Zum Inhalt

HKUSTAudioLlasa-1B-Multilingual

Llasa

Pairs LLaSA reference audio with its exact transcript for voice cloning.

Text to speechVoiceHub-nativellasaParameters: 1.8BLanguages: zh, en +9Training: nativeLicense: CC-BY-NC-4.0

Parameter metadata: Exact Safetensors total reported by the Hugging Face model API for the registered default checkpoint, retrieved 2026-08-13.

Use this model Checkpoint
Resources
Upstream GitHub VoiceHub source Open in Colab

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: Pairs LLaSA reference audio with its exact transcript for voice cloning.

Inputs and controls: Both reference fields are required together; VoiceHub rejects incomplete cloning context.

from pathlib import Path

from voicehub import AutoModelForTextToSpeech, TTSGenerationConfig

REFERENCE_AUDIO = Path("reference.wav")
REFERENCE_TEXT = "The reference transcript must exactly match the authorized audio."
if not REFERENCE_AUDIO.is_file():
    raise FileNotFoundError(REFERENCE_AUDIO)

model = AutoModelForTextToSpeech.from_pretrained(
    'HKUSTAudio/Llasa-1B-Multilingual',
    model_type='llasa',
    device="cuda",
    lazy_load=True,
)
output = model.generate(
    'VoiceHub keeps model integrations explicit and reproducible.',
    generation_config=TTSGenerationConfig(
        seed=42,
        output_file=Path("output.wav"),
    ),
    speaker_audio_path=str(REFERENCE_AUDIO),
    reference_text=REFERENCE_TEXT,
    max_new_tokens=1_024,
    top_p=0.9,
)
print(output.file_path, output.sample_rate, output.metadata)

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

Overview

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

Property Value
Task Text to speech
Architecture llasa
Runtime VoiceHub-native
Languages zh, en, de, fr, … complete audited list below
Capabilities text-to-speech, voice-cloning, multilingual, fine-tuning, safetensors, voicehub-native, native-runtime, raw-audio-fine-tuning, preencoded-code-fine-tuning
Reusable components —
Normalized output TTSOutput

Language support

Supported language abbreviations

zh, en, de, fr, ja, ko, nl, es, it, pt, pl

Paper and GitHub

Configuration

Load configuration without constructing the model:

from voicehub import AutoConfig

config = AutoConfig.for_model('llasa')
print(config.model_type)
Property Value
Canonical model type llasa
Configuration class LlasaConfig
Architecture class LlasaForTextToSpeech

Processing

Create the registered processor without allocating model weights:

from voicehub import AutoProcessor

processor = AutoProcessor.from_pretrained(
    'HKUSTAudio/Llasa-1B-Multilingual',
    model_type='llasa',
)
print(type(processor).__name__)

Inference

The Usage example returns TTSOutput through AutoModelForTextToSpeech.

Input and output contract

Property Value
Readiness integrated-raw
Data architecture codec-lm
Sample rate 16,000 Hz
Contract getter get_tts_dataset_spec('llasa')
Variant Required fields One of Boundary Other rules
raw-audio text audio / audio_codes Source —
tokenized input_ids, labels — Prepared —

Autoregressive text/audio-token or codec-language-model 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 native
Family causal-lm
Recipe single-phase
Default phase codec_language_model
Training checkpoint HKUSTAudio/Llasa-1B-Multilingual
Native training graph yes
Phase Kind Components Required inputs Loss keys
codec_language_model objective model input_ids, attention_mask, labels loss

The integration accepts its declared source or prepared contract directly. Call model.validate_training_support() first, then follow the training workflow.

Checkpoints, provenance, license, and limitations

Property Value
Default checkpoint HKUSTAudio/Llasa-1B-Multilingual
Hugging Face ID HKUSTAudio/Llasa-1B-Multilingual
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.llasa.modeling_llasa.LlasaForTextToSpeech
Configuration voicehub.models.llasa.configuration_llasa.LlasaConfig
Source provenance voicehub/models/llasa/source/SOURCE.json
License CC-BY-NC-4.0

The vendored XCodec2 component is restricted to non-commercial use. Commercial use: not allowed.

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

LlasaConfig

View source

LlasaConfig(**config_kwargs)

Parameters

  • **config_kwargs — Configuration fields validated by LlasaConfig.

Model

LlasaForTextToSpeech

View source

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

Parameters

  • pretrained_model_name_or_path — Hub ID or compatible local directory.
  • model_type — Canonical model type; use 'llasa'.
  • config — Optional preloaded LlasaConfig instance.
  • **model_kwargs — Model-specific loading arguments.
from voicehub import get_model_spec

spec = get_model_spec('llasa')
print(spec.display_name, spec.task.value)
Purpose Public object
Discover get_model_spec('llasa')
Load and run AutoModelForTextToSpeech
Configure LlasaConfig
Process AutoProcessor
Model implementation LlasaForTextToSpeech
Normalized output TTSOutput
Training contract get_training_spec('llasa')
Optimization lifecycle available_optimization_passes, apply_optimization_plan, optimization_manifest, restore_optimization_plan

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