[Bugfix] Fix Qwen3-TTS gradio streaming TTFP by using audio/pcm (#4346)

Signed-off-by: YU Xinyuan <yuxinyuan02@corp.netease.com>
Co-authored-by: YU Xinyuan <yuxinyuan02@corp.netease.com>
Co-authored-by: Yueqian Lin <70319226+linyueqian@users.noreply.github.com>

examples/online_serving/text_to_speech/qwen3_tts/gradio_demo.py

@@ -408,7 +408,10 @@ def create_app(api_base: str):
                 await resp.aclose()
                 await client.aclose()
 
-        return StreamingResponse(relay(), media_type="application/octet-stream")
+        # Use audio/pcm to bypass browsers (e.g. Chromium) MIME type sniffing, which
+        # buffers application/octet-stream responses before delivering chunks to
+        # fetch().body.getReader(), inflating TTFP to total_ms.
+        return StreamingResponse(relay(), media_type="audio/pcm")
 
     # ── Gradio UI ────────────────────────────────────────────────
     voices = fetch_voices(api_base)