Make a synthetic voice sing: write notes and syllables, shape breath and phrasing, compare a plain take with a performed one, and export the WAV. Procedural synthesis, no samples and no trained models.
AudioBefore you run it: plays sound, can save files to your device.