From 5385dda4feeb417bd94a29a5e62ec904b757a0ef Mon Sep 17 00:00:00 2001
From: Yueqian Lin <70319226+linyueqian@users.noreply.github.com>
Date: Mon, 14 Sep 2026 17:19:21 -0700
Subject: [PATCH] [Bugfix] Use current OpenAI audio event names in duplex
 client metrics (#7541)

Signed-off-by: Yueqian Lin <linyueqian@outlook.com>
---
 tests/clients/test_duplex_client.py                       | 8 ++++----
 .../e2e/features/fullduplex/test_omni_duplex_eval_cli.py  | 2 +-
 vllm_omni/clients/duplex.py                               | 2 +-
 3 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/tests/clients/test_duplex_client.py b/tests/clients/test_duplex_client.py
index 8a74a2087..ab2a38951 100644
--- a/tests/clients/test_duplex_client.py
+++ b/tests/clients/test_duplex_client.py
@@ -776,7 +776,7 @@ def test_event_collector_reports_global_metrics_across_responses():
         )
         collector.add(
             {
-                "type": "response.audio_transcript.delta",
+                "type": "response.output_audio_transcript.delta",
                 "response_id": response_id,
                 "delta": response_id,
             },
@@ -785,7 +785,7 @@ def test_event_collector_reports_global_metrics_across_responses():
         for received_at_s, duration_ms in zip(audio_times, durations, strict=True):
             collector.add(
                 {
-                    "type": "response.audio.delta",
+                    "type": "response.output_audio.delta",
                     "response_id": response_id,
                     "delta": base64.b64encode(b"audio").decode("ascii"),
                     "sample_rate_hz": 16_000,
@@ -856,7 +856,7 @@ def test_event_collector_prefers_server_request_start_for_ttf():
     for received_at_s, cumulative_audio_ms in ((10.2, 80), (10.25, 160), (10.36, 240)):
         collector.add(
             {
-                "type": "response.audio.delta",
+                "type": "response.output_audio.delta",
                 "response_id": "resp-a",
                 "delta": base64.b64encode(b"audio").decode("ascii"),
                 "sample_rate_hz": 16_000,
@@ -868,7 +868,7 @@ def test_event_collector_prefers_server_request_start_for_ttf():
             received_at_s=received_at_s,
         )
     collector.add(
-        {"type": "response.audio_transcript.delta", "response_id": "resp-a", "delta": "hello"},
+        {"type": "response.output_audio_transcript.delta", "response_id": "resp-a", "delta": "hello"},
         received_at_s=10.15,
     )
 
diff --git a/tests/e2e/features/fullduplex/test_omni_duplex_eval_cli.py b/tests/e2e/features/fullduplex/test_omni_duplex_eval_cli.py
index 7f4965768..b1b02ef65 100644
--- a/tests/e2e/features/fullduplex/test_omni_duplex_eval_cli.py
+++ b/tests/e2e/features/fullduplex/test_omni_duplex_eval_cli.py
@@ -389,7 +389,7 @@ def test_cli_generate_partial_resume_merges_metrics(tmp_path: Path, monkeypatch:
 def test_collect_duplex_session_metrics_matches_omniinteract_window():
     collector = EventCollector()
     audio = {
-        "type": "response.audio.delta",
+        "type": "response.output_audio.delta",
         "response_id": "r1",
         "format": "pcm16",
         "delta": base64.b64encode(bytes((1, 0)) * 2400).decode(),
diff --git a/vllm_omni/clients/duplex.py b/vllm_omni/clients/duplex.py
index 0504c241b..a1d9fc1df 100644
--- a/vllm_omni/clients/duplex.py
+++ b/vllm_omni/clients/duplex.py
@@ -1475,7 +1475,7 @@ class EventCollector:
             if (
                 event.get("type")
                 in {
-                    "response.audio_transcript.delta",
+                    "response.output_audio_transcript.delta",
                     "response.output_text.delta",
                     "response.text.delta",
                 }