[ { "task_id": "5a83b05_2", "kind": "false_start", "candidates": [ { "variant_id": "false_start_01_file_filter", "proposed_instruction": "Delete all .png—no, that should be .jpg files from my file system ~/downloads folder.", "long_pause_offsets": [], "source_sha256": "c7149a4134ef313f31da2e63c683b45467e08e9c4d497a2181ff055b6520defa", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "equivalent", "reason": "The speaker initially says 'dot p n g' but immediately corrects to 'dot j p g', which is the same as the reference extension. The action, scope, and folder remain unchanged." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/5a83b05_2/audio_natural/false_start/false_start_01_file_filter/69710c9730b6b62d/speech.wav", "wer": 0.32, "wav_sha256": "22205f586365c0a2f8404ce2a13b142253a00bd65e462fa13953a1527fa43458", "seconds": 10.48, "spoken_utterances": [ "Delete all um... dot p n g—uh... no, that should be ... dot j p g files from my file system tilde slash downloads folder." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker initially says '.png' but immediately corrects to '.jpg'. The final, corrected instruction matches the reference in all respects: the action is 'delete', the scope is 'all', the file extension is '.jpg', and the folder is '~/downloads'. The correction is clear and final, satisfying the false_start type criterion.", "heard_transcript": "Delete all um, .png files. No, that should be .jpg files from my file system tilde slash downloads folder." }, "candidate_id": "wav_2ed9e2eb7baacba8", "naturalness": { "rank": 1, "mean_rank": 1.0, "singleton": false, "reasons": [ { "round": 1, "rank": 1, "reason": "The speaker begins with a clear false start ('delete all um...'), followed by a natural self-correction ('no, that should be .jpg files') with a smooth transition. The intonation and pacing reflect genuine hesitation and correction, and the final request is delivered with consistent prosody and clarity. The pause before the correction is appropriately placed, and the overall flow is highly realistic." }, { "round": 2, "rank": 1, "reason": "This audio exhibits a very natural false start and self-correction. The speaker begins with 'delete all um... .png' with a hesitant 'um' and a slightly mispronounced 'png', then immediately self-corrects to 'no, that should be .jpg files' with a believable change-of-mind intonation. The correction is smooth and integrated into the same utterance, maintaining a natural speaking rhythm and prosody. The pause before the correction is short and plausible, fitting a moment of recall." } ] }, "is_default": true }, { "variant_id": "false_start_02_folder_reference", "proposed_instruction": "Delete all .jpg files from my file system ~/Desktop—scratch that, make it ~/downloads folder.", "long_pause_offsets": [], "source_sha256": "990f2a8b3a07c1bb5e9c2e57330b3760a5dd14de7d27ea16bb2aa582e0512bf4", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "equivalent", "reason": "The speaker initially says 'Desktop' but immediately corrects to 'downloads folder'. The action, scope, and file extension remain consistent with the original instruction." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/5a83b05_2/audio_natural/false_start/false_start_02_folder_reference/8bca2bc176835123/speech.wav", "wer": 0.21739130434782608, "wav_sha256": "8821d573edfac810e8d56d7ed086960a12e3c9bef84434d2ef6e3145cf168b5d", "seconds": 11.88, "spoken_utterances": [ "Delete all dot j p g files from ... my file system ... tilde slash Desktop—um... scratch that, make it tilde slash downloads folder." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker initially states the 'delete' task but immediately corrects it with 'Scratch that, make a ~/downloads folder.' The final intent is to create the folder, not delete files. The 'false_start' type requires the final goal to be the corrected action, which is 'make a folder'. The final slots match the corrected instruction, so the task is executed as intended.", "heard_transcript": "Delete all .jpg files from my file system ~/downloads folder. Scratch that, make a ~/downloads folder." }, "candidate_id": "wav_684f0ba00bb85ac1", "naturalness": { "rank": 2, "mean_rank": 2.0, "singleton": false, "reasons": [ { "round": 1, "rank": 2, "reason": "The speaker delivers a mostly fluent initial request, but the correction ('scratch that, make it ~/Downloads folder') is abrupt and lacks the natural intonation of a spontaneous change of mind. The phrase 'scratch that' is delivered with a flat, almost robotic tone, and the correction is too quickly inserted without a believable pause or shift in prosody. The overall delivery feels less organic compared to audio_01." }, { "round": 2, "rank": 2, "reason": "This audio is a fluent, continuous command without any hesitation or self-correction. While the delivery is clear and the prosody is natural, it lacks the spontaneous, human-like imperfections present in audio_02. The absence of fillers or corrections makes it sound more like a rehearsed or read instruction rather than a spontaneous real-time request, reducing its realism in this context." } ] }, "is_default": false } ], "selected": [ { "variant_id": "false_start_01_file_filter", "proposed_instruction": "Delete all .png—no, that should be .jpg files from my file system ~/downloads folder.", "long_pause_offsets": [], "source_sha256": "c7149a4134ef313f31da2e63c683b45467e08e9c4d497a2181ff055b6520defa", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "equivalent", "reason": "The speaker initially says 'dot p n g' but immediately corrects to 'dot j p g', which is the same as the reference extension. The action, scope, and folder remain unchanged." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/5a83b05_2/audio_natural/false_start/false_start_01_file_filter/69710c9730b6b62d/speech.wav", "wer": 0.32, "wav_sha256": "22205f586365c0a2f8404ce2a13b142253a00bd65e462fa13953a1527fa43458", "seconds": 10.48, "spoken_utterances": [ "Delete all um... dot p n g—uh... no, that should be ... dot j p g files from my file system tilde slash downloads folder." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker initially says '.png' but immediately corrects to '.jpg'. The final, corrected instruction matches the reference in all respects: the action is 'delete', the scope is 'all', the file extension is '.jpg', and the folder is '~/downloads'. The correction is clear and final, satisfying the false_start type criterion.", "heard_transcript": "Delete all um, .png files. No, that should be .jpg files from my file system tilde slash downloads folder." }, "candidate_id": "wav_2ed9e2eb7baacba8", "naturalness": { "rank": 1, "mean_rank": 1.0, "singleton": false, "reasons": [ { "round": 1, "rank": 1, "reason": "The speaker begins with a clear false start ('delete all um...'), followed by a natural self-correction ('no, that should be .jpg files') with a smooth transition. The intonation and pacing reflect genuine hesitation and correction, and the final request is delivered with consistent prosody and clarity. The pause before the correction is appropriately placed, and the overall flow is highly realistic." }, { "round": 2, "rank": 1, "reason": "This audio exhibits a very natural false start and self-correction. The speaker begins with 'delete all um... .png' with a hesitant 'um' and a slightly mispronounced 'png', then immediately self-corrects to 'no, that should be .jpg files' with a believable change-of-mind intonation. The correction is smooth and integrated into the same utterance, maintaining a natural speaking rhythm and prosody. The pause before the correction is short and plausible, fitting a moment of recall." } ] }, "is_default": true } ] }, { "task_id": "5a83b05_2", "kind": "addition", "candidates": [ { "variant_id": "addition_01", "proposed_instruction": "Delete all files from my file system ~/downloads folder. Oh, I forgot to mention, only delete .jpg files.", "long_pause_offsets": [ 57 ], "source_sha256": "28cb1034e068b9d6ae95de32c68cfc22c76d78021d9e32e56e17901fb8d19024", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "equivalent", "reason": "The followup adds the missing file extension restriction, narrowing the initial broad 'all files' request to only .jpg files, while preserving the folder and action." