is required to process and export a song.

2 The 8-Step Workflow: Starter vs Pro

Every song follows the same eight-step path. The early steps are identical on both plans; where they diverge is Step 3 (which humanization path you choose: Direct AI Vocal Humanization on Starter, or add your own voice on Pro). Here is the whole journey at a glance:

1. Step 1 — Upload the AI Song. Load the full AI mix (MP3 or WAV). The copyright scan runs instantly and the tempo is detected automatically.
2. Step 2 — Stem Split. One click separates the song into a clean lead vocal, optional backing vocals, and an untouched instrumental. The lead vocal becomes the app's “AI Vocal”; the instrumental becomes the “Backing Track”.
3. Step 3 — Humanize the Vocals. Choose your path: Direct AI Vocal Humanization (Starter), or Sing Along & Record / Personal Trained Voice (Pro).
4. Step 4 — Rebuild the Instrumental (automatic). We re-perform the AI bed with real instruments so the whole song passes, not just the vocal. On both Starter and Pro this runs automatically as part of the whole-song re-production.
5. Step 5 — Automatic Processing. The app time-aligns, EQ-shapes, ducks, de-esses, preserves breaths, adds saturation and micro-timing drift, and gently corrects pitch — building one finished master.
6. Step 6 — Advanced Settings (optional, Pro). Fine-tune any of it by hand — manual EQ, de-ess, transient, drift and pitch controls, plus A/B against the uncorrected take.
7. Step 7 — Final Blend & Compare. Tune the faithful blend on a real mastering console, then A/B the result against the original AI song.
8. Step 8 — Export. Render the full mix offline at full quality, download the WAV, and verify the final master against the AI detectors.

Now the detail on each plan:

Starter — Direct AI Vocal Humanization + a rebuilt instrumental

On Starter you upload your AI song and the app humanizes the AI vocal directly. It measures the vocal's actual pitch jitter, shimmer, HNR and vibrato-band energy and nudges each one toward the human-normal range, then runs the automatic vocal chain (de-ess, breath preservation, consonant emphasis, micro-timing drift, pitch shaping). At the same time it rebuilds the instrumental with real instruments — re-performing the AI bed so it keeps the same musical content but gets a brand-new waveform with none of the generator's spectral fingerprint. When you export, the final mix shows you that it passes AI detection for the whole song, not just the vocal. That's the whole Starter job, and it's a genuine one: the vocal is the part detectors latch onto hardest, and the rebuilt bed removes the rest of the tell-tale signature.

Where this lives in the app: Direct AI Vocal Humanization is Step 3 of the Re-Production Studio — the standard path. When you reach Step 3 you simply choose Direct AI Vocal Humanization (or Sing Along & Record for the Pro Nuclear Option) and pick up the controls there. It is not an Advanced Setting.

Inside Direct AI Vocal Humanization

These four tools adjust the AI vocal toward measured human-normal ranges, so Starter users get a genuinely humanized vocal, and the instrumental is rebuilt with real instruments alongside it. No separate real-vocal recording is required.

Pitch Jitter. Cycle-to-cycle F0 (pitch) micro-variation. Real voices naturally sit around 0.3–1.5% jitter; many AI vocals are unnaturally stable (too low) or, in some models, oddly unstable (too high). This control nudges jitter toward the natural range in either direction.

Vibrato / F0 Naturalness. Shapes the slow, musical pitch-contour movement (vibrato-band energy, roughly 3.5–8.5 Hz) so it resembles natural human pitch drift rather than a perfectly flat or robotically regular contour.

Shimmer. Cycle-to-cycle amplitude micro-variation (loudness flicker). Human voices typically measure 2–5% shimmer. AI vocals often run higher than this, sounding subtly buzzy or unstable, so this control usually reduces shimmer toward the natural range.

HNR / Breath. Harmonics-to-Noise Ratio — controls the breathy/noisy component blended into the harmonic signal. Many AI vocals have unnaturally clean harmonics (too high an HNR), so this control typically adds a touch of shaped breath noise to bring HNR down into the natural 13–25 dB range.

Auto Mode. On Starter and free plans, all four controls run in a smart automatic mode: the app analyzes your loaded AI vocal's actual jitter, shimmer, HNR, and vibrato-band energy and computes the correction each metric needs, without you touching a slider. Pro users can switch any control out of auto to fine-tune it manually in real time.

