Add Listen section: 5 playable stereo demos (user left / agent right), refreshed sample logs
2ea57e9 verified | { | |
| "name": "05_igbo_heavy", | |
| "desc": "Igbo-heavier user (shows the weak spot: LLM Igbo beyond phrases). User turns TTS-synthesized (Flash, female Igbo voice).", | |
| "user_turns": [ | |
| { | |
| "label": "TTS-synthesized user turn [identical input audio recovered from the 2026-10-24 demo]", | |
| "text": "<ibo>Kedu? Biko, nyere m aka.</ibo> <eng>I need a taxi to the airport.</eng>", | |
| "wav": "/scratch/duplex_v2/user_audio/05_igbo_heavy_turn0.wav", | |
| "orig_label": "TTS-synthesized user turn" | |
| }, | |
| { | |
| "label": "TTS-synthesized user turn [identical input audio recovered from the 2026-10-24 demo]", | |
| "text": "<ibo>Ee, ọ dị mma.</ibo> <eng>I will leave at six.</eng>", | |
| "wav": "/scratch/duplex_v2/user_audio/05_igbo_heavy_turn1.wav", | |
| "orig_label": "TTS-synthesized user turn" | |
| }, | |
| { | |
| "label": "TTS-synthesized user turn [identical input audio recovered from the 2026-10-24 demo]", | |
| "text": "<ibo>Daalụ, ka ọ dị.</ibo>", | |
| "wav": "/scratch/duplex_v2/user_audio/05_igbo_heavy_turn2.wav", | |
| "orig_label": "TTS-synthesized user turn" | |
| } | |
| ], | |
| "turns": [ | |
| { | |
| "turn": 0, | |
| "user_label": "TTS-synthesized user turn [identical input audio recovered from the 2026-10-24 demo]", | |
| "user_ref": "<ibo>Kedu? Biko, nyere m aka.</ibo> <eng>I need a taxi to the airport.</eng>", | |
| "user_start": 0.7, | |
| "user_onset": 0.753, | |
| "backchannels": [], | |
| "asr": "kedu biko nyere m aka i need a taxis to the airport", | |
| "asr_s": 0.093, | |
| "eot_wait_s": 0.576, | |
| "speech_end": 5.632, | |
| "target": { | |
| "mix": { | |
| "eng": 0.58, | |
| "ibo": 0.42 | |
| }, | |
| "matrix": "eng", | |
| "pattern": [ | |
| "greeting:ibo", | |
| "discourse:ibo", | |
| "inter-sentential" | |
| ] | |
| }, | |
| "target_desc": "Match the user's language mix: English about 60%, Igbo about 40% of the words. The main (matrix) language is English. The user greeted in Igbo: open with a short Igbo greeting. Sprinkle a Igbo discourse word or two the way the user did.", | |
| "reply": "<ibo>Kedu?</ibo> <eng>Okay, I can help you book a taxi to the airport. How soon do you need it?</eng> <ibo>Biko</ibo>", | |
| "reply_mix": { | |
| "eng": 0.89, | |
| "ibo": 0.11 | |
| }, | |
| "tvd": 0.312, | |
| "how": "best_failed", | |
| "first_cand_ok": false, | |
| "rounds": [ | |
| { | |
| "round": 0, | |
| "llm_s": 0.302, | |
| "cands": [ | |
| { | |
| "text": "<ibo>Kedu?</ibo> <eng>Okay, I can help you book a taxi to the airport. How soon do you need it?</eng> <ibo>Biko</ibo>", | |
| "ok": false, | |
| "why": [ | |
| "mix drift tvd=0.31" | |
| ], | |
| "tvd": 0.312 | |
| }, | |
| { | |
| "text": "<ibo>Kedu?</ibo> <eng>Okay, I can help you book a taxi to the airport. How soon do you need it?</eng> <ibo>Na so?</ibo>", | |
