Fix posture-check window: epoch timestamps captured at speech-end/answer-end
Browse files- tests/e2e_roleplay.py +5 -3
tests/e2e_roleplay.py
CHANGED
|
@@ -281,7 +281,8 @@ async def conversation():
|
|
| 281 |
np.copyto(np.frombuffer(frame.data, dtype=np.int16), piece)
|
| 282 |
await source.capture_frame(frame)
|
| 283 |
t_speech_end = time.perf_counter()
|
| 284 |
-
|
|
|
|
| 285 |
|
| 286 |
while audio_state["first"] is None:
|
| 287 |
await asyncio.sleep(0.05)
|
|
@@ -297,6 +298,7 @@ async def conversation():
|
|
| 297 |
if time.perf_counter() - t_speech_end > 90:
|
| 298 |
break
|
| 299 |
bot_pcm = np.concatenate(audio_state["chunks"])
|
|
|
|
| 300 |
segs, _ = verifier.transcribe(bot_pcm.astype(np.float32) / 32768.0, language="en")
|
| 301 |
text = " ".join(s.text.strip() for s in segs).strip()
|
| 302 |
ok = any(k in text.lower() for k in keywords)
|
|
@@ -307,8 +309,8 @@ async def conversation():
|
|
| 307 |
turns.append({"t_perceived": t_perceived, "text": text,
|
| 308 |
"t_speech_end": t_speech_end,
|
| 309 |
"t_end": audio_state["last"],
|
| 310 |
-
"t_speech_end_abs":
|
| 311 |
-
"t_end_abs":
|
| 312 |
await asyncio.sleep(0.5)
|
| 313 |
|
| 314 |
await room.disconnect()
|
|
|
|
| 281 |
np.copyto(np.frombuffer(frame.data, dtype=np.int16), piece)
|
| 282 |
await source.capture_frame(frame)
|
| 283 |
t_speech_end = time.perf_counter()
|
| 284 |
+
t_speech_end_abs = time.time() # captured AT speech end, not at append
|
| 285 |
+
print(f"[client] turn {i}: '{q_text}' pushed (abs {t_speech_end_abs:.3f})", flush=True)
|
| 286 |
|
| 287 |
while audio_state["first"] is None:
|
| 288 |
await asyncio.sleep(0.05)
|
|
|
|
| 298 |
if time.perf_counter() - t_speech_end > 90:
|
| 299 |
break
|
| 300 |
bot_pcm = np.concatenate(audio_state["chunks"])
|
| 301 |
+
t_end_abs = time.time() # end of the answer (silence wait included)
|
| 302 |
segs, _ = verifier.transcribe(bot_pcm.astype(np.float32) / 32768.0, language="en")
|
| 303 |
text = " ".join(s.text.strip() for s in segs).strip()
|
| 304 |
ok = any(k in text.lower() for k in keywords)
|
|
|
|
| 309 |
turns.append({"t_perceived": t_perceived, "text": text,
|
| 310 |
"t_speech_end": t_speech_end,
|
| 311 |
"t_end": audio_state["last"],
|
| 312 |
+
"t_speech_end_abs": t_speech_end_abs,
|
| 313 |
+
"t_end_abs": t_end_abs})
|
| 314 |
await asyncio.sleep(0.5)
|
| 315 |
|
| 316 |
await room.disconnect()
|