Dataset Viewer
Auto-converted to Parquet Duplicate
id
stringlengths
31
31
conversation_id
stringlengths
28
28
task
stringclasses
1 value
audio
audioduration (s)
16.4
32.8
instruction
stringclasses
184 values
choices
listlengths
2
4
choice_keys
listlengths
2
4
answer
stringclasses
4 values
answer_index
int64
0
3
tts_model
stringclasses
4 values
tts_runtime
stringclasses
4 values
voice_ids
listlengths
2
4
duration_seconds
float64
16.4
32.8
render_attempt
int64
0
6
passes_checks
bool
1 class
details
stringlengths
3.43k
4.33k
fact_attribution-train-00611-q0
fact_attribution-train-00611
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they are learning Turkish?
[ "Kim", "Robin" ]
[ "A", "B" ]
A
0
pocket
mlx-audio 0.5.8 (Apple M5 Pro)
[ "1349f12546a467cd3c34629c35875b03f557a5fed57de9d35424fda9d632fde5", "e782613e49b76514ae56290bde4d5a05c57345de9fd8486d64b3031b9f4b1b4a" ]
25.318
0
true
{"names": ["Robin", "Kim"], "speakers": [0, 1, 1, 0, 1, 0, 1, 0], "intro_turns": 2, "target_turn": 2, "fact": {"kind": "language", "value": "Turkish", "speaker": 1, "ask": "are learning Turkish"}, "facts": [{"kind": "language", "value": "Turkish", "speaker": 1, "ask": "are learning Turkish"}, {"kind": "language", "valu...
fact_attribution-train-00611-q1
fact_attribution-train-00611
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they play the cello?
[ "Kim", "Robin" ]
[ "A", "B" ]
A
0
pocket
mlx-audio 0.5.8 (Apple M5 Pro)
[ "1349f12546a467cd3c34629c35875b03f557a5fed57de9d35424fda9d632fde5", "e782613e49b76514ae56290bde4d5a05c57345de9fd8486d64b3031b9f4b1b4a" ]
25.318
0
true
{"names": ["Robin", "Kim"], "speakers": [0, 1, 1, 0, 1, 0, 1, 0], "intro_turns": 2, "target_turn": 4, "fact": {"kind": "instrument", "value": "the cello", "speaker": 1, "ask": "play the cello"}, "facts": [{"kind": "language", "value": "Turkish", "speaker": 1, "ask": "are learning Turkish"}, {"kind": "language", "value"...
fact_attribution-train-00611-q2
fact_attribution-train-00611
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they go cycling?
[ "Robin", "Kim" ]
[ "A", "B" ]
B
1
pocket
mlx-audio 0.5.8 (Apple M5 Pro)
[ "1349f12546a467cd3c34629c35875b03f557a5fed57de9d35424fda9d632fde5", "e782613e49b76514ae56290bde4d5a05c57345de9fd8486d64b3031b9f4b1b4a" ]
25.318
0
true
{"names": ["Robin", "Kim"], "speakers": [0, 1, 1, 0, 1, 0, 1, 0], "intro_turns": 2, "target_turn": 6, "fact": {"kind": "sport", "value": "cycling", "speaker": 1, "ask": "go cycling"}, "facts": [{"kind": "language", "value": "Turkish", "speaker": 1, "ask": "are learning Turkish"}, {"kind": "language", "value": "German",...
fact_attribution-train-00794-q0
fact_attribution-train-00794
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they are afraid of heights?
[ "Jordan", "Morgan", "Charlie" ]
[ "A", "B", "C" ]
C
2
omni
PyTorch CUDA (RTX 3090)
[ "27ec03b0394190262a832f292eed16048a6951baf4023ba97390615d4029d26f", "6d22b902afebc8f631688060845837fbe94a44c427dd7ebf3238ed4be6da98f4", "b57dc54de92c36391c6681b49573f75531d23a6c952f5bc2ac25ba4b7b7e52b7" ]
23.056
0
true
{"names": ["Morgan", "Charlie", "Jordan"], "speakers": [0, 1, 2, 1, 0, 1, 2, 0, 2], "intro_turns": 3, "target_turn": 3, "fact": {"kind": "fear", "value": "heights", "speaker": 1, "ask": "are afraid of heights"}, "facts": [{"kind": "fear", "value": "heights", "speaker": 1, "ask": "are afraid of heights"}, {"kind": "fear...
fact_attribution-train-00794-q1
fact_attribution-train-00794
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they grow lavender on their balcony?