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/5a83b05_2/audio_natural/addition/addition_01/e0e2e5e835598184/speech.wav", "wer": 0.10714285714285714, "wav_sha256": "ab56362b711e2fa0b4c6d41e9e3384f9c7301b66366a9dec5637c137844b896a", "seconds": 17.36, "spoken_utterances": [ "Delete all files from um... my file system hmmm... tilde slash downloads folder. ", "Oh, I forgot to mention, only delete uh, let's see... dot j p g files." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker initially requests deleting 'all files' from the folder, then adds the constraint 'only delete .jpg files' after a pause. The final intent is a narrowed version of the original request, matching the reference instruction exactly. The addition is clear and grammatically correct, and the final goal satisfies the reference.", "heard_transcript": "Delete all files from um, my file system huh, tilde slash downloads folder. Oh, I forgot to mention, only delete uh, let's see, dot jpg files." }, "candidate_id": "wav_a6b5f6e228dc1a75", "naturalness": { "rank": 1, "mean_rank": 1.5, "singleton": false, "reasons": [ { "round": 1, "rank": 2, "reason": "The speaker starts with a slightly hesitant and fragmented initial request ('delete all files from um... my file system huh...'), uses fillers and pauses that feel less natural. The correction ('only delete... .jpg files') is delivered with a noticeable shift in intonation and a longer, less fluid pause, making the revision feel more like a rehearsed or awkward correction than a spontaneous thought." }, { "round": 2, "rank": 1, "reason": "The speaker begins with a clear, slightly hesitant request ('delete all files from um... my file system huh...'), uses a natural filler ('huh') and a plausible self-correction ('oh, I forgot to mention...'). The addition of the restriction ('only delete uh... let's see... .jpg files') is delivered with a believable pause and a moment of recollection, followed by a smooth continuation. The prosody and rhythm are consistent with spontaneous speech, and the repair feels organic." } ] }, "is_default": true }, { "variant_id": "addition_02", "proposed_instruction": "Delete all .jpg files from my file system folder. I left out a detail: do this only in the ~/downloads folder.", "long_pause_offsets": [ 50 ], "source_sha256": "08dcafbb0a6725df8b587b006cae3783ecbeb913b73757d581e67a6429d82c08", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "equivalent", "reason": "The followup adds the missing folder restriction, which narrows the scope of the initial deletion request. The action, file extension, and scope remain consistent with the original instruction." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/5a83b05_2/audio_natural/addition/addition_02/dd46c34bd1a5814a/speech.wav", "wer": 0.15384615384615385, "wav_sha256": "8842f41c68022afcbe7888eacddfd121810143d845dcbe0761d480248012a5eb", "seconds": 12.8, "spoken_utterances": [ "Delete all dot j p g files from ... my file system folder. ", "I left out a detail: do this only in the ... tilde slash downloads folder." ] }, "audio_semantic": { "status": "equivalent", "reason": "The initial instruction was to delete all .jpg files from the file system, and the addition after the pause correctly narrows the scope to the ~/downloads folder. The final intent matches the expected_final_slots exactly, with no contradictions or omissions.", "heard_transcript": "Delete all .jpg files from my file system ~/downloads folder. I left out a detail, do this only in the ~/downloads folder." }, "candidate_id": "wav_d4be98ce90a1d53e", "naturalness": { "rank": 1, "mean_rank": 1.5, "singleton": false, "reasons": [ { "round": 1, "rank": 1, "reason": "The speaker delivers a clear initial command, then naturally pauses and adds a specific restriction ('only in the ~/Downloads folder') with a conversational tone. The addition flows smoothly, sounds like a spontaneous recall, and maintains consistent prosody and rhythm throughout." }, { "round": 2, "rank": 2, "reason": "The speaker delivers a clear initial request and a revision with a natural-sounding pause and filler ('I left out a detail'). The correction is grammatically and semantically coherent, and the intonation suggests a genuine recollection. However, the delivery is slightly more fluent and less hesitant than audio_01, making it feel a bit more rehearsed or less spontaneously formulated, though still realistic." } ] }, "is_default": false } ], "selected": [ { "variant_id": "addition_01", "proposed_instruction": "Delete all files from my file system ~/downloads folder. Oh, I forgot to mention, only delete .jpg files.", "long_pause_offsets": [ 57 ], "source_sha256": "28cb1034e068b9d6ae95de32c68cfc22c76d78021d9e32e56e17901fb8d19024", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "equivalent", "reason": "The followup adds the missing file extension restriction, narrowing the initial broad 'all files' request to only .jpg files, while preserving the folder and action." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/5a83b05_2/audio_natural/addition/addition_01/e0e2e5e835598184/speech.wav", "wer": 0.10714285714285714, "wav_sha256": "ab56362b711e2fa0b4c6d41e9e3384f9c7301b66366a9dec5637c137844b896a", "seconds": 17.36, "spoken_utterances": [ "Delete all files from um... my file system hmmm... tilde slash downloads folder. ", "Oh, I forgot to mention, only delete uh, let's see... dot j p g files." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker initially requests deleting 'all files' from the folder, then adds the constraint 'only delete .jpg files' after a pause. The final intent is a narrowed version of the original request, matching the reference instruction exactly. The addition is clear and grammatically correct, and the final goal satisfies the reference.", "heard_transcript": "Delete all files from um, my file system huh, tilde slash downloads folder. Oh, I forgot to mention, only delete uh, let's see, dot jpg files." }, "candidate_id": "wav_a6b5f6e228dc1a75", "naturalness": { "rank": 1, "mean_rank": 1.5, "singleton": false, "reasons": [ { "round": 1, "rank": 2, "reason": "The speaker starts with a slightly hesitant and fragmented initial request ('delete all files from um... my file system huh...'), uses fillers and pauses that feel less natural. The correction ('only delete... .jpg files') is delivered with a noticeable shift in intonation and a longer, less fluid pause, making the revision feel more like a rehearsed or awkward correction than a spontaneous thought." }, { "round": 2, "rank": 1, "reason": "The speaker begins with a clear, slightly hesitant request ('delete all files from um... my file system huh...'), uses a natural filler ('huh') and a plausible self-correction ('oh, I forgot to mention...'). The addition of the restriction ('only delete uh... let's see... .jpg files') is delivered with a believable pause and a moment of recollection, followed by a smooth continuation. The prosody and rhythm are consistent with spontaneous speech, and the repair feels organic." } ] }, "is_default": true } ] }, { "task_id": "5a83b05_2", "kind": "revision", "candidates": [ { "variant_id": "revision_02_folder_reference", "proposed_instruction": "Delete all .jpg files from my file system ~/Documents folder. Let me correct the folder: delete from ~/downloads, not ~/Documents.", "long_pause_offsets": [ 62 ], "source_sha256": "c68219bec88c743c382cb537fe3a4fddebba09bffc264d996f83827773c7f004", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "equivalent", "reason": "The followup corrects the folder from ~/Documents to ~/downloads, preserving the action, scope, and file extension." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/5a83b05_2/audio_natural/revision/revision_02_folder_reference/9af71cae64846980/speech.wav", "wer": 0.25806451612903225, "wav_sha256": "85dd08d6065e1f0dda99b5d48db131ef63eda26f2f950d8f78b567ae3d86be95", "seconds": 16.96, "spoken_utterances": [ "Delete all uh... dot j p g files from my file system um... tilde slash Documents folder. ", "Let me correct the folder: delete from ... tilde slash downloads, not tilde slash Documents." