Serious Issues with Real-Time Streaming

[mingheyuemankong]

In the official streaming inference implementation, audio is accumulated from the beginning during inference. See:
https://github.com/QwenLM/Qwen3-ASR/blob/main/qwen_asr/inference/qwen3_asr.py#L800-L807

As the conversation progresses, the accumulated audio length increases, eventually leading to stuttering or even unresponsiveness. Therefore, this streaming is essentially pseudo-streaming because it does not perform incremental inference, which is in serious conflict with the actual claims. I would like to know if true streaming can be provided.