Generating Vocals in Suno v5.5: Building Voice Datasets for Timbre and Style
The "Your Voice" feature in Suno v5.5 lets you create vocals using custom voice datasets. The system breaks down the voice into timbre, intonation, and articulation. Timbre is shaped by the vocal tract's geometry, intonation by pitch variations with vibrato and slides, and articulation by tongue and lip movements. This setup offers flexibility: the timbre model pairs with intonation and articulation for any language.
Voice recognizability hinges on balancing these elements. If the output doesn't sound right, check which component is off.
Timbre Categories
Singing timbre is described using these categories:
- Power (Strong): Resonant voice with a singer's formant around 2.5 kHz, typical in opera.
- Normal (Neutral): Balanced without strain.
- Breathy (Breathing): Airy with breathiness, close to whispered notes.
Tools like Synth-V add style descriptors: Soft, Power, Bright, Dark. Vocaloid emphasizes Breathiness, Dynamics, and Opening.
Preparing Voice Datasets
Tests used 11 datasets (DS) ranging from 6 seconds to 4 minutes. Untuned vocals and varied speech styles. Generations in 9 styles with Weirdness 0–90 and Style/Audio Influence 50–100.
First 7 DS:
- Power (vocals, strong).
- Normal (vocals).
- Theater (speech, dramatic).
- Crying (emotional crying).
- Mixed (combo of 1–4).
- Speech Expressive (poetry with expression).
- Speech Inexpressive (monotone poetry).
Additional 4 DS for precision:
- Breathy (airy, low-end cleaned up).
- RVC DS (Russian/English vocals/speech).
- One Song (single song with power/normal/breathy).
- Song Set (clips from 8 songs).
Datasets with distinct intonation (Crying, RVC) or timbre (Power, Breathy) yield recognizable results. Speech datasets deliver precise rap with emotion and rhythm.
Arranging with Your Own Melody
For arrangements, start with a MIDI sketch: melody, rhythm, and harmony in dry form (no effects). Convert to MP3 and use Cover mode + Your Voice + Lyrics + Style.
Sketch Tips:
- Simple notes: Quarter notes for chords/bass allow variations (eighths, triplets).
- Melody: GM Clarinet or Lead 6 (voice); hummed tune with tuning; Synth-V/Vocaloid.
- Harmony: Stick to basics (avoid Am7/G, Csus2); the system may simplify to major/minor.
Tested on 8 DS: folk song, "Make Me Feel," "Red Hair Girl." High Audio Influence boosts similarity.
Using finished tracks as sketches limits style freedom and quality.
Key Factors for Preserving Timbre
What Matters:
- Pitch range in sketch matches DS; system extends it but morphs if data is sparse.
- Prompt/Style aligns with DS: R&B adds slides that clash with power voices.
- High Audio Influence prioritizes the DS.
Limit lyric variability, tweaking verses to control melody.
Key Takeaways
- Breaking voice into timbre, intonation, and articulation enables flexible synthesis.
- Multiple DS per style: one universal set won't cover every song.
- Simple dry sketches maximize arrangement variations.
- High Audio Influence and matching Style keep it recognizable.
- Speech DS produce pro-level rap with emotion.
— Editorial Team
No comments yet.