add-tts-model
Integrate a new text-to-speech model into vLLM-Omni from HuggingFace reference implementation through production-ready serving with streaming and CUDA graph acceleration. Use when adding a new TTS model, wiring stage separation for speech synthesis, enabling online voice generation serving, debugging TTS integration behavior, or building audio output pipelines.
Best for DevOps teams deploying TTS…Works with GitHubLow risk
⌘source
- author
@vllm-project- language
- Python
★4.4kstars
787forks
Apr 18, 2026
✦overview.md
Key Features
- ·Integrates new TTS models into vLLM-Omni from HuggingFace references
- ·Enables production-ready serving with streaming support
- ·Utilizes CUDA graph acceleration for performance
- ·Wires stage separation for speech synthesis pipelines
- ·Supports online voice generation serving
- ·Facilitates debugging of TTS integration behavior
Use Cases
- →Adding a new text-to-speech model to vLLM-Omni
- →Building audio output pipelines for speech synthesis
- →Debugging TTS integration behavior in production systems
- →Enabling online voice generation serving with streaming
Best for
- ✓DevOps teams deploying TTS models
- ✓Performance-critical speech synthesis applications
Not ideal for
- !One-off model experiments without production deployment
- !Non-CUDA environments
FAQs
name
add-tts-model
description
Integrate a new text-to-speech model into vLLM-Omni from HuggingFace reference implementation through production-ready serving with streaming and CUDA graph acceleration. Use when adding a new TTS model, wiring stage separation for speech synthesis, enabling online voice generation serving, debugging TTS integration behavior, or building audio output pipelines.
TTS Model Integration Workflow
Overview
HF Reference -> Stage Separation -> Online Serving -> Async Chunk -> CUDA Graph
(Phase 1) (Phase 2) (Phase 3) (Phase 4) (Phase 5)
Phase 1: HuggingFace Reference
Goal: Understand the reference implementation and verify it produces correct audio.
Steps
- Run the reference model end-to-end using the official HuggingFace / GitHub code
- Document the architecture:
- What are the sub-models? (AR decoder, codec decoder, vocoder, etc.)
- What is the token vocabulary? (semantic codes, RVQ codebooks, special tokens)
- What is the output format? (sample rate, channels, codec type)
- Capture reference outputs for comparison during integration
- Identify the config structure:
config.jsonfields,model_type, sub-model configs
Key Questions
- How many codebooks? What are the codebook sizes?
- What special tokens exist? (
<|voice|>,<|audio_start|>,<|im_end|>, etc.) - What is the token-to-ID mapping for codec codes?
- What is the hop length / frame rate of the codec?
$install
1-click copynpx skills add vllm-project/vllm-omni --skill add-tts-modelSafety assessment
★
Clarity score
How clear and easy to understand the SKILL.md instructions are, rated from 1 to 5.
3/ 5
goodMostly clear, but there are still a few confusing or poorly structured parts.
◎
Actionability score
How directly an agent can act on the SKILL.md instructions, rated from 1 to 5.
3/ 5
mediumPartially actionable with several concrete steps, but still missing important details.