audioJS

Speech prosody editor

Shape how a sentence sounds. Adjust its pitch, emphasis and pace.

Drop a speech recording here
or click to choose, or paste with Ctrl+V
Experimental · one voice · mono output

Choose a short, clean recording to begin.

Nothing is uploaded. Everything runs in this tab with @audio/stretch-wsola and @audio/encode-wav.

Edit delivery, one syllable at a time

Each syllable is a note on a piano roll. Drag it to another pitch, snap a phrase to semitones, straighten wobbly notes, smooth or exaggerate the intonation, or change a phrase's duration. Pinch to zoom into the recording and drag across the plot to select a phrase.

Pitch and timing edits use the WORLD reference speech engine, compiled to WebAssembly and run locally. Harvest tracks pitch continuously across each phrase and glottal cycle marks make it exact per cycle. Phrases are split into syllables at loudness dips, and pitch the tracker is unsure of follows the voice around it. Every edited voiced phrase is rebuilt whole at the new pitch and duration: the waveform engine resamples the recording's own voice cycles and restores their formants, the vocoder engine rebuilds the voice for large changes. Unvoiced audio in a retimed span is stretched with WSOLA and consonant bursts keep their length. WAV export contains audio only, without source tags.

Questions

Can this automatically fix the tone of a question?

Rise tilts the pitch of your selection up or down toward its end. It does not infer sentence meaning. Select the last word, listen and adjust it to match your intended delivery.

Can it tune speech or chant to notes, like Auto-Tune or Melodyne?

Yes. Each syllable becomes a note you can drag; it snaps to the semitones of A4 = 440 Hz. Snap to notes moves the centres of the selected notes onto the nearest semitone, Straighten flattens the pitch within each note, and together at 100% they give flat notes on the grid with natural glides between them. Smooth softens fast wobbles while keeping the melody.

Will it preserve the voice?

For changes within an octave the waveform engine resamples the recording's own voice cycles to the new pitch and restores their formants, so pulse shapes, breath and jitter stay the voice's own. Larger changes use the WORLD vocoder, which rebuilds the voice from its vocal-tract spectrum and breathiness and restores the original loudness contour. Consonants, silence and untouched phrases keep their original samples. Compare with the original before saving.

Why are parts of the curve missing?

Silence and unvoiced consonants do not have a reliable fundamental pitch. Those regions stay unpitched. Very short clips may also have too little audio for detection.

More audio utilities