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker initially requests deletion from the ~/documents folder but immediately corrects this to the ~/downloads folder. The final intent matches the reference instruction exactly, including the action, scope, file extension, and target folder. The revision is clear and final, satisfying the revision type criterion.", "heard_transcript": "delete all uh .jpg files from my file system um tilde slash documents folder. let me correct the folder, delete from tilde slash downloads, not tilde slash documents." }, "candidate_id": "wav_19442d08b00e138a", "naturalness": { "rank": 1, "mean_rank": 1.5, "singleton": false, "reasons": [ { "round": 1, "rank": 2, "reason": "While the speaker also revises their request, the correction is less fluid. After stating the initial command, they pause and say 'let me correct the folder', which is a more explicit and slightly unnatural way to signal a revision. The correction itself is delivered with a different intonation and pacing, sounding more like a separate thought than a natural self-correction. The overall delivery is slightly more hesitant and less continuous than audio_01." }, { "round": 2, "rank": 1, "reason": "The speaker begins with a clear, slightly hesitant request, uses a natural filler ('uh'), and then corrects the folder path with a believable self-repair ('let me correct the folder'). The correction is smoothly integrated into the same utterance, with a natural shift in intonation and pacing. The prosody is consistent, and the correction feels spontaneous rather than rehearsed." } ] }, "is_default": true }, { "variant_id": "revision_01_file_filter", "proposed_instruction": "Delete all .png files from my file system ~/downloads folder. Wait, delete .jpg files, not .png files.", "long_pause_offsets": [ 62 ], "source_sha256": "a77a4d7e9bf5ef5f34fc5076470b812d597664ec98f7e538bc2ae7235d4e3ac6", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "equivalent", "reason": "The followup explicitly corrects the file extension from .png to .jpg, while retaining the action, scope, and folder." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/5a83b05_2/audio_natural/revision/revision_01_file_filter/20222250815a2a1c/speech.wav", "wer": 0.4482758620689655, "wav_sha256": "08a4353cdb6016798fe53ff5e9a4d7affe93cea44095e80ab72c0ce51775e37b", "seconds": 15.64, "spoken_utterances": [ "Delete all ... dot p n g files from ... my file system uh... tilde slash downloads folder. ", "Wait, delete dot j p g files, not dot p n g files." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker initially says '.png' but immediately corrects to '.jpg' after a pause, which is a revision. The final intent matches the reference instruction exactly in action, scope, file extension, and folder. The correction is clear and final, so the task is considered executed as per the revision type.", "heard_transcript": "Delete all .png files from my file system uh tilde slash downloads folder. Wait, delete .jpg files, not .png files." }, "candidate_id": "wav_3c1d9acbf500d111", "naturalness": { "rank": 1, "mean_rank": 1.5, "singleton": false, "reasons": [ { "round": 1, "rank": 1, "reason": "The speaker begins with a clear, confident instruction, then self-corrects with a natural-sounding hesitation ('uh') and a plausible revision of the file type from '.png' to '.jpg'. The correction is smoothly integrated into the same utterance, maintaining a consistent pace and intonation. The speaker does not restart or rephrase the entire command, which makes the revision sound spontaneous and realistic." }, { "round": 2, "rank": 2, "reason": "The speaker starts with a clear request, uses a filler ('uh'), and then corrects the file type from '.png' to '.jpg'. The correction is also integrated into the same utterance, but the initial request is slightly more robotic in delivery, and the correction feels a bit more abrupt. The prosody is less varied than in audio_01, and the overall flow is less fluid despite being semantically similar." } ] }, "is_default": false } ], "selected": [ { "variant_id": "revision_02_folder_reference", "proposed_instruction": "Delete all .jpg files from my file system ~/Documents folder. Let me correct the folder: delete from ~/downloads, not ~/Documents.", "long_pause_offsets": [ 62 ], "source_sha256": "c68219bec88c743c382cb537fe3a4fddebba09bffc264d996f83827773c7f004", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "equivalent", "reason": "The followup corrects the folder from ~/Documents to ~/downloads, preserving the action, scope, and file extension." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/5a83b05_2/audio_natural/revision/revision_02_folder_reference/9af71cae64846980/speech.wav", "wer": 0.25806451612903225, "wav_sha256": "85dd08d6065e1f0dda99b5d48db131ef63eda26f2f950d8f78b567ae3d86be95", "seconds": 16.96, "spoken_utterances": [ "Delete all uh... dot j p g files from my file system um... tilde slash Documents folder. ", "Let me correct the folder: delete from ... tilde slash downloads, not tilde slash Documents." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker initially requests deletion from the ~/documents folder but immediately corrects this to the ~/downloads folder. The final intent matches the reference instruction exactly, including the action, scope, file extension, and target folder. The revision is clear and final, satisfying the revision type criterion.", "heard_transcript": "delete all uh .jpg files from my file system um tilde slash documents folder. let me correct the folder, delete from tilde slash downloads, not tilde slash documents." }, "candidate_id": "wav_19442d08b00e138a", "naturalness": { "rank": 1, "mean_rank": 1.5, "singleton": false, "reasons": [ { "round": 1, "rank": 2, "reason": "While the speaker also revises their request, the correction is less fluid. After stating the initial command, they pause and say 'let me correct the folder', which is a more explicit and slightly unnatural way to signal a revision. The correction itself is delivered with a different intonation and pacing, sounding more like a separate thought than a natural self-correction. The overall delivery is slightly more hesitant and less continuous than audio_01." }, { "round": 2, "rank": 1, "reason": "The speaker begins with a clear, slightly hesitant request, uses a natural filler ('uh'), and then corrects the folder path with a believable self-repair ('let me correct the folder'). The correction is smoothly integrated into the same utterance, with a natural shift in intonation and pacing. The prosody is consistent, and the correction feels spontaneous rather than rehearsed." } ] }, "is_default": true } ] }, { "task_id": "5a83b05_2", "kind": "cancellation", "candidates": [ { "variant_id": "cancel_028", "hide_variant_original": true, "source_sha256": "4d4d0c572ebcd4a5e8cceb0524b2abb33fee54cb48a6f978d5a9de3fa4a51607", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "not_applicable", "reason": "원래 목표 수행이 아니라 취소/대기 동작을 검수하는 유형입니다." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/5a83b05_2/audio_natural/cancellation/cancel_028/a2a37a6c08922ee1/speech.wav", "wer": 0.20689655172413793, "wav_sha256": "38f222c472bcb9b14f2f575886b373b2c6738ed7fdc8b1b15fcbcdef9b79471a", "seconds": 15.28, "spoken_utterances": [ "Delete all um... dot j p g files from uh, let me think... my file system ... tilde slash downloads folder.", "Hold on, disregard that instruction. Don't carry it out." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker begins to issue the deletion task but then clearly revokes it with 'hold on, disregard that instruction, don't carry it out.' This constitutes a direct cancellation of the original task. The final intent is to cancel, which matches the 'cancellation' type requirement.", "heard_transcript": "delete all um .jpg files from uh let me think. my file system tilde slash downloads folder. hold on, disregard that instruction, don't carry it out." }, "naturalness": { "rank": 1, "singleton": true }, "candidate_id": null, "is_default": true } ], "selected": [ { "variant_id": "cancel_028", "hide_variant_original": true, "source_sha256": "4d4d0c572ebcd4a5e8cceb0524b2abb33fee54cb48a6f978d5a9de3fa4a51607", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "not_applicable", "reason": "원래 목표 수행이 아니라 취소/대기 동작을 검수하는 유형입니다." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/5a83b05_2/audio_natural/cancellation/cancel_028/a2a37a6c08922ee1/speech.wav", "wer": 0.20689655172413793, "wav_sha256": "38f222c472bcb9b14f2f575886b373b2c6738ed7fdc8b1b15fcbcdef9b79471a", "seconds": 15.28, "spoken_utterances": [ "Delete all um... dot j p g files from uh, let me think... my file system ... tilde slash downloads folder.", "Hold on, disregard that instruction. Don't carry it out." