Zum Inhalt springen
KI-Atlas

Musik- und Klangerzeugung

Musik oder einen Klangeffekt aus einer Beschreibung erzeugen

Sprache & MusikAnfänger #30
EingabeTextAudio

Der vollständige Artikel liegt auf Englisch vor; Titel und Zusammenfassung sind lokalisiert.

WAS DIESE FÄHIGKEIT BEDEUTET

Takes a text description, optionally with lyrics or a style reference, and outputs audio — a full song, an ambient bed or a sound effect. It must handle melody, harmony, arrangement and timbre at once, adding a layer of musical structure on top of speech synthesis. Unlike text-to-speech the output is not language but rhythmically and tonally organised audio.

Wie sie technisch umgesetzt wird

The mainstream uses an audio latent representation with diffusion or autoregressive generation: audio is first compressed into discrete or continuous tokens, generated under conditioning, and decoded back to a waveform. Structural tags during training teach the model the order of intro, verse and chorus. Lyrics and vocals are produced by separate alignment and synthesis modules and then mixed with the accompaniment into a finished track.

Repräsentative Produkte

3

Beteiligte Organisationen

Typische Verwendungen

  • Background music for short video and podcasts
  • Sound effects for games and apps
  • Scores for ads and promotional films
  • Creative demos and arrangement ideas

Wie sie bewertet wird

FAD
Distribution distance between generated and real music; lower is better
MOS
Mean opinion score for listening quality
Human preference
Pairwise judgement of melodic and arrangement appeal

Grenzen und schwierige Punkte

  • Long-form structure collapses: chorus returns and arrangement layers fail to hold together
  • Lyrics and melody misalign and enunciation blurs, with mispronounced or swallowed syllables
  • Imitating the style of a living artist under copyright carries risk and needs checking before commercial use

Konzepte dahinter