[ "Charlie", "Morgan", "Jordan" ]
[ "A", "B", "C" ]
C
2
omni
PyTorch CUDA (RTX 3090)
[ "27ec03b0394190262a832f292eed16048a6951baf4023ba97390615d4029d26f", "6d22b902afebc8f631688060845837fbe94a44c427dd7ebf3238ed4be6da98f4", "b57dc54de92c36391c6681b49573f75531d23a6c952f5bc2ac25ba4b7b7e52b7" ]
23.056
0
true
{"names": ["Morgan", "Charlie", "Jordan"], "speakers": [0, 1, 2, 1, 0, 1, 2, 0, 2], "intro_turns": 3, "target_turn": 6, "fact": {"kind": "garden", "value": "lavender", "speaker": 2, "ask": "grow lavender on their balcony"}, "facts": [{"kind": "fear", "value": "heights", "speaker": 1, "ask": "are afraid of heights"}, {"...
fact_attribution-train-00794-q2
fact_attribution-train-00794
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they moved here six months ago?
[ "Morgan", "Charlie", "Jordan" ]
[ "A", "B", "C" ]
A
0
omni
PyTorch CUDA (RTX 3090)
[ "27ec03b0394190262a832f292eed16048a6951baf4023ba97390615d4029d26f", "6d22b902afebc8f631688060845837fbe94a44c427dd7ebf3238ed4be6da98f4", "b57dc54de92c36391c6681b49573f75531d23a6c952f5bc2ac25ba4b7b7e52b7" ]
23.056
0
true
{"names": ["Morgan", "Charlie", "Jordan"], "speakers": [0, 1, 2, 1, 0, 1, 2, 0, 2], "intro_turns": 3, "target_turn": 7, "fact": {"kind": "moved", "value": "six months", "speaker": 0, "ask": "moved here six months ago"}, "facts": [{"kind": "fear", "value": "heights", "speaker": 1, "ask": "are afraid of heights"}, {"kind...
fact_attribution-train-00615-q0
fact_attribution-train-00615
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they start the morning with an espresso?
[ "Casey", "Charlie", "Jordan", "Morgan" ]
[ "A", "B", "C", "D" ]
A
0
pocket
mlx-audio 0.5.8 (Apple M5 Pro)
[ "4b75457e388a1e7eea495988ad5a995299cb60f58dfa05785d9dfae3b2d38970", "cae65d2662c7ae70efb84d525e81dbd3a7fb47f303a1437e7b0d34b4df08ee4b", "73f03ee2a629ad1703529101fabf1907bbb332a29a5c4db64c0351a66313b93c", "879d1788415da2721d5443d09fc6969138cfcaf717b6a8e4db0ffcedb9d660d0" ]
22.374
0
true
{"names": ["Charlie", "Jordan", "Casey", "Morgan"], "speakers": [0, 1, 2, 3, 2, 3, 2, 0], "intro_turns": 4, "target_turn": 4, "fact": {"kind": "drink", "value": "an espresso", "speaker": 2, "ask": "start the morning with an espresso"}, "facts": [{"kind": "drink", "value": "an espresso", "speaker": 2, "ask": "start the ...
fact_attribution-train-00615-q1
fact_attribution-train-00615
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they plan to go camping this weekend?
[ "Charlie", "Morgan", "Jordan", "Casey" ]
[ "A", "B", "C", "D" ]
D
3
pocket
mlx-audio 0.5.8 (Apple M5 Pro)
[ "4b75457e388a1e7eea495988ad5a995299cb60f58dfa05785d9dfae3b2d38970", "cae65d2662c7ae70efb84d525e81dbd3a7fb47f303a1437e7b0d34b4df08ee4b", "73f03ee2a629ad1703529101fabf1907bbb332a29a5c4db64c0351a66313b93c", "879d1788415da2721d5443d09fc6969138cfcaf717b6a8e4db0ffcedb9d660d0" ]
22.374
0
true
{"names": ["Charlie", "Jordan", "Casey", "Morgan"], "speakers": [0, 1, 2, 3, 2, 3, 2, 0], "intro_turns": 4, "target_turn": 6, "fact": {"kind": "weekend", "value": "go camping", "speaker": 2, "ask": "plan to go camping this weekend"}, "facts": [{"kind": "drink", "value": "an espresso", "speaker": 2, "ask": "start the mo...
fact_attribution-train-00615-q2
fact_attribution-train-00615
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they plan to try a new restaurant this weekend?