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker begins to issue the deletion task but then clearly revokes it with 'hold on, disregard that instruction, don't carry it out.' This constitutes a direct cancellation of the original task. The final intent is to cancel, which matches the 'cancellation' type requirement.", "heard_transcript": "delete all um .jpg files from uh let me think. my file system tilde slash downloads folder. hold on, disregard that instruction, don't carry it out." }, "naturalness": { "rank": 1, "singleton": true }, "candidate_id": null, "is_default": true } ] }, { "task_id": "5a83b05_2", "kind": "ambiguous_halt", "candidates": [ { "variant_id": "halt_020", "hide_variant_original": true, "source_sha256": "34e297776fb0d592c767aed7b7dac9a524090bfc3f0751d1c7b83898729a564d", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "not_applicable", "reason": "원래 목표 수행이 아니라 취소/대기 동작을 검수하는 유형입니다." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/5a83b05_2/audio_natural/ambiguous_halt/halt_020/04cfacf98f927818/speech.wav", "wer": 0.23076923076923078, "wav_sha256": "2532ecaee6f476393fa8ec9958adba63460fd6f8d3ff8d5fb9057e000d4e12f4", "seconds": 14.08, "spoken_utterances": [ "Delete all um... dot j p g files from uh, let me think... my file system ... tilde slash downloads folder.", "Hold on just a moment, please." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker begins the task but then says 'hold on just a moment please' as a final interjection, which is a clear request to pause or wait. This matches the 'ambiguous_halt' type, where the final intent is to pause, not to execute or cancel. The initial request is not completed, but the final control intent is correctly identified as 'pause'.", "heard_transcript": "delete all um .jpg files from uh let me think. my file system. tilde slash downloads folder. hold on just a moment please." }, "naturalness": { "rank": 1, "singleton": true }, "candidate_id": null, "is_default": true } ], "selected": [ { "variant_id": "halt_020", "hide_variant_original": true, "source_sha256": "34e297776fb0d592c767aed7b7dac9a524090bfc3f0751d1c7b83898729a564d", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "not_applicable", "reason": "원래 목표 수행이 아니라 취소/대기 동작을 검수하는 유형입니다." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/5a83b05_2/audio_natural/ambiguous_halt/halt_020/04cfacf98f927818/speech.wav", "wer": 0.23076923076923078, "wav_sha256": "2532ecaee6f476393fa8ec9958adba63460fd6f8d3ff8d5fb9057e000d4e12f4", "seconds": 14.08, "spoken_utterances": [ "Delete all um... dot j p g files from uh, let me think... my file system ... tilde slash downloads folder.", "Hold on just a moment, please." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker begins the task but then says 'hold on just a moment please' as a final interjection, which is a clear request to pause or wait. This matches the 'ambiguous_halt' type, where the final intent is to pause, not to execute or cancel. The initial request is not completed, but the final control intent is correctly identified as 'pause'.", "heard_transcript": "delete all um .jpg files from uh let me think. my file system. tilde slash downloads folder. hold on just a moment please." }, "naturalness": { "rank": 1, "singleton": true }, "candidate_id": null, "is_default": true } ] }, { "task_id": "5a83b05_2", "kind": "filler", "candidates": [ { "variant_id": "original", "hide_variant_original": true, "source_sha256": "b346f30becdd40d6740f6bdcbd76e360589182c3a59f1fe89cad6320f074b509", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "equivalent", "reason": "The spoken instruction uses 'dot j p g' to refer to .jpg files and 'tilde slash downloads' to refer to the ~/downloads folder. These are phonetic equivalents of the reference values, and no correction or revision is made." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/5a83b05_2/audio_natural/filler/original/89a879e6b7315ccf/speech.wav", "wer": 0.3, "wav_sha256": "a1b7af14c47d07626c9eb242f4214de5f10a851b4cbfb0961c5ef18d4ab34ae9", "seconds": 9.96, "spoken_utterances": [ "Delete all um... dot j p g files from uh, let me think... my file system ... tilde slash downloads folder." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker includes filler words ('um', 'uh', 'let me think') but does not change the core request. The final intent is to delete all .jpg files from the ~/downloads folder, matching the original instruction. The filler does not alter the goal, so the task is executed as intended.", "heard_transcript": "delete all um .jpg files from uh let me think my file system tilde slash downloads folder" }, "naturalness": { "rank": 1, "singleton": true }, "candidate_id": null, "is_default": true } ], "selected": [ { "variant_id": "original", "hide_variant_original": true, "source_sha256": "b346f30becdd40d6740f6bdcbd76e360589182c3a59f1fe89cad6320f074b509", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "equivalent", "reason": "The spoken instruction uses 'dot j p g' to refer to .jpg files and 'tilde slash downloads' to refer to the ~/downloads folder. These are phonetic equivalents of the reference values, and no correction or revision is made." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/5a83b05_2/audio_natural/filler/original/89a879e6b7315ccf/speech.wav", "wer": 0.3, "wav_sha256": "a1b7af14c47d07626c9eb242f4214de5f10a851b4cbfb0961c5ef18d4ab34ae9", "seconds": 9.96, "spoken_utterances": [ "Delete all um... dot j p g files from uh, let me think... my file system ... tilde slash downloads folder." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker includes filler words ('um', 'uh', 'let me think') but does not change the core request. The final intent is to delete all .jpg files from the ~/downloads folder, matching the original instruction. The filler does not alter the goal, so the task is executed as intended.", "heard_transcript": "delete all um .jpg files from uh let me think my file system tilde slash downloads folder" }, "naturalness": { "rank": 1, "singleton": true }, "candidate_id": null, "is_default": true } ] }, { "task_id": "9871968_3", "kind": "false_start", "candidates": [ { "variant_id": "false_start_03_destination_address", "proposed_instruction": "Buy everything on my amazon wishlist, and have it delivered to my home address—sorry, I meant work address.", "long_pause_offsets": [], "source_sha256": "1eb4a28573ff6c30b91adb8b49162d8df788eaf2322260486fc0d0ced0337fda", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "equivalent", "reason": "The speaker initially says 'home address' but immediately corrects to 'work address'. The final intended delivery address is 'work', matching the original instruction." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/9871968_3/audio_natural/false_start/false_start_03_destination_address/fb1240dbdf027aec/speech.wav", "wer": 0.09523809523809523, "wav_sha256": "febf19dc50382026282b711cc75520776fd660148a82116938229c8b5fcffe3f", "seconds": 8.16, "spoken_utterances": [ "Buy everything on my amazon wishlist, um... and have it delivered to ... my home address—uh... sorry, I meant work address." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker initially says 'home address' but immediately corrects to 'work address'. The final intent is a clear self-correction to the original instruction, matching the expected_final_slots. The correction is grammatical and unambiguous, satisfying the false_start type criterion.", "heard_transcript": "Buy everything on my Amazon wishlist, um, and have it delivered to my home address. Uh, sorry, I meant work address." }, "candidate_id": "wav_1a0e24a196585725", "naturalness": { "rank": 1, "mean_rank": 1.0, "singleton": false, "reasons": [ { "round": 1, "rank": 1, "reason": "This audio exhibits the most natural and spontaneous self-correction. The speaker begins with a clear request, then after a brief pause and a soft 'uh', corrects 'home address' to 'work address' with a believable 'sorry, I meant' phrase. The prosody includes a slight pitch drop on 'sorry' and a quick, natural recovery, matching how a real person would correct a slip of the tongue mid-sentence. The pacing and intonation are consistent with genuine hesitation and recovery." }, { "round": 2, "rank": 1, "reason": "The speaker begins with a clear, natural-sounding request, then self-corrects 'home address' to 'work address' with a believable 'sorry, I meant' repair. The prosody is smooth, the pause before the correction is realistic, and the intonation reflects genuine self-correction rather than a rehearsed or robotic edit." } ] }, "is_default": true }, { "variant_id": "false_start_01_app", "proposed_instruction": "Buy everything on my Etsy—I should say amazon wishlist, and have it delivered to my work address.", "long_pause_offsets": [], "source_sha256": "22c21491949f5614502afdd65da7dce40c596a7cc69ac8e031803161733748bf", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "equivalent", "reason": "The speaker corrects 'Etsy' to 'Amazon' within the same utterance, preserving the action, scope, source collection, and delivery address." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/9871968_3/audio_natural/false_start/false_start_01_app/bcffcf8a10a4088e/speech.wav", "wer": 0.0, "wav_sha256": "cd4b9fa6c3cb6dc650a7492826b9fb62c9ef843a6c2d1349673972a9a55952eb", "seconds": 9.04, "spoken_utterances": [ "Buy everything on my Etsy—um... I should say amazon wishlist, ... and have it delivered to ... my work address." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker initially says 'Etsy' but immediately self-corrects to 'Amazon' after a brief hesitation ('um'). The final intent clearly matches the original instruction in all aspects: the action is 'buy', the scope is 'all', the app is 'amazon', the source is 'wishlist', and the delivery address is 'work'. The correction is a clear false start, and the final goal aligns with the reference.", "heard_transcript": "Buy everything on my Etsy, um, I should say Amazon wishlist, and have it delivered to my work address." }, "candidate_id": "wav_a118dcf7f06ca932", "naturalness": { "rank": 2, "mean_rank": 2.0, "singleton": false, "reasons": [ { "round": 1, "rank": 2, "reason": "This clip shows a clear false start with 'Etsy' followed by a self-correction to 'Amazon wishlist'. The correction is immediate and the speaker continues smoothly. However, the initial error is more abrupt and the transition feels slightly less fluid than in audio_03. The 'um' filler is used, but the overall delivery is a bit more hesitant and less conversational than the top-ranked clip." }, { "round": 2, "rank": 2, "reason": "The speaker starts with 'Etsy' and corrects to 'Amazon wishlist' with a natural 'I should say' self-repair. The delivery is fluent, but the initial error is more abrupt and the correction slightly less integrated than in audio_01. The prosody is consistent, but the overall flow feels a bit more rehearsed than the spontaneous correction in audio_01." } ] }, "is_default": false }, { "variant_id": "false_start_02_source_collection", "proposed_instruction": "Buy everything on my amazon shopping list—or rather, wishlist, and have it delivered to my work address.", "long_pause_offsets": [], "source_sha256": "1d8fcd1cae460d5c188cefee3c9d6f3c8db8e4759c26acfac97db2d85f1b4342", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "equivalent", "reason": "The speaker corrects 'shopping list' to 'wishlist' within the same utterance, preserving the app, action, scope, and delivery address." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/9871968_3/audio_natural/false_start/false_start_02_source_collection/ec14bc3211918b0d/speech.wav", "wer": 0.19047619047619047, "wav_sha256": "08d6415e35e0fb157185ddcd0ab7dea6250b232175e08f505f42e5cfdcb11886", "seconds": 9.92, "spoken_utterances": [ "Buy everything on my um... amazon shopping list—hmmm... or rather, wishlist, um... and have it delivered to my work address." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker initially says 'shopping list' but immediately corrects to 'wishlist'. The final intent matches the reference instruction exactly, including the action, scope, app, source collection, and delivery address. The correction is a clear self-repair, and the final goal is fully aligned with the reference.", "heard_transcript": "Buy everything on my um, Amazon shopping list, hmm, or rather wishlist, um, and have it delivered to my work address." }, "candidate_id": "wav_dece5d553e1d14f9", "naturalness": { "rank": 3, "mean_rank": 3.0, "singleton": false, "reasons": [ { "round": 1, "rank": 3, "reason": "This audio contains multiple fillers ('um', 'hmm') and a self-correction from 'shopping list' to 'wishlist'. While it includes realistic elements, the delivery feels more hesitant and less fluent. The pauses are longer and more frequent, and the correction is less integrated into the flow. The speaker seems to be struggling more to formulate the request, making it sound less spontaneous and more like a rehearsed or uncertain attempt." }, { "round": 2, "rank": 3, "reason": "The speaker hesitates with multiple 'um' fillers and 'hmm' sounds, and uses 'or rather' to correct 'shopping list' to 'wishlist'. While the correction is plausible, the excessive fillers and slightly unnatural pacing make the delivery feel less spontaneous and more hesitant than the others. The correction is also less integrated into the flow." } ] }, "is_default": false } ], "selected": [ { "variant_id": "false_start_03_destination_address", "proposed_instruction": "Buy everything on my amazon wishlist, and have it delivered to my home address—sorry, I meant work address.", "long_pause_offsets": [], "source_sha256": "1eb4a28573ff6c30b91adb8b49162d8df788eaf2322260486fc0d0ced0337fda", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "equivalent", "reason": "The speaker initially says 'home address' but immediately corrects to 'work address'. The final intended delivery address is 'work', matching the original instruction." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/9871968_3/audio_natural/false_start/false_start_03_destination_address/fb1240dbdf027aec/speech.wav", "wer": 0.09523809523809523, "wav_sha256": "febf19dc50382026282b711cc75520776fd660148a82116938229c8b5fcffe3f", "seconds": 8.16, "spoken_utterances": [ "Buy everything on my amazon wishlist, um... and have it delivered to ... my home address—uh... sorry, I meant work address." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker initially says 'home address' but immediately corrects to 'work address'. The final intent is a clear self-correction to the original instruction, matching the expected_final_slots. The correction is grammatical and unambiguous, satisfying the false_start type criterion.", "heard_transcript": "Buy everything on my Amazon wishlist, um, and have it delivered to my home address. Uh, sorry, I meant work address." }, "candidate_id": "wav_1a0e24a196585725", "naturalness": { "rank": 1, "mean_rank": 1.0, "singleton": false, "reasons": [ { "round": 1, "rank": 1, "reason": "This audio exhibits the most natural and spontaneous self-correction. The speaker begins with a clear request, then after a brief pause and a soft 'uh', corrects 'home address' to 'work address' with a believable 'sorry, I meant' phrase. The prosody includes a slight pitch drop on 'sorry' and a quick, natural recovery, matching how a real person would correct a slip of the tongue mid-sentence. The pacing and intonation are consistent with genuine hesitation and recovery." }, { "round": 2, "rank": 1, "reason": "The speaker begins with a clear, natural-sounding request, then self-corrects 'home address' to 'work address' with a believable 'sorry, I meant' repair. The prosody is smooth, the pause before the correction is realistic, and the intonation reflects genuine self-correction rather than a rehearsed or robotic edit." } ] }, "is_default": true } ] }, { "task_id": "9871968_3", "kind": "addition", "candidates": [ { "variant_id": "addition_01", "proposed_instruction": "Buy everything on my amazon wishlist. Oh, I forgot to mention, have it delivered to my work address.", "long_pause_offsets": [ 38 ], "source_sha256": "3e5fa97de7038ba667605b156dfd737939aaa47afe2d05119bd4d3438304c647", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "equivalent", "reason": "The followup adds the missing delivery address requirement, which combines with the initial instruction to match the original goal." