Breath and Phrasing
Breath marks decide how a line is understood, and a synthetic voice will not breathe for you.
Related
A breath is punctuation
A line of lyrics has no comma printed on it, yet a singer breathes somewhere and the listener uses that break to parse the sentence. A synthetic voice cuts on notes rather than meaning, so it joins what should break and holds what should close. To judge whether a take sounds human, count the breaths before judging the tone.
Five things to change
Where the breath falls, how long it lasts, how much adjacent notes overlap, how the line ends, and where the consonants land. Position follows both lyrics and melody and usually sits on a syntactic boundary, one breath every six to ten syllables. Length runs from about eighty to two hundred milliseconds: shorter and the breath disappears, longer and the pulse sags. Overlap or separation controls the joins; too much smears the words, too little turns singing into recitation. Endings are the hardest part, because a long note has to release early and a short one has to keep a little air.
Settings
| Parameter | What it does | Common practice | When pushed too far |
|---|---|---|---|
| Breath position | Parses the phrase | On a syntactic boundary, every six to ten syllables | Too frequent reads as gasping, too rare as chanting |
| Breath length | Creates the intake | Eighty to two hundred milliseconds | Past half a beat, the pulse falls apart |
| Note overlap | Controls joins | Twenty to sixty milliseconds | More smears words, less makes them jump |
| Ending decay | Closes the phrase | Finish inside the last beat | Drawn out sounds fake, cut early sounds stiff |
| Consonant onset | Places the attack | Ten to thirty milliseconds ahead of the beat | Too far ahead sounds like rushing |
Endings and consonants
Even with breaths in the right places a line can still fail, and the fault is usually at the end. Synthesis tends to fill the last note for its full value and stop dead, while a singer releases half a beat early and leaves a little air behind. Consonants also need to arrive slightly before the beat so the word feels spoken rather than switched on. Turn these choices into a template and the next line costs almost nothing.
Where it breaks
Breaths packed too closely sound like panic, and breath noise added for realism often sounds pasted on. Fast songs need short breaths and slow ones can afford long breaths, but neither should cover the beat. Backing parts do not need a breath at every phrase, and several lines breathing together sound suspiciously tidy. Spoken narration and singing follow different rules, so do not carry a vocal template into a voice-over.
Listen in the library
All tracks from midicn-lib — inline badge is the license tier (C1 commercial / C2 non-commercial / C3 study only).
Along the same line
Sources
- Richard Miller, The Structure of Singing（Schirmer）与 James McKinney, The Diagnosis and Correction of Vocal Faults 中关于气口与分句的章节。
- 合成歌声系统文档里对呼吸标记、音素时长与连奏参数的说明。
- Vocal pedagogy texts on breath placement and phrasing, with reference to singing-voice system documentation.