qwen3_tts

Model type

qwen3_tts

Class

Qwen3TTSForConditionalGeneration

Task

tts

Source

models/qwen3_tts.py

Description

Qwen3-TTS composite model for the full multi-model TTS pipeline.

Contains:

  • talker: Talker decoder (inputs_embeds → logits + hidden)

  • code_predictor: Code predictor (hidden → 15 code groups)

  • embedding: Text + codec embedding model

  • talker_step_embedder: Per-frame talker input construction

  • talker_prefill_embedder: Prompt prefill/trailing-text construction

  • speaker_encoder: Optional ECAPA-TDNN speaker encoder

HuggingFace class: Qwen3TTSForConditionalGeneration

Usage

mobius build --model <MODEL_ID> output_dir/

from mobius import build

model = build("<MODEL_ID>")