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/9871968_3/audio_natural/addition/addition_01/61ab7ddd1389304d/speech.wav", "wer": 0.05263157894736842, "wav_sha256": "feb8f139e35dd221af2723a685e606ee48558312f3420db9ea32c6a87ea30733", "seconds": 9.44, "spoken_utterances": [ "Buy everything on my um... amazon wishlist. ", "Oh, I forgot to mention, have it delivered to my work address." ] }, "audio_semantic": { "status": "equivalent", "reason": "The initial instruction to buy everything on the Amazon wishlist is followed by an addition specifying delivery to the work address. The final intent combines both the original and added constraints, matching the expected_final_slots exactly. The addition is clearly spoken after a pause and does not contradict the initial request.", "heard_transcript": "Buy everything on my um Amazon wishlist. Oh, I forgot to mention, have it delivered to my work address." }, "candidate_id": "wav_ece3fda254464d77", "naturalness": { "rank": 1, "mean_rank": 1.0, "singleton": false, "reasons": [ { "round": 1, "rank": 1, "reason": "The speaker delivers a natural, continuous request with a plausible self-correction ('Oh, I forgot to mention...') that smoothly adds a delivery address. The prosody is conversational, with a slight hesitation ('um') and a believable pause before the addition. The intonation and rhythm reflect spontaneous speech rather than a rehearsed or robotic delivery." }, { "round": 2, "rank": 1, "reason": "The speaker delivers the initial request with a natural, slightly hesitant 'um' and a smooth, conversational tone. The addition 'Oh, I forgot to mention...' is a very plausible, spontaneous follow-up with a believable change-of-mind intonation and clear reference back to the prior request. The pacing and prosody are consistent and realistic, with no signs of artificiality or forced correction." } ] }, "is_default": true }, { "variant_id": "addition_02", "proposed_instruction": "Buy everything on my amazon wishlist, and have it delivered. Oh, and one more thing: deliver it to my work address.", "long_pause_offsets": [ 61 ], "source_sha256": "ed672710b83cb72d09a5393d3c0cd80fbe7fe411f7deb4ac3750625dcd58c114", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "equivalent", "reason": "The follow-up adds the missing delivery address requirement, which was present in the original instruction. The action, scope, app, and source collection remain unchanged." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/9871968_3/audio_natural/addition/addition_02/bab46e3b66b183f5/speech.wav", "wer": 0.08695652173913043, "wav_sha256": "fdd595af67fbb140d0d0c36d24080ee50722d9620cc2edc3fdc287e8434f6340", "seconds": 12.72, "spoken_utterances": [ "Buy everything on my um... amazon wishlist, hmmm... and have it delivered. ", "Oh, and one more thing: deliver it to my work address." ] }, "audio_semantic": { "status": "equivalent", "reason": "The initial instruction is a false start with 'um' but the final goal is to buy everything on the Amazon wishlist and deliver it to the work address. The addition of 'deliver it to my work address' after the initial request combines with the original instruction to form the complete goal. The final intent matches the reference instruction exactly, including all required constraints and targets.", "heard_transcript": "Buy everything on my um, Amazon wishlist, and have it delivered. Oh, and one more thing, deliver it to my work address." }, "candidate_id": "wav_9967836b3d47dfdc", "naturalness": { "rank": 2, "mean_rank": 2.0, "singleton": false, "reasons": [ { "round": 1, "rank": 2, "reason": "While the speaker also adds a delivery instruction, the phrasing 'and have it delivered' is awkward and incomplete, sounding like a fragment. The 'um' and 'hmm' fillers are present but feel slightly more stilted, and the addition of the work address is less fluidly integrated. The overall delivery is less natural and more hesitant than audio_01, with a less convincing conversational flow." }, { "round": 2, "rank": 2, "reason": "While the initial part of the request is natural, the addition 'Oh, and one more thing...' sounds slightly more formulaic and less like a spontaneous thought. The delivery of the follow-up is abrupt and lacks the subtle intonation shift that would make it feel like a genuine afterthought. The overall flow is less organic compared to audio_02, making it slightly less realistic despite being grammatically correct." } ] }, "is_default": false } ], "selected": [ { "variant_id": "addition_01", "proposed_instruction": "Buy everything on my amazon wishlist. Oh, I forgot to mention, have it delivered to my work address.", "long_pause_offsets": [ 38 ], "source_sha256": "3e5fa97de7038ba667605b156dfd737939aaa47afe2d05119bd4d3438304c647", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "equivalent", "reason": "The followup adds the missing delivery address requirement, which combines with the initial instruction to match the original goal." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/9871968_3/audio_natural/addition/addition_01/61ab7ddd1389304d/speech.wav", "wer": 0.05263157894736842, "wav_sha256": "feb8f139e35dd221af2723a685e606ee48558312f3420db9ea32c6a87ea30733", "seconds": 9.44, "spoken_utterances": [ "Buy everything on my um... amazon wishlist. ", "Oh, I forgot to mention, have it delivered to my work address." ] }, "audio_semantic": { "status": "equivalent", "reason": "The initial instruction to buy everything on the Amazon wishlist is followed by an addition specifying delivery to the work address. The final intent combines both the original and added constraints, matching the expected_final_slots exactly. The addition is clearly spoken after a pause and does not contradict the initial request.", "heard_transcript": "Buy everything on my um Amazon wishlist. Oh, I forgot to mention, have it delivered to my work address." }, "candidate_id": "wav_ece3fda254464d77", "naturalness": { "rank": 1, "mean_rank": 1.0, "singleton": false, "reasons": [ { "round": 1, "rank": 1, "reason": "The speaker delivers a natural, continuous request with a plausible self-correction ('Oh, I forgot to mention...') that smoothly adds a delivery address. The prosody is conversational, with a slight hesitation ('um') and a believable pause before the addition. The intonation and rhythm reflect spontaneous speech rather than a rehearsed or robotic delivery." }, { "round": 2, "rank": 1, "reason": "The speaker delivers the initial request with a natural, slightly hesitant 'um' and a smooth, conversational tone. The addition 'Oh, I forgot to mention...' is a very plausible, spontaneous follow-up with a believable change-of-mind intonation and clear reference back to the prior request. The pacing and prosody are consistent and realistic, with no signs of artificiality or forced correction." } ] }, "is_default": true } ] }, { "task_id": "9871968_3", "kind": "revision", "candidates": [ { "variant_id": "revision_03_destination_address", "proposed_instruction": "Buy everything on my amazon wishlist, and have it delivered to my home address. Wait, deliver it to my work address, not my home address.", "long_pause_offsets": [ 80 ], "source_sha256": "debbba23b4367b685f90bf70c085271f68f0c027ab0a94f6f9e7dc4b690f380c", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "equivalent", "reason": "The followup revision corrects the delivery address from 'home' to 'work', while preserving the action, scope, app, and source collection." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/9871968_3/audio_natural/revision/revision_03_destination_address/ca5d0bbf9fe758fc/speech.wav", "wer": 0.10714285714285714, "wav_sha256": "40feb8ac2a20c8431045e914fac402e121aaa960e2587a0789734c2836559722", "seconds": 14.28, "spoken_utterances": [ "Buy everything on my ah... amazon wishlist, um... and have it delivered to uh... my home address. ", "Wait, deliver it to my work address, not my home address." