vllm-omni-audio-tts

Generate speech and audio by orchestrating Qwen3-TTS, MiMo-Audio, and Stable-Audio within vLLM-Omni. This approach simplifies cross-model workflows for text-to-speech, voice cloning, and sound effect generation through a unified API.

Voice cloning from reference samples is supported for text-to-speech generation. You provide a text prompt and reference audio to generate realistic speech that mimics the cloned voice persona using the configured models.

Create narrated audio with synchronized background music by using multi-model audio generation. You can combine cloned presenter speech with audio composed of music and sound effects generated by Stable-Audio and MiMo-Audio.

You need Omni client and server setup to run text-to-audio tasks. The workflow requires configuring the listed models and environment details described in the documentation for both offline and server modes.

vLLM-Omni supports voice design and unique voice persona creation. You can design custom voices and generate audio content tailored to specific character profiles or application requirements using Qwen3-TTS.