Objective Acoustic Humanization Score. Once an AI vocal is loaded, a score panel shows a composite 0–100 rating (weighted across jitter, shimmer, HNR, and vibrato against measured human-normal reference ranges), both before and after humanization is applied, so you can see the actual measured improvement. Timing micro-variation is analyzed and displayed but intentionally excluded from the composite score, since acceptable timing variation is much more style- and genre-dependent than the other three metrics. Sub-scores are reported honestly — including when a metric doesn't fully reach the human-normal range — rather than only showing favorable numbers.

Pro — add your Personal Trained Voice and the “Nuclear Option”

Every plan — Starter included — rebuilds the instrumental with real instruments (Step 4) so the whole song passes, not just the vocal. Pro picks up from there and adds two extra moves:

• Your Personal Trained Voice. Point the app at a private model of your real voice — captured, not generated — and it sings the lead, pitch-matched to the original. Because the voice is genuinely human, the vocal passes detection at the source. (Full walkthrough in section 3.)
• The “Nuclear Option” — Sing Along & Record. Detectors change over time. The single most certain way to never be flagged in the future is to add a real human voice. Record yourself singing the same part (or upload a take) and the app time-aligns it to the AI vocal, gently pitch-matches it so it sits in the same key and melody, and blends it in underneath as human texture — pitched to sound identical to the AI song you uploaded while humanizing it. If you'd rather your own voice replace the AI vocal, that's the same feature: your real take, matched to the song's pitch and timing, becomes the humanized lead. This is a Pro feature.

How to think about it: Starter makes the whole song — humanized vocals and a rebuilt instrumental — read as human. Pro goes further and puts your own real voice in the song, the most future-proof answer of all. If a detector update ever starts flagging your track again, the Nuclear Option is the answer that doesn't depend on any model's current blind spots.

What the automatic pipeline does

Once your files are loaded the app starts working on its own — there's nothing to click. Watch the checklist fill in:

• Time-aligning real vocal to AI vocal — cross-correlation lines up your recording with the AI vocal automatically.
• Analyzing sibilance & breath dynamics — scans for harsh "S" sounds and quiet breaths.
• Analyzing & gently correcting pitch — nudges your notes closer to the AI vocal's notes without erasing your natural pitch character.
• Applying micro-timing drift & consonant emphasis — adds human-like timing variation and punches up consonant attacks.
• Building the EQ / duck / de-ess / saturation chain — shapes your vocal to sit underneath the AI vocal as texture.
• Removing AI fingerprints — the finished master is passed through our built-in fingerprint cleaner (on by default). It strips the AI generator's fingerprints across six layers — spectral, timing, dynamics, stereo, noise-floor and metadata — using changes below the threshold of human hearing, so the mix sounds the same but no longer carries the tell-tale signatures Suno/Udio leave behind. No third-party cleaner needed. The same cleaner is applied to the instrumental too — on every plan the AI bed is rebuilt with real instruments and de-fingerprinted.

Below the checklist, two live visualizations update automatically: a waveform view and a pitch contour view. You don't need to touch anything here — it's a live look under the hood, not a control panel.

Preview the blend

Use the transport controls (▶ Play, ⏸ Pause, progress bar) to listen to the finished blend. Each track — AI Vocal, Real Vocal, Backing Track — has its own strip with Solo, Mute, and a volume slider. This is exactly what will be rendered when you export, and it's where the final-mix detector readout shows you whether the result passes — the whole song (humanized vocals and a rebuilt instrumental) on every plan.

Export

Click Render & Export WAV. The app renders the full mix offline at full quality and gives you a download link. Starter plan exports use one of your 5 monthly downloads (shown just above the export button); Pro plan downloads are unlimited. If you try to export without an active plan, or after using up your monthly Starter downloads, you'll see a message explaining exactly what's needed.

3 Your Personal Trained Voice Pro feature

This is the part that makes AVH Studios one of a kind. Every other “humanizer” tries to smooth the AI vocal so it stops reading as synthetic. We do something different: we put your real voice in the song.

What it is

A Personal Trained Voice is a private model of your own voice, built from a few short passages you read aloud — no singing required. You read the prompts, the app stitches the clips into one clean training sample, and it trains a model that lives on your account and is never shared.

When you humanize a song, that model sings the lead. Because the voice in the output is captured from a real human throat — not generated — the vocal is genuinely human, not a synthetic imitation. The AI fingerprint is gone at the source, not painted over.