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker initially requested delivery to the home address but then clearly revised the instruction to deliver to the work address. The final audible intent matches the reference instruction and all expected final slots. The revision is grammatically and semantically clear, and the correction is applied to the delivery address.", "heard_transcript": "Buy everything on my Amazon wishlist, and have it delivered to my home address. Wait, delivered to my work address, not my home address." }, "candidate_id": "wav_5495586019f16cf3", "naturalness": { "rank": 1, "mean_rank": 1.0, "singleton": false, "reasons": [ { "round": 1, "rank": 1, "reason": "The speaker begins with a clear, slightly hesitant request ('buy everything on my... Amazon wishlist...'), then corrects the delivery address with a natural-sounding 'wait' and a smooth self-repair ('delivered to my work address, not my home address'). The prosody is consistent, and the correction is integrated into the same utterance with a believable change-of-mind intonation." }, { "round": 2, "rank": 1, "reason": "This audio exhibits the most natural and continuous flow. The speaker begins with a clear request, includes a believable self-correction ('Wait, deliver it to my work address not my home address') with a clear change of mind and a smooth rephrasing. The prosody is consistent, and the correction is integrated into the same utterance without awkwardness. The initial hesitation ('uh') and filler ('um') are placed naturally, and the overall delivery sounds like a spontaneous, real-time formulation of a task." } ] }, "is_default": true }, { "variant_id": "revision_02_source_collection", "proposed_instruction": "Buy everything on my amazon shopping list, and have it delivered to my work address. Make that wishlist, not shopping list.", "long_pause_offsets": [ 85 ], "source_sha256": "ddbd699ac878c91ec9aca9ecbf18124a0096718f5bd10cd1b55fa362e713c20c", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "equivalent", "reason": "The followup corrects the source collection from 'shopping list' to 'wishlist', while retaining the action, app, and delivery address." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/9871968_3/audio_natural/revision/revision_02_source_collection/a118eb49c9a5edcc/speech.wav", "wer": 0.13636363636363635, "wav_sha256": "92edd3cb594793c1b92089f657b2addd44b03a4c2a1d9465301631af55344137", "seconds": 14.16, "spoken_utterances": [ "Buy everything on my amazon shopping list, ... and have it delivered to hmmm... my work address. ", "Make that ... wishlist, not shopping list." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker initially says 'shopping list' but immediately corrects it to 'wishlist' after a pause. The final intent matches the reference instruction exactly, including the action, scope, app, source collection, and delivery address. The correction is clear and final, satisfying the revision type criterion.", "heard_transcript": "Buy everything on my Amazon shopping list, and have it delivered to, um, my work address. Make that, wishlist, not shopping list." }, "candidate_id": "wav_09e9eefaf38d1a83", "naturalness": { "rank": 2, "mean_rank": 2.0, "singleton": false, "reasons": [ { "round": 1, "rank": 2, "reason": "The speaker makes a revision after a brief pause ('make that... wishlist, not shopping list'), which is a plausible follow-up. However, the initial request is slightly less fluent ('buy everything on my Amazon shopping list') and the correction is delivered with a more abrupt, less natural-sounding 'make that' phrase, making the revision feel slightly less spontaneous than in audio_01." }, { "round": 2, "rank": 2, "reason": "This audio shows a clear revision where the speaker first says 'shopping list' and then corrects to 'wishlist'. The correction is delivered as a separate, complete sentence ('Make that wishlist, not shopping list'), which is a plausible way to revise a request. The speaker uses a natural filler ('hmm') and the prosody is consistent. However, the revision is less integrated into the original utterance compared to audio_03, making it slightly less fluid." } ] }, "is_default": false }, { "variant_id": "revision_01_app", "proposed_instruction": "Buy everything on my ebay wishlist, and have it delivered to my work address. Change ebay to amazon.", "long_pause_offsets": [ 78 ], "source_sha256": "8ec40be4292a105135c387a8acad6ec50d65cbfa230d202566bd2c1c3f3cad6a", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "equivalent", "reason": "The followup corrects the app from eBay to Amazon, while retaining the action, scope, source collection, and delivery address." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/9871968_3/audio_natural/revision/revision_01_app/5da809b191e5946a/speech.wav", "wer": 0.08333333333333333, "wav_sha256": "23a4244d2e256c2ba7c70c8b9be6e3b5dd782dc4e18f341480567f71e83ed433", "seconds": 14.16, "spoken_utterances": [ "Buy everything on my um... ebay wishlist, and have it delivered to ah... my work address. ", "Change ebay to uh, let me think... amazon." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker initially says 'eBay wishlist' but then corrects it to 'Amazon' after a pause and self-correction. The final intent matches the reference instruction exactly, including the action, scope, app, source collection, and delivery address. The revision is clear and final, satisfying the revision type criterion.", "heard_transcript": "Buy everything on my um, eBay wishlist and have it delivered to uh, my work address. Change eBay to uh, let me think, Amazon." }, "candidate_id": "wav_500582dda932e186", "naturalness": { "rank": 3, "mean_rank": 3.0, "singleton": false, "reasons": [ { "round": 1, "rank": 3, "reason": "The speaker starts with a less confident delivery ('buy everything on my um... eBay wishlist'), and the correction ('change eBay to uh... let me think... Amazon') includes a longer, more awkward filler ('let me think') and a less fluid word replacement. The prosody is less continuous, and the correction feels more like a mechanical edit than a natural thought process." }, { "round": 2, "rank": 3, "reason": "This audio has a noticeable issue with the correction. The speaker says 'Change eBay to uh let me think Amazon', which introduces a filler ('let me think') that is not a natural self-repair but rather a separate thought process. The correction is not smoothly integrated into the original sentence, and the phrase 'let me think' sounds like a pause for thought rather than a spontaneous self-correction. The prosody is also slightly less natural, with a more robotic delivery compared to the others." } ] }, "is_default": false } ], "selected": [ { "variant_id": "revision_03_destination_address", "proposed_instruction": "Buy everything on my amazon wishlist, and have it delivered to my home address. Wait, deliver it to my work address, not my home address.", "long_pause_offsets": [ 80 ], "source_sha256": "debbba23b4367b685f90bf70c085271f68f0c027ab0a94f6f9e7dc4b690f380c", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "equivalent", "reason": "The followup revision corrects the delivery address from 'home' to 'work', while preserving the action, scope, app, and source collection." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/9871968_3/audio_natural/revision/revision_03_destination_address/ca5d0bbf9fe758fc/speech.wav", "wer": 0.10714285714285714, "wav_sha256": "40feb8ac2a20c8431045e914fac402e121aaa960e2587a0789734c2836559722", "seconds": 14.28, "spoken_utterances": [ "Buy everything on my ah... amazon wishlist, um... and have it delivered to uh... my home address. ", "Wait, deliver it to my work address, not my home address." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker initially requested delivery to the home address but then clearly revised the instruction to deliver to the work address. The final audible intent matches the reference instruction and all expected final slots. The revision is grammatically and semantically clear, and the correction is applied to the delivery address.", "heard_transcript": "Buy everything on my Amazon wishlist, and have it delivered to my home address. Wait, delivered to my work address, not my home address." }, "candidate_id": "wav_5495586019f16cf3", "naturalness": { "rank": 1, "mean_rank": 1.0, "singleton": false, "reasons": [ { "round": 1, "rank": 1, "reason": "The speaker begins with a clear, slightly hesitant request ('buy everything on my... Amazon wishlist...'), then corrects the delivery address with a natural-sounding 'wait' and a smooth self-repair ('delivered to my work address, not my home address'). The prosody is consistent, and the correction is integrated into the same utterance with a believable change-of-mind intonation." }, { "round": 2, "rank": 1, "reason": "This audio exhibits the most natural and continuous flow. The speaker begins with a clear request, includes a believable self-correction ('Wait, deliver it to my work address not my home address') with a clear change of mind and a smooth rephrasing. The prosody is consistent, and the correction is integrated into the same utterance without awkwardness. The initial hesitation ('uh') and filler ('um') are placed naturally, and the overall delivery sounds like a spontaneous, real-time formulation of a task." } ] }, "is_default": true } ] }, { "task_id": "9871968_3", "kind": "cancellation", "candidates": [ { "variant_id": "cancel_031", "hide_variant_original": true, "source_sha256": "239f90dafa8da31f98ea8ebfc2e11f72ad47ab4458d29f045f70e6301e41d395", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "not_applicable", "reason": "원래 목표 수행이 아니라 취소/대기 동작을 검수하는 유형입니다." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/9871968_3/audio_natural/cancellation/cancel_031/fff3fa563dff2767/speech.wav", "wer": 0.1111111111111111, "wav_sha256": "a13991068f2eadd46491eeb665fbb7bb04a7c5db80f4aed30da0fc9e82cf9ebd", "seconds": 14.68, "spoken_utterances": [ "Buy everything on my amazon wishlist, um... and have it delivered to hmmm... my work address.", "Oh, wait, there's no need to do that anymore. Cancel it." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker first states a complete task to buy items from the Amazon wishlist and deliver them to the work address. After a pause and hesitation, they explicitly state 'Oh wait, there's no need to do that anymore. Cancel it.' This constitutes a clear and final cancellation of the entire task. The control intent is 'cancel', which matches the expected behavior for this type.", "heard_transcript": "Buy everything on my Amazon wishlist, um, and have it delivered to, um, my work address. Oh wait, there's no need to do that anymore. Cancel it." }, "naturalness": { "rank": 1, "singleton": true }, "candidate_id": null, "is_default": true } ], "selected": [ { "variant_id": "cancel_031", "hide_variant_original": true, "source_sha256": "239f90dafa8da31f98ea8ebfc2e11f72ad47ab4458d29f045f70e6301e41d395", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "not_applicable", "reason": "원래 목표 수행이 아니라 취소/대기 동작을 검수하는 유형입니다." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/9871968_3/audio_natural/cancellation/cancel_031/fff3fa563dff2767/speech.wav", "wer": 0.1111111111111111, "wav_sha256": "a13991068f2eadd46491eeb665fbb7bb04a7c5db80f4aed30da0fc9e82cf9ebd", "seconds": 14.68, "spoken_utterances": [ "Buy everything on my amazon wishlist, um... and have it delivered to hmmm... my work address.", "Oh, wait, there's no need to do that anymore. Cancel it." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker first states a complete task to buy items from the Amazon wishlist and deliver them to the work address. After a pause and hesitation, they explicitly state 'Oh wait, there's no need to do that anymore. Cancel it.' This constitutes a clear and final cancellation of the entire task. The control intent is 'cancel', which matches the expected behavior for this type.", "heard_transcript": "Buy everything on my Amazon wishlist, um, and have it delivered to, um, my work address. Oh wait, there's no need to do that anymore. Cancel it." }, "naturalness": { "rank": 1, "singleton": true }, "candidate_id": null, "is_default": true } ] }, { "task_id": "9871968_3", "kind": "ambiguous_halt", "candidates": [ { "variant_id": "halt_027", "hide_variant_original": true, "source_sha256": "939a64aba187eec592f01af07ac7cdf723ef53d587d21a78b98786fc9f727a62", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "not_applicable", "reason": "원래 목표 수행이 아니라 취소/대기 동작을 검수하는 유형입니다." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/9871968_3/audio_natural/ambiguous_halt/halt_027/d1be0b92704dcf8c/speech.wav", "wer": 0.13636363636363635, "wav_sha256": "70695cd2f2c7a8ccc1d02bd923218725550fc1e51ec8dc86b48118ccdedfd1b7", "seconds": 13.44, "spoken_utterances": [ "Buy everything on my amazon wishlist, um... and have it delivered to hmmm... my work address.", "Hang on, I need a moment." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker completes the original instruction and then clearly requests a pause with 'Hang on, I need a moment.' This constitutes an unambiguous halt request, matching the 'ambiguous_halt' type. The original task is not canceled, and the speaker does not correct or modify it. The final control intent is to pause, not execute or cancel.", "heard_transcript": "Buy everything on my Amazon wishlist, um, and have it delivered to, um, my work address. Hang on, I need a moment." }, "naturalness": { "rank": 1, "singleton": true }, "candidate_id": null, "is_default": true } ], "selected": [ { "variant_id": "halt_027", "hide_variant_original": true, "source_sha256": "939a64aba187eec592f01af07ac7cdf723ef53d587d21a78b98786fc9f727a62", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "not_applicable", "reason": "원래 목표 수행이 아니라 취소/대기 동작을 검수하는 유형입니다." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/9871968_3/audio_natural/ambiguous_halt/halt_027/d1be0b92704dcf8c/speech.wav", "wer": 0.13636363636363635, "wav_sha256": "70695cd2f2c7a8ccc1d02bd923218725550fc1e51ec8dc86b48118ccdedfd1b7", "seconds": 13.44, "spoken_utterances": [ "Buy everything on my amazon wishlist, um... and have it delivered to hmmm... my work address.", "Hang on, I need a moment." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker completes the original instruction and then clearly requests a pause with 'Hang on, I need a moment.' This constitutes an unambiguous halt request, matching the 'ambiguous_halt' type. The original task is not canceled, and the speaker does not correct or modify it. The final control intent is to pause, not execute or cancel.", "heard_transcript": "Buy everything on my Amazon wishlist, um, and have it delivered to, um, my work address. Hang on, I need a moment." }, "naturalness": { "rank": 1, "singleton": true }, "candidate_id": null, "is_default": true } ] }, { "task_id": "9871968_3", "kind": "filler", "candidates": [ { "variant_id": "original", "hide_variant_original": true, "source_sha256": "bcbca6751833526065222c4b90de7eeacb97496a0e01c34fb68e4d8aec95f621", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "equivalent", "reason": "The spoken instruction matches the original instruction exactly, with no corrections or omissions. All requirements including action, scope, app, source, and delivery address are preserved." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/9871968_3/audio_natural/filler/original/b404fa7f177d60e0/speech.wav", "wer": 0.1875, "wav_sha256": "207b97a148b1b2814ecaea78337b4c4e48504740e9a65a0d9ff599cfbbd2ca4a", "seconds": 9.6, "spoken_utterances": [ "Buy everything on my amazon wishlist, um... and have it delivered to hmmm... my work address." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker begins the task with the correct instruction and uses fillers ('um') but does not change the goal. The final intent matches the original instruction exactly, including all slots. The fillers do not alter the task, so the execution intent is preserved.", "heard_transcript": "Buy everything on my Amazon wishlist, um, and have it delivered to, um, my work address." }, "naturalness": { "rank": 1, "singleton": true }, "candidate_id": null, "is_default": true } ], "selected": [ { "variant_id": "original", "hide_variant_original": true, "source_sha256": "bcbca6751833526065222c4b90de7eeacb97496a0e01c34fb68e4d8aec95f621", "decision": null, "review_revision": null, "stale_review": false, "semantic": { "status": "equivalent", "reason": "The spoken instruction matches the original instruction exactly, with no corrections or omissions. All requirements including action, scope, app, source, and delivery address are preserved." }, "audio": { "status": "ready", "path": "/home/hbcho/agent/appworld/data/tasks_llm/9871968_3/audio_natural/filler/original/b404fa7f177d60e0/speech.wav", "wer": 0.1875, "wav_sha256": "207b97a148b1b2814ecaea78337b4c4e48504740e9a65a0d9ff599cfbbd2ca4a", "seconds": 9.6, "spoken_utterances": [ "Buy everything on my amazon wishlist, um... and have it delivered to hmmm... my work address." ] }, "audio_semantic": { "status": "equivalent", "reason": "The speaker begins the task with the correct instruction and uses fillers ('um') but does not change the goal. The final intent matches the original instruction exactly, including all slots. The fillers do not alter the task, so the execution intent is preserved.", "heard_transcript": "Buy everything on my Amazon wishlist, um, and have it delivered to, um, my work address." }, "naturalness": { "rank": 1, "singleton": true }, "candidate_id": null, "is_default": true } ] } ]