fix(tts): correct cosy audio end interval

ai_agents/agents/ten_packages/extension/cosy_tts_python/extension.py

@@ -50,6 +50,8 @@ def __init__(self, name: str) -> None:
         self.recorder_map: dict[str, PCMWriter] = {}
         # Timestamp when TTS request was sent to service
         self.request_start_ts: datetime | None = None
+        # Timestamp when the first audio chunk was received
+        self.first_chunk_ts: datetime | None = None
         # Total audio duration for current request in milliseconds
         self.request_total_audio_duration_ms: int | None = None
         # Time to first byte for current request in milliseconds
@@ -196,6 +198,7 @@ async def request_tts(self, t: TTSTextInput) -> None:
                 self.total_audio_bytes = 0  # Reset for new request
                 self.request_ttfb = None
                 self.first_chunk = True
+                self.first_chunk_ts = None
                 self.chunk_count = 0
                 self.request_start_ts = datetime.now()
                 self.is_first_message_of_request = True
@@ -534,6 +537,8 @@ async def _handle_first_audio_chunk(self) -> None:
         4. Logs the operation
         """
         if self.request_start_ts:
+            self.first_chunk_ts = datetime.now()
+
             await self.send_tts_audio_start(
                 self.current_request_id,
             )
@@ -573,9 +578,12 @@ async def _handle_tts_audio_end(
                     self.total_audio_bytes, self.config.sample_rate
                 )
             )
-            request_event_interval = int(
-                (datetime.now() - self.request_start_ts).total_seconds() * 1000
-            )
+            request_event_interval = 0
+            if self.first_chunk_ts is not None:
+                request_event_interval = int(
+                    (datetime.now() - self.first_chunk_ts).total_seconds()
+                    * 1000
+                )
 
             # Flush PCMWriter for current request to ensure dump file is written
             if (

ai_agents/agents/ten_packages/extension/cosy_tts_python/manifest.json

@@ -1,7 +1,7 @@
 {
   "type": "extension",
   "name": "cosy_tts_python",
-  "version": "0.4.2",
+  "version": "0.4.3",
   "dependencies": [
     {
       "type": "system",

ai_agents/agents/ten_packages/extension/cosy_tts_python/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "cosy-tts-python"
-version = "0.4.2"
+version = "0.4.3"
 requires-python = ">=3.10"
 dependencies = [
     "dashscope>=1.26.4",

ai_agents/agents/ten_packages/extension/cosy_tts_python/tests/test_metrics.py

@@ -28,6 +28,7 @@ def __init__(self):
         self.ttfb_value = -1
         self.audio_frame_received = False
         self.audio_end_received = False
+        self.request_event_interval_ms = -1
 
     def on_start(self, ten_env_tester: TenEnvTester) -> None:
         """Called when test starts, sends a TTS request."""
@@ -61,6 +62,11 @@ def on_data(self, ten_env: TenEnvTester, data) -> None:
 
         elif name == "tts_audio_end":
             self.audio_end_received = True
+            json_str, _ = data.get_property_to_json(None)
+            payload = json.loads(json_str)
+            self.request_event_interval_ms = payload.get(
+                "request_event_interval_ms", -1
+            )
             # Stop the test only after both TTFB and audio end are received
             if self.ttfb_received:
                 ten_env.log_info("Received tts_audio_end, stopping test.")
@@ -98,7 +104,7 @@ async def simulate_audio_stream():
                 """Simulate TTS service sending data asynchronously"""
                 queue = stream_state["queue"]
                 # Delay to simulate network latency and TTS processing time
-                await asyncio.sleep(0.25)  # 250ms TTFB
+                await asyncio.sleep(0.6)  # 600ms TTFB
 
                 # Send first audio chunk
                 await queue.put((False, MESSAGE_TYPE_PCM, b"\x11\x22\x33"))
@@ -139,11 +145,18 @@ async def simulate_audio_stream():
     assert tester.audio_frame_received, "Did not receive any audio frame."
     assert tester.audio_end_received, "Did not receive the tts_audio_end event."
     assert tester.ttfb_received, "TTFB metric was not received."
+    assert (
+        tester.request_event_interval_ms >= 0
+    ), "tts_audio_end did not include request_event_interval_ms."
 
-    # Check if the TTFB value is reasonable. It should be around 250ms with the delay
-    # we introduced. Allow 50ms margin for timing variations and system scheduling.
+    # Check if the TTFB value is reasonable. The larger delay keeps it well
+    # above the post-first-chunk interval despite scheduling variability.
     assert (
-        tester.ttfb_value >= 200
-    ), f"Expected TTFB to be >= 200ms, but got {tester.ttfb_value}ms."
+        tester.ttfb_value >= 500
+    ), f"Expected TTFB to be >= 500ms, but got {tester.ttfb_value}ms."
+    assert tester.request_event_interval_ms < tester.ttfb_value, (
+        "request_event_interval_ms should start at the first audio chunk and "
+        "exclude TTFB."
+    )
 
     print(f"✅ TTFB metric test passed. Received TTFB: {tester.ttfb_value}ms.")

ai_agents/agents/ten_packages/extension/cosy_tts_python/tests/test_robustness.py

@@ -23,6 +23,7 @@ class ExtensionTesterRobustness(ExtensionTester):
     def __init__(self):
         super().__init__()
         self.first_request_error: Optional[dict[str, Any]] = None
+        self.first_request_event_interval_ms: Optional[int] = None
         self.second_request_successful = False
         self.ten_env: Optional[TenEnvTester] = None
 
@@ -75,6 +76,14 @@ def on_data(self, ten_env: TenEnvTester, data) -> None:
             # After receiving the error for the first request, immediately send the second one.
             self.send_second_request()
 
+        elif (
+            name == "tts_audio_end"
+            and payload.get("request_id") == "tts_request_to_fail"
+        ):
+            self.first_request_event_interval_ms = payload.get(
+                "request_event_interval_ms"
+            )
+
         elif (
             name == "tts_audio_end"
             and payload.get("request_id") == "tts_request_to_succeed"
@@ -181,6 +190,10 @@ async def get_audio_data(self):
     assert (
         tester.first_request_error.get("code") == 1000
     ), f"Expected error code 1000 (NON_FATAL_ERROR), got {tester.first_request_error.get('code')}"
+    assert tester.first_request_event_interval_ms == 0, (
+        "A request with no audio should report a zero event interval, got "
+        f"{tester.first_request_event_interval_ms}ms"
+    )
 
     # 2. Verify that vendor_info was included in the error
     vendor_info = tester.first_request_error.get("vendor_info")