Vocals and pitch editing

Choose sampled or generated vocals #

Convert to Vocal Notes analyzes existing audio into sampled note segments. DiffSinger and UTAU generate singing from notes and lyrics. A sampled vocal correction does not require a singing voicebank, while generated singing does.

Automatic Per-Note Audio results and their playback are available in Free when the add-on is enabled. Manual per-note audio authoring requires Pro. DiffSinger and UTAU require Pro Vocal Synth, which is beta, plus usable voice content and its rendering dependencies.

Create sampled vocal notes #

Start with an arranger audio clip and choose Convert to Vocal Notes. You can also choose Record Per-Note Audio for a new take. The analyzer creates notes for detected pitched segments and attaches the source audio to them.

Check the phrase boundaries, pitches, and source offsets after analysis. Unvoiced consonants, breath, noise, and dense accompaniment may not form useful pitched notes. If a recording yields no pitches, check the fallback audio clip rather than assuming the take was discarded.

• Loop the sample attached to one note

Open the pitch editor #

Use Open Pitch Editor on an eligible sampled or vocal note. A sample note can also offer Enable Pitch Editor or Disable Pitch Editor. The editor is tied to the selected note and its authoring entitlement.

Opening a sampled note alone does not prepare fresh pitch analysis in this source-reviewed version. If the note has no existing pitch data, do not expect a detected pitch trace or audible resynthesis merely from opening the editor. This workflow has not passed a native walkthrough for this draft.

Its tools are Select, Pitch, Time, Amp, Split, Heal, Fmt, and Dia. When existing pitch data is available, start with Pitch to inspect the pitch curve, select the part you intend to correct, and make a small change. Time edits timing; Amp edits level; Split and Heal change segment boundaries. Fmt controls formant editing. Dia edits the transition diamond at a detected segment boundary.

Corr is correction strength, Ctr is pitch-center strength, and Dft is drift correction, each from 0 to 100. Mod sets modulation depth from 0 to 200. Fmt shifts formants from -2400 to +2400 cents. The initial toolbar values are Corr 100, Ctr 0, Dft 0, Mod 100, and Fmt 0; loading a note shows its stored settings.

The visible A/B button currently changes its toggle state without connecting an original-versus-edited playback comparison. Do not use it as evidence that you are hearing the untouched source. Use an unchanged source clip for comparison.

Generate singing with DiffSinger or UTAU #

Select a DiffSinger voice or UTAU voicebank for the track, draw the phrase as notes, and enter lyrics. Edit Lyrics edits a note lyric. Selected notes can use Fill Lyrics for a phrase. Choose a language supported by the voice and listen to a short test phrase before preparing a full song.

Voicebank and language lists describe available content. They do not mean that every voice can sing every listed language. Rendering runs outside the realtime note trigger path and can take time. Check the rendered phrase after lyric or timing changes.

Choose a discovered UTAU bank in the Instrument Browser, or drag a recognized local voicebank folder onto the editor to copy it into DonutStudio's UTAU voices folder and assign it to the receiving track. Discovery requires a valid oto.ini. To choose an external resampler, open Settings > System > UTAU and choose Browse.... If the configured path does not name an existing file, rendering uses the built-in WORLD path. Check a short phrase for bank alias, subbank and pronunciation compatibility before preparing a full song.

While editing a note lyric, Tab commits it and moves to the next note. Shift+Tab commits it and moves to the previous note. Escape cancels the edit.

Vocal note properties #

Choose Note Properties... in the vocal note menu for per-note settings for the selected singing engine. Language overrides the track language. DiffSinger has Phonemes, Gender, and Velocity. Phonemes editing requires a single selected note; Gender ranges from -1 to +1 and Velocity from 0.5 to 2.0. Here Velocity controls articulation speed rather than ordinary MIDI loudness.

Note Properties > Language offers Auto and English, Japanese, Mandarin Chinese, Korean, Spanish, French, German, Portuguese, Italian and Russian routes. A note override takes precedence over track language. Auto estimates English, Japanese, Korean, Chinese or Russian from the writing system. Select another Latin-script language explicitly. Built-in Mandarin accepts Pinyin and a limited Hanzi map; unmapped Hanzi can become a generic "a" phoneme. Check that the selected DiffSinger voice supports the language, inspect phonemes and render a short phrase.

UTAU has Flags, Voice color, and preutterance and overlap adjustments in milliseconds. Flags depend on the resampler and voicebank. Voice color lists subbanks found in the selected voicebank. Keep default values until a short phrase renders correctly, then change one setting at a time.

Apply commits the property changes and requests updated vocal rendering. Cancel leaves the dialog changes unapplied. Pro Vocal Synth is required at the editing boundary.

Fill lyrics for selected notes #

In the Piano Roll, select the notes for a phrase and choose Fill Lyrics from the vocal toolbar. Notes are processed in start-time order. Enter lyric items and choose Space, Line, or Char to split them. G2P previews phonemes using each note's language override and the track language. Apply fills matching notes; extra items are ignored, and notes without an item keep their lyrics. Changing a DiffSinger or UTAU lyric regenerates its phonemes and requests an updated vocal render.

Edit generated DiffSinger pitch #

For DiffSinger tracks with Vocal Synth authoring access, the Piano Roll overlays the rendered phrase's generated F0 curve. Draw Pitch paints pitch values; dragging an existing curve point adjusts it, and Shift-drag adjusts the note's range with its ends held in place. Right-click with Draw Pitch for quantize, straighten, smooth, and reset options. Erase Pitch clears user overrides only in unfrozen frames. Freeze Pitch toggles a short region around the click; dragging an existing point also freezes the edited frames. These controls are separate from Open Pitch Editor, which edits a selected sampled note. Playback refresh after existing-curve edits and pitch-menu actions is unverified in this draft.

Adjust phoneme timing in the Piano Roll #

Draft: phoneme boundary editing has source review; native gesture and rendering behavior remain unverified.

At 80 pixels per beat or more, the Piano Roll draws phoneme boundaries on vocal notes with at least two phonemes and enough screen width; labels appear when a segment has room for one. Drag a boundary to redistribute time between neighboring phonemes. Neighbors limit boundary movement. The exact minimum segment length after timing redistribution remains unverified. Releasing the drag requests an updated vocal render. Without stored phoneme timings, positions use a consonant/vowel estimate. Timing stays within the note; there is no separate pre-onset region.

Shape DiffSinger expression over time #

On a DiffSinger track, the Automation Panel offers Gender, Breathiness, Tension, Energy, Expressiveness, Velocity, Tone Shift, Speaker Morph, and per-speaker lanes when the selected voice has multiple speakers. Click to add a point, drag to move it, or right-click to remove it. These expression curves are used when rendering a phrase. Dynamics is a separate level envelope. The Device Chain also has Volume, Gender, Breathiness, Tension, Energy, Expressiveness, and Tone controls. Volume sets track level. Gender, Breathiness, Tension, and Tone provide track defaults. Rendering from the Energy and Expressiveness knob defaults, and immediate audio refresh after knob changes, are unverified in this draft. Use the automation lanes for changes over time. Per-speaker automation curves control the speaker blend during phrase rendering, separately from the track expression controls.