Skip to content

Commit 373adc3

Browse files
committed
fix(#350): verify #465 fix via fresh twilio_inbound recording
End-to-end proof of the Say-line + judge-quality-criterion fix: - Agent segment transcript is now "Thank you for calling. I will hold the line while you complete your scenario." — exactly the deterministic <Say> line the originator's TwiML now plays. No more Greek hallucination on line silence. - Judge passed all 4 criteria including the new transcript-coherence check: "Any captured agent audio transcribes as coherent English". - Recording downsampled 24kHz → 8kHz to fit pre-commit's 1MB cap; fidelity is telephony-grade and matches the demo's mulaw 8kHz transport reality. The bug Rogério called out (silent line after the quick question, Whisper hallucinating Greek) cannot reoccur on this demo: the originator now always plays a known utterance.
1 parent f327b21 commit 373adc3

5 files changed

Lines changed: 16 additions & 16 deletions

File tree

1.53 KB
Binary file not shown.

python/recordings/twilio_inbound/manifest.json

Lines changed: 16 additions & 16 deletions
Original file line numberDiff line numberDiff line change
@@ -1,43 +1,43 @@
11
{
2-
"generated_at": "2026-05-12T17:32:16.301777+00:00",
3-
"duration": 29.760395499994047,
2+
"generated_at": "2026-05-14T18:12:08.513670+00:00",
3+
"duration": 30.18283020800027,
44
"segment_count": 2,
55
"segments": [
66
{
77
"idx": 0,
8-
"file": "segments/00-user-1853ms.wav",
8+
"file": "segments/00-user-19126ms.wav",
99
"role": "user",
10-
"start_time": 1.8534086250001565,
11-
"end_time": 4.484551249988726,
12-
"duration": 2.63114262498857,
10+
"start_time": 19.126248375000614,
11+
"end_time": 21.765976292000232,
12+
"duration": 2.639727916999618,
1313
"transcript": "Hello, I have a quick question."
1414
},
1515
{
1616
"idx": 1,
17-
"file": "segments/01-agent-4484ms.wav",
17+
"file": "segments/01-agent-21766ms.wav",
1818
"role": "agent",
19-
"start_time": 4.484716457998729,
20-
"end_time": 29.760395499994047,
21-
"duration": 25.275679041995318,
22-
"transcript": "\u0391\u03c5\u03c4\u03cc \u03c4\u03bf \u03b2\u03b9\u03b2\u03bb\u03af\u03bf"
19+
"start_time": 21.766091167000013,
20+
"end_time": 30.18283020800027,
21+
"duration": 8.416739041000255,
22+
"transcript": "Thank you for calling. I will hold the line while you complete your scenario."
2323
}
2424
],
2525
"events": [
2626
{
27-
"time": 1.8534086250001565,
27+
"time": 19.126248375000614,
2828
"type": "user_start_speaking"
2929
},
3030
{
31-
"time": 4.484551249988726,
31+
"time": 21.765976292000232,
3232
"type": "user_stop_speaking"
3333
},
3434
{
35-
"time": 4.484716457998729,
35+
"time": 21.766091167000013,
3636
"type": "agent_start_speaking",
37-
"latency": 0.0001652080100029707
37+
"latency": 0.00011487499978102278
3838
},
3939
{
40-
"time": 29.760395499994047,
40+
"time": 30.18283020800027,
4141
"type": "agent_stop_speaking"
4242
}
4343
]
-37.6 KB
Binary file not shown.
117 KB
Binary file not shown.

python/recordings/twilio_inbound/segments/01-agent-4484ms.wav renamed to python/recordings/twilio_inbound/segments/01-agent-21766ms.wav

470 KB
Binary file not shown.

0 commit comments

Comments
 (0)