microsoftVibeVoice-Realtime-0.5B

VibeVoice¶

Loads the audited VibeVoice realtime stages without claiming an unverified text-to-waveform loop.

Text to speechVoiceHub-nativevibevoice-ttsParameters: 1BLanguage: enTraining: preprocessedLicense: Checkpoint-specific

Parameter metadata: Exact serialized tensor-element total from VoiceHub's audited native primary checkpoint; a distinct learned-parameter total is not available.

Use this model Checkpoint

Resources

Usage¶

Complete the VoiceHub installation once, then run this repository-authored example. Model pages intentionally contain no package-install command.

This example is maintained against VoiceHub's public API; it is not copied from an upstream demo or package README.

Model-specific path: Loads the audited VibeVoice realtime stages without claiming an unverified text-to-waveform loop.

Inputs and controls: High-level cached-prompt synthesis intentionally fails closed until cache serialization, chunk boundaries, and waveform parity are verified.

from voicehub import AutoModelForTextToSpeech

model = AutoModelForTextToSpeech.from_pretrained(
    'microsoft/VibeVoice-Realtime-0.5B',
    model_type='vibevoice',
    device="cuda",
    lazy_load=True,
)
model.load()
required_stages = (
    "forward_lm",
    "forward_tts_lm",
    "sample_speech_latents",
    "decode_speech_latents",
)
missing = [name for name in required_stages if not hasattr(model.model, name)]
if missing:
    raise RuntimeError(f"Missing audited VibeVoice stage(s): {', '.join(missing)}")
print("High-level synthesis is not verified; available native stages:", required_stages)

Use authorized recordings. Verify hardware needs and pin a revision in production.

Overview¶

vibevoice is a VoiceHub text to speech integration. This page is generated from its registry contract. Open the vibevoice Colab notebook.

Property	Value
Task	Text to speech
Architecture	`vibevoice-tts`
Runtime	`VoiceHub-native`
Languages	`en`
Capabilities	`text-to-speech`, `voice-prompt`, `fine-tuning`, `default-checkpoint-inference-only`, `safetensors`, `voicehub-native`, `native-runtime`, `preprocessed-training`, `verified-low-level-realtime-stages`, `high-level-generation-fails-closed`
Reusable components	—
Normalized output	`TTSOutput`

Language support¶

Supported language abbreviations

en

Paper and GitHub¶

Paper: VibeVoice Technical Report
Upstream GitHub: VibeVoice
VoiceHub source: VoiceHub model implementation

Configuration¶

Load configuration without constructing the model:

from voicehub import AutoConfig

config = AutoConfig.for_model('vibevoice')
print(config.model_type)

Property	Value
Canonical model type	`vibevoice`
Configuration class	`VibeVoiceConfig`
Architecture class	`VibeVoiceForTextToSpeech`

Processing¶

Create the registered processor without allocating model weights:

from voicehub import AutoProcessor

processor = AutoProcessor.from_pretrained(
    'microsoft/VibeVoice-Realtime-0.5B',
    model_type='vibevoice',
)
print(type(processor).__name__)

Inference¶

The Usage example returns TTSOutput through AutoModelForTextToSpeech.

Input and output contract¶

Property	Value
Readiness	`preprocessed`
Data architecture	`hybrid`
Sample rate	24,000 Hz
Contract getter	`get_tts_dataset_spec('vibevoice')`

Variant	Required fields	One of	Boundary	Other rules
`lm-diffusion-batch`	`input_ids`, `attention_mask`, `speech_tensors`, `speech_masks`, `speeches_loss_input`, `speech_semantic_tensors`, `acoustic_input_mask`, `acoustic_loss_mask`	—	Prepared	—

Multi-component language-model, diffusion, acoustic, or GAN data. See the data workflow.

Training and optimization¶

Use available_optimization_passes() to discover reversible public passes. Unsupported runtime or hardware fails closed before mutation.