How to build it (guided voice capture)

Open the guided capture from the Humanize step and follow five short prompts:

1. Mic check. Confirms your microphone is working and picks a clean input level.
2. Range test. You hum or sing a few notes so the app can measure your natural vocal range.
3. Read. You read a few short passages aloud in your normal speaking voice — this captures your tone and character.
4. Vowels. You hold a few vowel sounds so the model learns your sustained tone.
5. Train. The app stitches everything into one training sample and builds your voice model — usually under a minute.

Train once, reuse forever. Your Personal Trained Voice is saved to your account. Build it once and it sings the lead on every song you humanize from then on — no re-recording.

Why it is the strongest move

• Your voice, your identity. The output sounds like you — your tone and character — instead of a generic AI timbre.
• Detector-proof by design. Detectors flag synthetic artifacts. A real human voice has none to flag, so it stays safe as the models evolve.
• Private and yours. The model is stored on your account and never shared.

Pair it with the rebuilt instrumental in Step 4 and you get something no detector can argue with: a complete song where the voice is real and the bed is clean — the whole track humanized, not just the vocal.

4 Advanced Settings Reference Pro feature

Starter and no-plan users can see every control in the Advanced Settings panel (so you know exactly what Pro unlocks), but the controls themselves are locked until you upgrade. Everything already has a sensible default applied automatically — you only need this if you want to fine-tune the sound yourself.

Timing

Fine-tune timing offset. The app auto-aligns your vocal, but you can nudge it earlier or later by hand using the −100ms/−10ms/+10ms/+100ms buttons or by typing an exact millisecond value. The offset display shows what the automatic analysis originally detected.

Tone Shaping

Real Vocal Level (Duck). How far below the AI vocal your real vocal sits in volume, in dB. Recommended range is 12–18 dB below — enough to add texture without competing for attention.

High-Pass Filter. Cuts low-frequency rumble and handling noise from your recording.

Low-Mid Cut (Mud). A targeted dip around the low-mids (default 300 Hz) to stop your vocal from making the mix sound muddy when stacked underneath the AI vocal.

Low-Pass "Body" Filter. The key move in the whole chain — rolls off the higher frequencies of your real vocal so it acts as warm "body" and texture rather than a second, competing lead vocal.

Presence Tuck (optional). A narrow cut around 3 kHz if your real vocal is still poking through too much on top of the AI vocal.

Warmth Shelf (optional). A low-shelf boost around 200 Hz if the blended layer feels too thin.

Saturation / Warmth Drive. Soft, analog-style harmonic saturation. Higher values add more grit and warmth; 0% keeps it clean.

Realism Features

De-esser. Automatically detects and tames harsh "S"/"SH" sibilance frame-by-frame. Adjust the amount of reduction and target frequency, or disable it entirely.

Breath Preservation. Naturally-recorded breaths in your take are lifted slightly above the duck level so they remain audible instead of getting buried. This preserves your real breaths — it never synthesizes fake ones.

Consonant Emphasis. Automatically detects the sharp attack of consonants like P, B, T, and K and gives them a small boost, since real human voices tend to attack these harder than AI-generated vocals.

Micro-Timing Drift. Adds an extremely subtle, slowly-modulating delay so your vocal isn't perfectly locked to the beat at the sample level — mimicking the natural timing instability of a real human performance.

Pitch Correction ("Melodyne-Lite")

Pitch Correction Strength. How strongly your vocal's pitch is pulled toward the AI vocal's pitch. Defaults to a gentle 40% on purpose — full "snap to pitch" correction can ironically make a real voice sound more robotic and mechanically consistent, not less. Push it higher only if your take has genuinely distracting pitch issues.

Max Correction Range. Caps how many semitones any single note can be pulled, which limits artifacts on notes that are very far off.

Use pitch-corrected real vocal. Uncheck this to instantly A/B your original, uncorrected recording against the corrected version — useful for confirming the correction is actually helping.

Utility Buttons

Reset to Recommended Defaults. Returns every Advanced Setting to the app's tested default values in one click.

Re-run Automatic Processing. After changing any Advanced Setting, click this to reprocess your vocal with the new values. The pipeline checklist and visualizations update just like the first automatic pass.

This exact reference is also available as a collapsible dropdown directly inside the Advanced Settings panel in the app, so you can check it while you work.

5 The AI Detectors We Test Against

We score your final master against a real, published AI-music detector — the same model that runs inside the app — and we are transparent about every detector that matters: how each one works, and exactly how to verify your finished master on the SubmitHub AI Song Checker.