| "ok": false, | |
| "why": [ | |
| "mix drift tvd=0.45" | |
| ], | |
| "tvd": 0.45 | |
| }, | |
| { | |
| "text": "<ibo>Kedu?</ibo> <eng>Yes, I can help you get a taxi to the airport. How soon do you need it?</eng> <ibo>Biko,</ibo> <eng>what time are you leaving?</eng>", | |
| "ok": false, | |
| "why": [ | |
| "mix drift tvd=0.32" | |
| ], | |
| "tvd": 0.318 | |
| } | |
| ] | |
| }, | |
| { | |
| "round": 1, | |
| "llm_s": 0.236, | |
| "cands": [ | |
| { | |
| "text": "<ibo>Kedu?</ibo> <eng>Biko, I go get you a taxi to the airport. How far you need to go?</eng>", | |
| "ok": false, | |
| "why": [ | |
| "mix drift tvd=0.44" | |
| ], | |
| "tvd": 0.44 | |
| }, | |
| { | |
| "text": "<ibo>Kedu?</ibo> <eng>Biko, I go get a taxi for you to airport now. You dey wait for the car at your gate?</eng>", | |
| "ok": false, | |
| "why": [ | |
| "mix drift tvd=0.43" | |
| ], | |
| "tvd": 0.43 | |
| }, | |
| { | |
| "text": "<ibo>Kedu?</ibo> <eng>Biko, I dey get a taxi for you to airport. How far you dey from airport?</eng>", | |
| "ok": false, | |
| "why": [ | |
| "mix drift tvd=0.44" | |
| ], | |
| "tvd": 0.44 | |
| } | |
| ] | |
| } | |
| ], | |
| "mix_s": 0.0085, | |
| "llm_s": 0.538, | |
| "tts_first_s": 0.678, | |
| "spec_start_s": 0.256, | |
| "compute_to_first_audio_s": 1.351, | |
| "tts_breakdown": { | |
| "fe": 0.035, | |
| "lm": 0.09, | |
| "flow_voc": 0.202, | |
| "n_tok": 8, | |
| "audio_s": 0.32 | |
| }, | |
| "tts_chunks_s": [ | |
| 0.678, | |
| 1.775, | |
| 1.017, | |
| 0.328 | |
| ], | |
| "first_audio": 7.239, | |
| "reply_end": 15.054, | |
| "latency_s": 1.607 | |
| }, | |
| { | |
| "turn": 1, | |
| "user_label": "TTS-synthesized user turn [identical input audio recovered from the 2026-10-24 demo]", | |
| "user_ref": "<ibo>Ee, ọ dị mma.</ibo> <eng>I will leave at six.</eng>", | |
| "user_start": 15.754, | |
| "user_onset": 15.811, | |
| "backchannels": [], | |
| "asr": "ee ọ dị mma ewe live at six", | |
| "asr_s": 0.077, | |
| "eot_wait_s": 0.576, | |
| "speech_end": 18.4, | |
| "target": { | |
| "mix": { | |
| "ibo": 0.65, | |
| "eng": 0.35 | |
| }, | |
| "matrix": "ibo", | |
| "pattern": [ | |
| "inter-sentential" | |
| ] | |
| }, | |
| "target_desc": "Match the user's language mix: Igbo about 65%, English about 35% of the words. The main (matrix) language is Igbo.", | |
| "reply": "<ibo>Ee, ọ dị mma.</ibo> <eng>Got it, I’ll book the taxi for six o’clock.</eng> <ibo>Ọ dị mma na?</ibo>", | |
| "reply_mix": { | |
| "eng": 0.56, | |
| "ibo": 0.44 | |
| }, | |
| "tvd": 0.21, | |
| "how": "first_round", | |
| "first_cand_ok": false, | |
| "rounds": [ | |
| { | |
| "round": 0, | |
| "llm_s": 0.268, | |
| "cands": [ | |
| { | |
| "text": "<ibo>Ee, ọ dị mma.</ibo> <eng>Okay, I’ll make sure the taxi arrives by six.</eng> <ibo>Biko</ibo>", | |
| "ok": false, | |
| "why": [ | |
| "mix drift tvd=0.33" | |
| ], | |
| "tvd": 0.326 | |
| }, | |
| { | |
