vocal-route
/vocal-route
Dobiera i prowadzi sposób, w jaki piosenka dostaje główny wokal, gdy autor nie będzie śpiewać, i sprawdza go na jednym refrenie.
- Autor
- NatikChhabra ↗
- Repozytorium
- claude-arsenal ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 13.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 2
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 13.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Dobiera i prowadzi sposób, w jaki piosenka dostaje główny wokal, gdy autor nie będzie śpiewać, i sprawdza go na jednym refrenie.
Wyzwalacze z opisu
Oryginalny opis (EN)
Choose and run the way a song gets its lead vocal when the writer (a guitarist who can't sing) won't be the final sung voice: guide plus pitch fix plus voice conversion, vocal synthesis, generation from his guitar, rap, or a real singer, picked by language, melody fidelity and release terms, proved on one chorus with blind listeners before the full song. Use whenever a song needs vocals, for "make it sing", "AI vocals", "voice for my song", or a vocal that sounds fake.
Jak działa
- 0. Inputs
- 1. Pick the route (one table lookup, then a reason)
- 2. Range and key
- 3. Chorus first, blind
- 4. Full vocal
- 5. Credits and rights (before the song leaves the machine)
- Rules
- Report
Podgląd SKILL.md
---
name: vocal-route
description: >-
Choose and run the way a song gets its lead vocal when the writer (a guitarist who can't sing) won't be the final sung
voice: guide plus pitch fix plus voice conversion, vocal synthesis, generation from his guitar, rap, or a real singer,
picked by language, melody fidelity and release terms, proved on one chorus with blind listeners before the full song.
Use whenever a song needs vocals, for "make it sing", "AI vocals", "voice for my song", or a vocal that sounds fake.
---
# vocal-route
status: draft (full workflow; promote to beta after the first real song)
The owner loves singing and can't sing well. The vocal must sound great AND keep the song his: his melody, his words,
his phrasing. Every tool's real abilities, prices and terms: [references/routes.md](references/routes.md). Sources:
research/03_vocal_tools.md.
## 0. Inputs
SONG.md (lyrics passing lyric_lint, melody guide, language, release intent) and the guitar take if one exists. No
SONG.md = run skill song-from-idea first.
## 1. Pick the route (one table lookup, then a reason)
| Situation | Default route |
|---|---|
| English, he wants his exact melody | B (Synthesizer V 2 Pro Voice-to-MIDI from his guide), or A |
| Hindi, Hinglish or Punjabi lyrics | A (his guide, pitch-fixed, converted to a voice he owns or licensed) or his own vocal pitch-fixed; no commercial singing synth sings Hindi (2 Nov 2026) |
| He wants his own real voice on the record, nothing generated or converted | F (his own natural voice: melody written into his measured range, talk-sung verses, comped and lightly tuned) |
| A fast demo to hear the song | C (Suno cover of his guide, or ACE-Step locally) |
| The style allows rap or spoken word | D (his own voice) |
| The song is the centrepiece of a release and a willing singer exists | E |
| Monetised release | never a free-plan generator (Suno free: Suno owns it; Eleven free: non-commercial); never Udio (no downloads) |
Write the pick and the reason in SONG.md's decisions log. Unclear between two = run both on the chorus in step 3.
## 2. Range and key
- Measure a real voice, never guess it: `uv run --with librosa --with soundfile python ${CLAUDE_SKILL_DIR}/../../tools/voice_range.py <take.wav> --mode sung --melody <low> <high>`
(a 30 s sung scale or hummed chorus; `--mode speech` on any old spoken take gives a first estimate). It prints the
comfortable range, the shift and the capo, and flags clipped takes.
- Find the chosen voice's comfortable range (the synth voicebank's stated range, the converted voice's model, the
singer's own). Transpose the song (capo for guitar) so the chorus peak sits 1-2 semitones under the top of it.
- Update Key, Capo and Range in SONG.md before any guide is recorded: a re-key later means a new guide.
## 3. Chorus first, blind
1. Make ONLY the first chorus with the chosen route (two routes if step 1 was unclear).
2. Clean it: timing to the grid where it should sit, breaths kept, no clicks, de-ess.
3. Measure: `python ${CLAUDE_SKILL_DIR}/../../tools/loudness.py <chorus mix> --vocal <vocal stem> --inst <inst stem>`
(the phone-speaker check: the vocal must not sit under the band-limited instrumental in more than 10% of sung
windows).
4. Blind listen. Claude cannot hear audio, so the judges are people: the owner plus at least 2 listeners who were not
told how it was made. Ask them only: "Would you replay it? Does any part sound fake or robotic, where?" One "sounds
AI" with a timestamp = fix that phrase (breaths, timing, vibrato, consonants) and re-test with a new listener.
A text-only fresh agent can still read the lyric sheet for mis-stressed words before the listen.
5. Pass = every listener would replay it and nobody flags a fake spot. Then the full song.
## 4. Full vocal
- Same route, same settings and voice for every section (a voice change mid-song is a bug, a feature only when it is a
second character). Doubles and harmonies from the same voice, panned and lower.
- Comp, tune, time, de-ess; keep 2-3 natural breaths per phrase group. Export the dry lead, doubles, harmonies and
ad-libs as separate stems for skill mix-master.
## 5. Credits and rights (before the song leaves the machine)
Write `songs/<slug>/CREDITS.md`: route, tool and version, plan tier, voice model or voicebank and its licence, who
owns the guide voice, any human singer and their written consent, the disclosure each platform needs (references
"Release rules"). Missing licence = the song stays private.
## Rules
- Only voices he owns or has written permission for. Never clone or imitate a real artist, never prompt with an
artist's name.
- The release intent in SONG.md decides the tool tier. A song made on a free plan cannot be "upgraded" to commercial
later on Suno.
- A vocal that a blind listener flags as AI does not ship, however good the rest is.
## Report
Route and reason, range and key, the chorus result with each listener's words, loudness numbers, open rights items.
Pliki w paczce (2)
- SKILL.md 5 KB
- references/routes.md 10 KB
Wersje
W bazie nie ma innych skilli o nazwie vocal-route.
Migawka bazy z 13.11.2026 · wersja paczki efb35a628f5e (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 90vocal-chain
Ustawia łańcuch przetwarzania nagranego wokalu, np. popowy lub agresywny rapowy, aby wokal dobrze siedział w miksie. - 95local-song
Komponuje i produkuje oryginalną piosenkę ze śpiewem lokalnie na własnym komputerze, potem czyści wokal, masteruje ją i tworzy teledysk karaoke z tekstem. - 100arrange-lyrics
Układa słowa kompozytora na melodii nad jego nagraniem fortepianowym, pokazuje je na partyturze i śpiewa wybranym głosem, a potem sprawdza. - 100mix-master
Miksuje ścieżki wokalu i instrumentów oraz masteruje utwór pod streaming, wideo i grę, mierząc głośność i szczyty. - 90song-from-idea
Prowadzi od pomysłu, postaci lub riffa do gotowego arkusza piosenki: formy, tempa, akordów, melodii i tekstu. - 90mc-vocal-direction
Określa sposób śpiewania wokalu i projektuje harmonie: dykcję spółgłosek, samogłoski, rytm i cztery typy harmonii.