Training contract¶

Property	Value
Support	`preprocessed`
Family	`composite`
Recipe	`single-phase`
Default phase	`lm_diffusion`
Training checkpoint	`microsoft/VibeVoice-1.5B`
Native training graph	`yes`

Phase	Kind	Components	Required inputs	Loss keys
`lm_diffusion`	objective	`model`	`input_ids`, `attention_mask`, `speech_tensors`, `speech_masks`, `speeches_loss_input`, `speech_semantic_tensors`, `acoustic_input_mask`, `acoustic_loss_mask`	`loss`, `ce_loss`, `diffusion_loss`

Prepare the exact tensors listed in the data contract before this step. Call model.validate_training_support() first, then follow the training workflow.

Checkpoints, provenance, license, and limitations¶

Property	Value
Default checkpoint	`microsoft/VibeVoice-Realtime-0.5B`
Hugging Face ID	`microsoft/VibeVoice-Realtime-0.5B` Repository availability verified through the Hugging Face model API on 2026-08-11; pin a revision before production use.
Checkpoint status	Registry default; pin an immutable revision for production and reproducible evidence
Optional dependency extra	Core package
Hardware and runtime	Usage selects `cuda`; verify checkpoint-specific requirements
Real-checkpoint evidence	Release evidence; a registry default alone is not execution evidence
Implementation	`voicehub.models.vibevoice.modeling_vibevoice.VibeVoiceForTextToSpeech`
Configuration	`voicehub.models.vibevoice.configuration_vibevoice.VibeVoiceConfig`
Source provenance	`voicehub/models/vibevoice/source/SOURCE.json`
License	Checkpoint-specific

No VoiceHub-specific license override is registered. Verify the checkpoint and upstream source terms before use.

Confirm the checkpoint revision, access terms, provenance, and license.

Limitations¶

No integration-specific checkpoint limitation is registered. Verify the selected checkpoint revision and its documented runtime requirements.
Validate memory, precision, and optional dependencies on the target system.
Public optimizations fail closed when the runtime or hardware cannot satisfy their validation contract; an unavailable pass is not reported as applied.
Contract tests do not replace the linked released-checkpoint evidence.

Public API¶

Use the stable configuration, processor, and task-model facades below.

Configuration

`VibeVoiceConfig`¶

View source

VibeVoiceConfig(**config_kwargs)

Parameters¶

**config_kwargs — Configuration fields validated by VibeVoiceConfig.

Model

`VibeVoiceForTextToSpeech`¶

View source

AutoModelForTextToSpeech.from_pretrained(
    pretrained_model_name_or_path,
    *,
    model_type='vibevoice',
    config=None,
    **model_kwargs,
)

Parameters¶

pretrained_model_name_or_path — Hub ID or compatible local directory.
model_type — Canonical model type; use 'vibevoice'.
config — Optional preloaded VibeVoiceConfig instance.
**model_kwargs — Model-specific loading arguments.

from voicehub import get_model_spec

spec = get_model_spec('vibevoice')
print(spec.display_name, spec.task.value)

Purpose	Public object
Discover	`get_model_spec('vibevoice')`
Load and run	`AutoModelForTextToSpeech`
Configure	`VibeVoiceConfig`
Process	`AutoProcessor`
Model implementation	`VibeVoiceForTextToSpeech`
Normalized output	`TTSOutput`
Training contract	`get_training_spec('vibevoice')`
Optimization lifecycle	`available_optimization_passes`, `apply_optimization_plan`, `optimization_manifest`, `restore_optimization_plan`

See all model guides, inference, and the training matrix.

Usage¶

Overview¶

Language support¶

Paper and GitHub¶

Configuration¶

Processing¶

Inference¶

Input and output contract¶

Training and optimization¶

Training contract¶

Checkpoints, provenance, license, and limitations¶

Limitations¶

Public API¶

VibeVoiceConfig¶

Parameters¶

VibeVoiceForTextToSpeech¶

Parameters¶

`VibeVoiceConfig`¶

`VibeVoiceForTextToSpeech`¶