HEARTY’S BOOTH

Mac: http://127.0.0.1:7860
Phone: https://8eff1a727970e8aedf.gradio.live

Launch: double-click Desktop/Hearty's Booth.command
Or: cd /Users/gionjubes/Developer/Treblo-Local-Studio && .venv/bin/python manage_studio.py dashboard
Private login: .phone-access.json in this project folder.
Phone access requires the Mac awake; the HTTPS link expires after one week.
This is an installable web dashboard, not a native App Store app.

Load a saved song at the top, then open Tab 3's verse editor:
1. Set the verse start and end.
2. Choose Drake, Brent Faiyaz or Jay-Z in Singer model, or upload/select a voice reference.
3. Choose Change singer, or Replace lyrics and enter the new words.
4. Render, audition, then Continue editing this version for another change.
Vevo2 edits phrases up to 30 seconds; ACE edits sections up to 90 seconds.
Brent Faiyaz and Jay-Z now use their paired retrieval indexes. In Reference
selection and vocal mix, adjust Voice retrieval strength (default 0.75).
Set 0 to compare without retrieval. The current Drake package has no index.
Actual before/after samples: Desktop/Treblo_Complete_Studio/voice_comparisons/.
Retrieval execution is verified; exact artist identity and listening preference
remain unverified. Keep the result you prefer after auditioning it.

Tab 1: Strict instrumental is ON by default; lyrics are ignored and known names use music-only profiles; custom names condition the model directly. Turn it off to generate singing.
Tab 2: 4,160 public tags and User-Named Style; type Jay-Z, Drake, Sade or a custom name.
Tab 3: extension, remix, inpainting, voices and lyric edits.
Tab 4: native waveform zoom/selection/looping, zero-cross snap, crop/replace/undo, references, stems and WAV/FLAC/MP3/OGG/M4A exports.

Outputs: ~/Desktop/Treblo_Complete_Studio
Code: studio.py, precision_editor.py, assets/precision-editor.*, plus the voice workers
Source package: Complete-Studio-Code.zip
Build instructions: UNIVERSAL-BUILD.md
Installation and evidence: infrastructure-log.json and logs/vocal-editing/
Strict instrumental evidence: logs/strict-instrumental/

The music renderer uses its separate GCP project and bounded two-hour CPU sessions.
No Treblo API is required for independent rendering. The optional hosted fallback is unconfigured; see HEARTYS-BOOTH.md for actual feature coverage. Artist labels come from community uploaders; identity,
model-specific licensing, exact wording and Treblo sonic parity are unverified.

STRICT INSTRUMENTAL: Keep the checkbox ON in Tab 1, then enter Drake, Travis Scott
or any artist/custom name in User-Named Style. Lyrics and vocal profile cues are
ignored. Only the cleaned instrumental mix is exported; cleanup failure blocks
saving. Turn the checkbox off to generate singing. Custom names are saved for
reuse and condition the model directly; recognition and sound fidelity are
unverified. Prepared musical profiles remain available as suggestions.

Strict mode suppresses vocals aggressively, but faint vocal bleed remains
possible. Audition the saved result before treating it as completely voice-free.

LENGTH: Tab 1 → Generation length mode → Length range. Minimum 0–270s;
maximum 30–300s; both in 30s steps. The selected target is shown in the receipt.

WAVEFORM: Tab 4 → import or load a saved song → drag to select → zoom / loop /
snap. Drag selection edges or enter sample/second boundaries. Crop, cut/copy,
paste, replace and undo save new versions. Uploaded replacements can keep their
own duration or exactly replace an equal-length selection. Export uses the
current waveform version. Use selection buttons to send it to AI, lyric/voice
editing, or a cropped style/voice reference. Stereo snapping and short fades
reduce seam clicks but cannot guarantee every boundary is click-free.

USER-NAMED STYLES: Tab 2 accepts any artist/style name and saves it for reuse.
Export your saved names from the same tab. All 4,160 public v3 tags were verified
against the current public source. The User-Named Style label is open-ended;
no complete private Treblo artist/model database was exposed or imported.

INDEPENDENT API: http://127.0.0.1:8766/v1
Start without waking compute: .venv/bin/python manage_studio.py api
Separate private key: .api-access.json (do not share it).
Routes, examples and explicit gaps: API-COMPATIBILITY.md.
Preserving source, tracks and installed models: RECOVERY.md.

VOCAL PERFORMANCE: Tab 1 → turn off Strict instrumental → ACE-Step → Vocal performance direction. Choose a preset or emotion, texture, phrasing, vibrato, pitch glides and breath/diction/dynamics. The preview and job receipt record the caption direction; sonic response varies. See VOCAL-DIRECTION.md.

WHOLE-SONG SINGER: Tab 1 → disable Strict instrumental → ACE-Step → enter lyrics and a vocal song prompt → Artist voice matching after generation → Generation singer. Choose Drake, Brent Faiyaz, Jay-Z or Uploaded voice reference (select a clean 1–25 second crop). Generate once; the matched mix, native song, matched vocals and original backing are saved together. Longer songs use overlapping 80-second windows. The method is native generation followed by neural vocal matching; it is independent of Treblo, and exact artist fidelity remains unverified.
