update md

Author: feizi.wx

docs/source/en/model_doc/qwen2_5_omni.md

@@ -175,6 +175,15 @@ print(text)
 
 ### Usage Tips
 
+#### Prompt for audio output
+If users need audio output, the system prompt must be set as "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech.", otherwise the audio output may not work as expected.
+```
+{
+    "role": "system",
+    "content": "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech.",
+}
+```
+
 #### Use audio output or not
 
 The model supports both text and audio outputs, if users do not need audio outputs, they can set `enable_audio_output` in the `from_pretrained` function. This option will save about `~2GB` of GPU memory but the `return_audio` option for `generate` function will only allow to be set at `False`.