UdioModel<T>
Models & Types · AiDotNet.Diffusion.Audio
Udio/Suno architecture model for full-song music generation with structural awareness.
Inheritance: DiffusionModelBase<T> → LatentDiffusionModelBase<T> → AudioDiffusionModelBase<T> → UdioModel<T>
For Beginners
Udio/Suno generates complete songs from text prompts.
How Udio/Suno works:
- Text prompt is encoded with genre-aware text encoder (1024-dim features)
- Structural conditioning identifies verse/chorus/bridge sections
- DiT backbone generates audio latents with musical structure awareness
- Flow matching provides efficient denoising in fewer steps
- High-fidelity audio VAE decodes latents to 44.1 kHz stereo audio
- Post-processing applies mastering for production-quality output
Key characteristics:
- Full-song generation (2-5 minutes) with musical structure
- Verse/chorus/bridge structural conditioning
- Genre-aware generation across any musical style
- High-fidelity 44.1 kHz stereo output
- Optional vocal + instrumental separation
- Flow matching for efficient sampling
When to use Udio/Suno:
- Complete song generation from text descriptions
- Music production prototyping and ideation
- Background music generation for content
- Genre-specific music creation
Limitations:
- Commercial API service (not open-source)
- Generation time increases with song duration
- Limited fine-grained control over arrangement
- Vocal quality varies by genre and language
How It Works
This model represents the Udio/Suno class of full-song music generation systems using a DiT backbone with structural conditioning for verse/chorus/bridge awareness. It generates complete songs at 44.1 kHz stereo with full musical structure from text prompts.
Architecture components:
- DiT backbone (2048 hidden, 32 layers, 16 heads) for music generation
- Structural conditioning for verse/chorus/bridge awareness
- Genre-aware text encoder (1024-dim) for style control
- High-fidelity audio VAE with 128 mel channels
- Flow matching scheduler for efficient inference
- Stereo output at 44.1 kHz sample rate
Technical specifications:
- Architecture: DiT with structural conditioning
- Hidden dimension: 2048
- Transformer layers: 32
- Attention heads: 16
- Context dimension: 1024
- Latent channels: 64
- Sample rate: 44,100 Hz (stereo)
- Default duration: 180 seconds (3 minutes)
- Maximum duration: 300 seconds (5 minutes)
- Mel channels: 128
- Scheduler: Flow matching
Reference: Conceptual representation of Udio/Suno-class music generation, 2024
Example
var udio = new UdioModel<float>();
var song = udio.GenerateMusic(
prompt: "An upbeat pop song with catchy chorus about summer",
durationSeconds: 120.0,
numInferenceSteps: 100,
guidanceScale: 5.0);
Constructors
| Constructor | Summary |
|---|---|
UdioModel(NeuralNetworkArchitecture<T> architecture, DiffusionModelOptions<T> options, INoiseScheduler<T> scheduler, DiTNoisePredictor<T> dit, AudioVAE<T> audioVAE, IConditioningModule<T> conditioner, int? seed) |
Properties
| Property | Summary |
|---|---|
IConditioningModule<T> Conditioner { get; } |
|
INoisePredictor<T> NoisePredictor { get; } |
|
IVAEModel<T> VAE { get; } |
|
bool SupportsAudioToAudio { get; } |
|
bool SupportsTextToAudio { get; } |
|
bool SupportsTextToMusic { get; } |
|
bool SupportsTextToSpeech { get; } |
|
int LatentChannels { get; } |
Methods
| Method | Summary |
|---|---|
ModelMetadata<T> GetModelMetadata() |