[New Model]: Add PersonaPlex (nvidia/personaplex-7b-v1) MVP support

The model to consider.

https://huggingface.co/nvidia/personaplex-7b-v1

The closest model vllm-omni already supports.

It is closest to Qwen3-Omni because both are autoregressive multimodal speech models with staged audio generation. Qwen3-TTS is also related for the speech synthesis/output path, but PersonaPlex adds speech-to-speech conversational behavior (including duplex interaction).

What's your difficulty of supporting the model you want?

Main difficulty is that PersonaPlex is full-duplex speech-to-speech, while current vLLM-Omni paths are mostly staged/turn-based. I plan to contribute Phase 1 (MVP) only: turn-based speech-in/speech-out support with model integration, architecture registration, stage input processor, stage config, basic offline/online examples, and minimal tests/docs updates. Full-duplex overlap/barge-in behavior will be out of scope for this PR and handled in a later phase.

Use case and motivation

real-time voice assistant (speech in, speech out). It is important because it adds strong speech-to-speech support to vllm-omni

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.