MusicGenTrainingSample<T>

Models & Types · AiDotNet.Audio.MusicGen

One MusicGen training example: a music clip, its text description and optional metadata.

Constructors

Constructor Summary
MusicGenTrainingSample()

Properties

Property Summary
IReadOnlyDictionary<string, string> Metadata { get; set; } Metadata merged into the description by condition merging (§3.1): any of key, bpm, genre, moods, instrument and keywords (lists comma-separated).
Tensor<T> Audio { get; set; } The clip as a waveform at the model’s sample rate (or give Codes).
Tensor<T> Codes { get; set; } The clip’s EnCodec codes [frames, codebooks], used instead of encoding Audio.
Tensor<T> Melody { get; set; } The melody to condition on, for a melody model; null uses the clip itself (Audio, as the paper trains), decoding Codes when only they are given.
string Description { get; set; } The text description; null trains the step unconditionally.