[ "Casey", "Charlie", "Morgan", "Jordan" ]
[ "A", "B", "C", "D" ]
B
1
pocket
mlx-audio 0.5.8 (Apple M5 Pro)
[ "4b75457e388a1e7eea495988ad5a995299cb60f58dfa05785d9dfae3b2d38970", "cae65d2662c7ae70efb84d525e81dbd3a7fb47f303a1437e7b0d34b4df08ee4b", "73f03ee2a629ad1703529101fabf1907bbb332a29a5c4db64c0351a66313b93c", "879d1788415da2721d5443d09fc6969138cfcaf717b6a8e4db0ffcedb9d660d0" ]
22.374
0
true
{"names": ["Charlie", "Jordan", "Casey", "Morgan"], "speakers": [0, 1, 2, 3, 2, 3, 2, 0], "intro_turns": 4, "target_turn": 7, "fact": {"kind": "weekend", "value": "try a new restaurant", "speaker": 0, "ask": "plan to try a new restaurant this weekend"}, "facts": [{"kind": "drink", "value": "an espresso", "speaker": 2, ...
fact_attribution-train-00690-q0
fact_attribution-train-00690
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they have two hamsters?
[ "Sam", "Taylor", "Alex", "Jamie" ]
[ "A", "B", "C", "D" ]
D
3
qwen
PyTorch CUDA (RTX 3090); replaced clips: PyTorch MPS (Apple M5 Pro)
[ "277c3eaed7e5b1919da372bb445a99c1f9427c1542beb381ed64af60a7013103", "abefb8d0d83c1405628099d716d34eecde6d9063c28cd1959154668baaf4699d", "de5a7fc7059508ea96816e528c10d1eb077cb099378907df25f7cf9a5576c735", "d093ac6548f79e456acb32bbdd285fd0d3e2a1012b5689e72e0c13b6f12d9e48" ]
20.678
1
true
{"names": ["Sam", "Alex", "Jamie", "Taylor"], "speakers": [0, 1, 2, 3, 2, 1, 3, 0], "intro_turns": 4, "target_turn": 4, "fact": {"kind": "pet", "value": "two hamsters", "speaker": 2, "ask": "have two hamsters"}, "facts": [{"kind": "pet", "value": "two hamsters", "speaker": 2, "ask": "have two hamsters"}, {"kind": "pet"...
fact_attribution-train-00690-q1
fact_attribution-train-00690
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they have a parrot?
[ "Alex", "Jamie", "Sam", "Taylor" ]
[ "A", "B", "C", "D" ]
A
0
qwen
PyTorch CUDA (RTX 3090); replaced clips: PyTorch MPS (Apple M5 Pro)
[ "277c3eaed7e5b1919da372bb445a99c1f9427c1542beb381ed64af60a7013103", "abefb8d0d83c1405628099d716d34eecde6d9063c28cd1959154668baaf4699d", "de5a7fc7059508ea96816e528c10d1eb077cb099378907df25f7cf9a5576c735", "d093ac6548f79e456acb32bbdd285fd0d3e2a1012b5689e72e0c13b6f12d9e48" ]
20.678
1
true
{"names": ["Sam", "Alex", "Jamie", "Taylor"], "speakers": [0, 1, 2, 3, 2, 1, 3, 0], "intro_turns": 4, "target_turn": 5, "fact": {"kind": "pet", "value": "a parrot", "speaker": 1, "ask": "have a parrot"}, "facts": [{"kind": "pet", "value": "two hamsters", "speaker": 2, "ask": "have two hamsters"}, {"kind": "pet", "value...
fact_attribution-train-00690-q2
fact_attribution-train-00690
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they go rowing?
[ "Alex", "Jamie", "Sam", "Taylor" ]
[ "A", "B", "C", "D" ]
D
3
qwen
PyTorch CUDA (RTX 3090); replaced clips: PyTorch MPS (Apple M5 Pro)
[ "277c3eaed7e5b1919da372bb445a99c1f9427c1542beb381ed64af60a7013103", "abefb8d0d83c1405628099d716d34eecde6d9063c28cd1959154668baaf4699d", "de5a7fc7059508ea96816e528c10d1eb077cb099378907df25f7cf9a5576c735", "d093ac6548f79e456acb32bbdd285fd0d3e2a1012b5689e72e0c13b6f12d9e48" ]
20.678
1
true
{"names": ["Sam", "Alex", "Jamie", "Taylor"], "speakers": [0, 1, 2, 3, 2, 1, 3, 0], "intro_turns": 4, "target_turn": 6, "fact": {"kind": "sport", "value": "rowing", "speaker": 3, "ask": "go rowing"}, "facts": [{"kind": "pet", "value": "two hamsters", "speaker": 2, "ask": "have two hamsters"}, {"kind": "pet", "value": "...
fact_attribution-train-00589-q0
fact_attribution-train-00589
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they love tacos?
