Add Listen section: 5 playable stereo demos (user left / agent right), refreshed sample logs
2ea57e9 verified | { | |
| "name": "03_pidgin_bargein", | |
| "desc": "Nigerian Pidgin throughout, with a barge-in. User turns TTS-synthesized (Flash, female voice).", | |
| "user_turns": [ | |
| { | |
| "label": "TTS-synthesized user turn [identical input audio recovered from the 2026-10-24 demo]", | |
| "text": "<pcm>How far? Abeg, wetin I fit cook tonight wey no go take long?</pcm>", | |
| "wav": "/scratch/duplex_v2/user_audio/03_pidgin_bargein_turn0.wav", | |
| "orig_label": "TTS-synthesized user turn" | |
| }, | |
| { | |
| "label": "TTS-synthesized user turn (barge-in 1.2 s into the reply) [identical input audio recovered from the 2026-10-24 demo]", | |
| "text": "<pcm>No be rice o, I no get rice for house.</pcm>", | |
| "barge_at": 1.2, | |
| "wav": "/scratch/duplex_v2/user_audio/03_pidgin_bargein_turn1.wav", | |
| "orig_label": "TTS-synthesized user turn (barge-in 1.2 s into the reply)" | |
| }, | |
| { | |
| "label": "TTS-synthesized user turn [identical input audio recovered from the 2026-10-24 demo]", | |
| "text": "<pcm>Okay, na so. Thank you well well.</pcm>", | |
| "wav": "/scratch/duplex_v2/user_audio/03_pidgin_bargein_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": "<pcm>How far? Abeg, wetin I fit cook tonight wey no go take long?</pcm>", | |
| "user_start": 0.7, | |
| "user_onset": 0.764, | |
| "backchannels": [], | |
| "asr": "how far i beg wetin i fit cook tonight when no go take long", | |
| "asr_s": 0.086, | |
| "eot_wait_s": 0.576, | |
| "speech_end": 4.768, | |
| "target": { | |
| "mix": { | |
| "pcm": 1.0 | |
| }, | |
| "matrix": "pcm", | |
| "pattern": [ | |
| "greeting:pcm" | |
| ] | |
| }, | |
| "target_desc": "Match the user's language mix: Nigerian Pidgin about 100% of the words. The main (matrix) language is Nigerian Pidgin. The user greeted in Nigerian Pidgin: open with a short Nigerian Pidgin greeting. Reply only in Nigerian Pidgin.", | |
| "reply": "<pcm>How far! You fit cook moi moi or amala with egusi soup, e no go pass twenty minutes. You dey need pepper?</pcm>", | |
| "reply_mix": { | |
| "pcm": 1.0 | |
| }, | |
| "tvd": 0.0, | |
| "how": "first_round", | |
| "first_cand_ok": true, | |
| "rounds": [ | |
| { | |
| "round": 0, | |
| "llm_s": 0.232, | |
| "cands": [ | |
| { | |
| "text": "<pcm>How far! You fit cook moi moi or amala with egusi soup, e no go pass twenty minutes. You dey need pepper?</pcm>", | |
| "ok": true, | |
| "why": [], | |
| "tvd": 0.0 | |
| }, | |
| { | |
| "text": "<pcm>How far! You fit cook efo riro with pepper soup, e no go take long. You get pepper for that?</pcm>", | |
| "ok": true, | |
| "why": [], | |
| "tvd": 0.0 | |
| }, | |
| { | |
| "text": "<pcm>How far! You fit cook pepper soup with garri, e no go pass twenty minutes. You get pepper for that?</pcm>", | |
| "ok": true, | |
| "why": [], | |
| "tvd": 0.0 | |
| } | |
| ] | |
| } | |
| ], | |
| "mix_s": 0.0099, | |
| "llm_s": 0.232, | |
| "tts_first_s": 0.662, | |
| "spec_start_s": 0.256, | |
| "compute_to_first_audio_s": 1.003, | |
| "tts_breakdown": { | |
| "fe": 0.0, | |
| "lm": 0.516, | |
| "flow_voc": 0.22, | |
| "n_tok": 33, | |
| "audio_s": 1.32 | |
| }, | |
| "tts_chunks_s": [ | |
| 0.662, | |
| 5.019, | |
| 0.736 | |
| ], | |
| "first_audio": 6.027, | |
| "reply_end": 19.286, | |
| "latency_s": 1.259 | |
| }, | |
| { | |
| "turn": 1, | |
| "user_label": "TTS-synthesized user turn (barge-in 1.2 s into the reply) [identical input audio recovered from the 2026-10-24 demo]", | |
| "user_ref": "<pcm>No be rice o, I no get rice for house.</pcm>", | |
| "user_start": 7.227, | |
| "user_onset": 7.493, | |
| "backchannels": [], | |
| "barge": { | |
| "user_onset": 7.493, | |
| "detected": 7.712, | |
| "agent_stopped": 7.752, | |
| "reaction_s": 0.259, | |
| "agent_chunks_played": 1, | |