| "text": "<ibo>Ee, ọ dị mma!</ibo> <eng>The flight leaves at six. Is that correct?</eng> <ibo>Biko</ibo>", | |
| "ok": false, | |
| "why": [ | |
| "mix drift tvd=0.30" | |
| ], | |
| "tvd": 0.303 | |
| }, | |
| { | |
| "text": "<ibo>Ee, ọ dị mma.</ibo> <eng>Got it, I’ll book the taxi for six o’clock.</eng> <ibo>Ọ dị mma na?</ibo>", | |
| "ok": true, | |
| "why": [], | |
| "tvd": 0.21 | |
| } | |
| ] | |
| } | |
| ], | |
| "mix_s": 0.0058, | |
| "llm_s": 0.268, | |
| "tts_first_s": 0.913, | |
| "spec_start_s": 0.256, | |
| "compute_to_first_audio_s": 1.278, | |
| "tts_breakdown": { | |
| "fe": 0.0, | |
| "lm": 0.634, | |
| "flow_voc": 0.217, | |
| "n_tok": 40, | |
| "audio_s": 1.6 | |
| }, | |
| "tts_chunks_s": [ | |
| 0.913, | |
| 1.908, | |
| 0.851 | |
| ], | |
| "first_audio": 19.935, | |
| "reply_end": 27.483, | |
| "latency_s": 1.535 | |
| }, | |
| { | |
| "turn": 2, | |
| "user_label": "TTS-synthesized user turn [identical input audio recovered from the 2026-10-24 demo]", | |
| "user_ref": "<ibo>Daalụ, ka ọ dị.</ibo>", | |
| "user_start": 28.183, | |
| "user_onset": 28.248, | |
| "backchannels": [], | |
| "asr": "daalụ kaọdị", | |
| "asr_s": 0.056, | |
| "eot_wait_s": 0.576, | |
| "speech_end": 29.664, | |
| "target": { | |
| "mix": { | |
| "ibo": 0.88, | |
| "eng": 0.12 | |
| }, | |
| "matrix": "ibo", | |
| "pattern": [ | |
| "closing:ibo" | |
| ] | |
| }, | |
| "target_desc": "Match the user's language mix: Igbo about 90%, English about 10% of the words. The main (matrix) language is Igbo. The user closed in Igbo: you may close in Igbo.", | |
| "reply": "<ibo>Daalụ kaọdị.</ibo> <eng>See you later!</eng> <ibo>Ọ dị mma.</ibo>", | |
| "reply_mix": { | |
| "eng": 0.38, | |
| "ibo": 0.62 | |
| }, | |
| "tvd": 0.257, | |
| "how": "first_round", | |
| "first_cand_ok": false, | |
| "rounds": [ | |
| { | |
| "round": 0, | |
| "llm_s": 0.247, | |
| "cands": [ | |
| { | |
| "text": "<ibo>Daalụ kaọdị.</ibo> <eng>Thank you, I’ll see you at six.</eng> <ibo>Ọ dị mma.</ibo>", | |
| "ok": false, | |
| "why": [ | |
| "mix drift tvd=0.46" | |
| ], | |
| "tvd": 0.462 | |
| }, | |
| { | |
| "text": "<ibo>Daalụ kaọdị.</ibo> <eng>See you later!</eng> <ibo>Ọ dị mma.</ibo>", | |
| "ok": true, | |
| "why": [], | |
| "tvd": 0.257 | |
| }, | |
| { | |
| "text": "<ibo>Daalụ kaọdị.</ibo> <eng>Thank you, have a safe trip.</eng>", | |
| "ok": false, | |
| "why": [ | |
| "mix drift tvd=0.63" | |
| ], | |
| "tvd": 0.63 | |
| } | |
| ] | |
| } | |
| ], | |
| "mix_s": 0.0016, | |
| "llm_s": 0.247, | |
| "tts_first_s": 0.864, | |
| "spec_start_s": 0.256, | |
| "compute_to_first_audio_s": 1.179, | |
| "tts_breakdown": { | |
| "fe": 0.0, | |
| "lm": 0.482, | |
| "flow_voc": 0.208, | |
| "n_tok": 31, | |
| "audio_s": 1.24 | |
| }, | |
| "tts_chunks_s": [ | |
| 0.864, | |
| 0.728, | |
| 0.69 | |
| ], | |
| "first_audio": 31.099, | |
| "reply_end": 35.299, | |
| "latency_s": 1.435 | |
| } | |
| ] | |
| } |