The Export step in the app also scores your rendered master against the in-app detector and shows a SubmitHub-style prediction, so you can catch a problem before you upload.

6 AI Re-Performance Collaboration Pro feature

This is the first-of-its-kind part of AVH Studios: a shared studio where two creators re-perform one AI song together, each in their own real voice. It’s built around a simple promise — nothing happens to your voice, or your money, without your explicit consent. It’s a Pro feature, designed for exactly two collaborators, and it works asynchronously, so you don’t have to be online at the same time.

Where to find it. Open Collab from the navigation bar. Create a session from a track, share the invite link with your one partner, and they join with their own account. Everything below is then shared between the two of you.

The two-key idea

Every collaboration here runs on two keys: one held by each of you. A voice is never used, a co-owned voice is never created, a use is never granted, and a split is never locked, unless both keys are turned. Either of you can revoke your key at any time, and a voice stays valid only while both of its parent voices stay cleared.

1. Voice Passports — bring your own voice

Each of you shares a Voice Passport: one of your Personal Trained Voices, offered into this song. Your partner must turn their key to clear it for the mix (and you turn yours for theirs). A passport can be revoked, and it’s scoped to the session — sharing a voice here never exposes it anywhere else.

2. Voice Blend — fuse two voices into one

Once both passports are cleared, you can fuse them into a Voice Blend at any ratio — for example 65% you, 35% your partner. The blend becomes a first-class voice you can cast to any section, so a single line can be sung in a voice that is literally a mix of both people. A blend stays valid only while both of its parent voices stay cleared.

3. Co-owned voice — one voice you both own

You can save a blend as a co-owned voice: a single, jointly-held record that travels across every song you make together. Creating it needs both owners’ approval (Gate A). Using it in any new song needs both owners’ approval again, for that specific use (Gate B) — so a co-owned voice can never be quietly dropped into a project one of you doesn’t know about.

4. Section casting & takes — pass the mic

The shared track is automatically split into verses and choruses (chorus-first). Cast any section to either voice, to a blend, or to a co-owned voice, then record or upload your take for that section. A take locks in only when both of you approve it. You can also leave arrangement notes pinned to any section, so “what to change” lives inside the song.

5. Split agreement — agree the money up front

Before the final production ships, agree how its earnings are split. One of you proposes the percentages, both of you consent, and the agreement is locked. Once locked, it can only ever change with a fresh proposal that both of you accept again — a tamper-proof handshake on the money.

6. Earnings ledger — every dollar, auto-allocated

The earnings ledger reads your locked split. Record an earning (a royalty statement, a sync fee, a sale) and it is instantly allocated to each partner by the locked percentages, with a per-partner balance and payout requests. Each earning snapshots the split at the moment it was recorded, so a later re-lock never rewrites history.

7. Consent nudges — nothing stalls

Whenever your partner needs your key — a voice shared, a co-owned use requested, a split proposed, a take uploaded — a nudge lands in your in-app notification centre (the 🔔 bell in the Collab header), and by email when an email provider is configured. So the collaboration keeps moving even across time zones.

Putting it together. Share passports → clear each other → (optionally) blend or save a co-owned voice → cast sections → record and approve takes → assemble the shared vocal → lock a split → let the ledger allocate every dollar. That’s the full AI Re-Performance Collaboration — the only one of its kind.

7 Tips for Best Results

• A rough, natural take beats a "perfect" one. Human imperfection is the signal, not the noise. Don't over-rehearse your recording.
• Keep your backing track in sync. Make sure it's the same edit/length as your AI vocal stem — the app aligns your real vocal to the AI vocal, but does not independently re-align the backing track.
• Use a decent microphone if you can, but don't stress about it. Even a basic USB mic or phone recording works — the processing chain is built around ducking, filtering, and blending your vocal underneath the AI vocal as texture.
• Try the A/B toggle before exporting. Uncheck "Use pitch-corrected real vocal" briefly to confirm the correction is genuinely helping and not overcorrecting.
• No detector bypass is guaranteed. This tool adds genuine human vocal characteristics back into an AI-generated performance. It is not a guaranteed workaround for any specific AI-detection system, and detection systems are updated over time.

8 Plans at a Glance

That's the full tool, start to finish. If you get stuck, the in-app hints under each section header describe what that step does — and the Advanced Settings tutorial dropdown inside the app always has this same reference one click away.

← Back to the app