| "agent_chunks_total": 3 | |
| }, | |
| "asr": "no be rice oh i no get rice for house", | |
| "asr_s": 0.082, | |
| "eot_wait_s": 0.576, | |
| "speech_end": 11.072, | |
| "target": { | |
| "mix": { | |
| "pcm": 1.0 | |
| }, | |
| "matrix": "pcm", | |
| "pattern": [] | |
| }, | |
| "target_desc": "Match the user's language mix: Nigerian Pidgin about 100% of the words. The main (matrix) language is Nigerian Pidgin. Reply only in Nigerian Pidgin.", | |
| "reply": "<pcm>Abeg, no worry! You fit cook beans with pepper soup—no rice, no problem. Wetin you fit use?</pcm>", | |
| "reply_mix": { | |
| "pcm": 1.0 | |
| }, | |
| "tvd": 0.0, | |
| "how": "first_round", | |
| "first_cand_ok": true, | |
| "rounds": [ | |
| { | |
| "round": 0, | |
| "llm_s": 0.221, | |
| "cands": [ | |
| { | |
| "text": "<pcm>Abeg, no worry! You fit cook beans with pepper soup—no rice, no problem. Wetin you fit use?</pcm>", | |
| "ok": true, | |
| "why": [], | |
| "tvd": 0.0 | |
| }, | |
| { | |
| "text": "<pcm>Abeg, no worry! You fit make garri and beans with pepper soup—no go take more than twenty minutes.</pcm>", | |
| "ok": true, | |
| "why": [], | |
| "tvd": 0.0 | |
| }, | |
| { | |
| "text": "<pcm>No wahala, you fit cook garri with beans, e no go take long and no need rice.</pcm>", | |
| "ok": true, | |
| "why": [], | |
| "tvd": 0.0 | |
| } | |
| ] | |
| } | |
| ], | |
| "mix_s": 0.0069, | |
| "llm_s": 0.221, | |
| "tts_first_s": 1.206, | |
| "spec_start_s": 0.256, | |
| "compute_to_first_audio_s": 1.528, | |
| "tts_breakdown": { | |
| "fe": 0.0, | |
| "lm": 0.755, | |
| "flow_voc": 0.223, | |
| "n_tok": 47, | |
| "audio_s": 1.88 | |
| }, | |
| "tts_chunks_s": [ | |
| 1.206, | |
| 2.214, | |
| 0.978 | |
| ], | |
| "first_audio": 12.856, | |
| "reply_end": 21.776, | |
| "latency_s": 1.784 | |
| }, | |
| { | |
| "turn": 2, | |
| "user_label": "TTS-synthesized user turn [identical input audio recovered from the 2026-10-24 demo]", | |
| "user_ref": "<pcm>Okay, na so. Thank you well well.</pcm>", | |
| "user_start": 22.476, | |
| "user_onset": 22.746, | |
| "backchannels": [], | |
| "asr": "oka na so thank you gbee well", | |
| "asr_s": 0.065, | |
| "eot_wait_s": 0.576, | |
| "speech_end": 26.464, | |
| "target": { | |
| "mix": { | |
| "pcm": 1.0 | |
| }, | |
| "matrix": "pcm", | |
| "pattern": [ | |
| "discourse:pcm" | |
| ] | |
| }, | |
| "target_desc": "Match the user's language mix: Nigerian Pidgin about 100% of the words. The main (matrix) language is Nigerian Pidgin. Sprinkle a Nigerian Pidgin discourse word or two the way the user did. Reply only in Nigerian Pidgin.", | |
| "reply": "<pcm>Na so, thank you well well! I dey glad I fit help. We go see later, no wahala.</pcm>", | |
| "reply_mix": { | |
| "pcm": 1.0 | |
| }, | |
| "tvd": 0.0, | |
| "how": "first_round", | |
| "first_cand_ok": true, | |
| "rounds": [ | |
| { | |
| "round": 0, | |
| "llm_s": 0.204, | |
| "cands": [ | |
| { | |
| "text": "<pcm>Na so, thank you well well! I dey glad I fit help. We go see later, no wahala.</pcm>", | |
| "ok": true, | |
| "why": [], | |
| "tvd": 0.0 | |
| }, | |
| { | |
| "text": "<pcm>Na so! Thank you well well. We go see later, no wahala.</pcm>", | |
| "ok": true, | |
| "why": [], | |
| "tvd": 0.0 | |
| }, | |
| { | |
| "text": "<pcm>Na so! Thank you well well, my friend. We go see later, no wahala.</pcm>", | |
| "ok": true, | |
| "why": [], | |
| "tvd": 0.0 | |
| } | |
| ] | |
| } | |
| ], | |
| "mix_s": 0.0048, | |
| "llm_s": 0.204, | |
| "tts_first_s": 1.289, | |
| "spec_start_s": 0.256, | |
| "compute_to_first_audio_s": 1.576, | |
| "tts_breakdown": { | |
| "fe": 0.0, | |
| "lm": 1.203, | |
| "flow_voc": 0.222, | |
| "n_tok": 73, | |
| "audio_s": 2.92 | |
| }, | |
| "tts_chunks_s": [ | |
| 1.289, | |
| 0.964, | |
| 1.425 | |
| ], | |
| "first_audio": 28.296, | |
| "reply_end": 35.696, | |
| "latency_s": 1.832 | |
| } | |
| ] | |
| } |