[ "Jordan", "Avery", "Sam" ]
[ "A", "B", "C" ]
C
2
omni
PyTorch CUDA (RTX 3090)
[ "1f5309738cba446fb0c2bbd22547cf2a96f9778b8b9c629c991ac619dc65a006", "6def6d473a2d517116ebc9a9dbd1c8c6723f398949886785128b453f15a23ca8", "7bc037a2131e7381fcee96afa8adb012cc7786a0ab7ab7de74b86ff4de22ecb5" ]
24.869
0
true
{"names": ["Sam", "Jordan", "Avery"], "speakers": [0, 1, 2, 0, 2, 0, 2, 0, 1], "intro_turns": 3, "target_turn": 3, "fact": {"kind": "food", "value": "tacos", "speaker": 0, "ask": "love tacos"}, "facts": [{"kind": "food", "value": "tacos", "speaker": 0, "ask": "love tacos"}, {"kind": "food", "value": "ramen", "speaker":...
fact_attribution-train-00589-q1
fact_attribution-train-00589
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they moved here three weeks ago?
[ "Sam", "Jordan", "Avery" ]
[ "A", "B", "C" ]
C
2
omni
PyTorch CUDA (RTX 3090)
[ "1f5309738cba446fb0c2bbd22547cf2a96f9778b8b9c629c991ac619dc65a006", "6def6d473a2d517116ebc9a9dbd1c8c6723f398949886785128b453f15a23ca8", "7bc037a2131e7381fcee96afa8adb012cc7786a0ab7ab7de74b86ff4de22ecb5" ]
24.869
0
true
{"names": ["Sam", "Jordan", "Avery"], "speakers": [0, 1, 2, 0, 2, 0, 2, 0, 1], "intro_turns": 3, "target_turn": 6, "fact": {"kind": "moved", "value": "three weeks", "speaker": 2, "ask": "moved here three weeks ago"}, "facts": [{"kind": "food", "value": "tacos", "speaker": 0, "ask": "love tacos"}, {"kind": "food", "valu...
fact_attribution-train-00589-q2
fact_attribution-train-00589
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they are learning German?
[ "Sam", "Jordan", "Avery" ]
[ "A", "B", "C" ]
B
1
omni
PyTorch CUDA (RTX 3090)
[ "1f5309738cba446fb0c2bbd22547cf2a96f9778b8b9c629c991ac619dc65a006", "6def6d473a2d517116ebc9a9dbd1c8c6723f398949886785128b453f15a23ca8", "7bc037a2131e7381fcee96afa8adb012cc7786a0ab7ab7de74b86ff4de22ecb5" ]
24.869
0
true
{"names": ["Sam", "Jordan", "Avery"], "speakers": [0, 1, 2, 0, 2, 0, 2, 0, 1], "intro_turns": 3, "target_turn": 8, "fact": {"kind": "language", "value": "German", "speaker": 1, "ask": "are learning German"}, "facts": [{"kind": "food", "value": "tacos", "speaker": 0, "ask": "love tacos"}, {"kind": "food", "value": "rame...
fact_attribution-train-00251-q0
fact_attribution-train-00251
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they are allergic to peanuts?
[ "Charlie", "Kim", "Riley" ]
[ "A", "B", "C" ]
A
0
pocket
mlx-audio 0.5.8 (Apple M5 Pro)
[ "f064cf4902ba261042926b900185871e1a2fdbdc07da9743348b15e35ad4361f", "74c3ac3d5d0c2c87f0aace19c7ab71cdcb993a79c355d76771b54e608cb92776", "48c18d19503981a44c54ad720e03c5e9fd1e975d484c6f3857b86d5fa6907b7d" ]
23.864
0
true
{"names": ["Riley", "Charlie", "Kim"], "speakers": [0, 1, 2, 1, 0, 2, 0, 1, 0], "intro_turns": 3, "target_turn": 3, "fact": {"kind": "allergy", "value": "peanuts", "speaker": 1, "ask": "are allergic to peanuts"}, "facts": [{"kind": "allergy", "value": "peanuts", "speaker": 1, "ask": "are allergic to peanuts"}, {"kind":...
fact_attribution-train-00251-q1
fact_attribution-train-00251
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they are allergic to strawberries?
[ "Riley", "Charlie", "Kim" ]
[ "A", "B", "C" ]
A
0
pocket
mlx-audio 0.5.8 (Apple M5 Pro)
[ "f064cf4902ba261042926b900185871e1a2fdbdc07da9743348b15e35ad4361f", "74c3ac3d5d0c2c87f0aace19c7ab71cdcb993a79c355d76771b54e608cb92776", "48c18d19503981a44c54ad720e03c5e9fd1e975d484c6f3857b86d5fa6907b7d" ]
23.864
0
true
{"names": ["Riley", "Charlie", "Kim"], "speakers": [0, 1, 2, 1, 0, 2, 0, 1, 0], "intro_turns": 3, "target_turn": 4, "fact": {"kind": "allergy", "value": "strawberries", "speaker": 0, "ask": "are allergic to strawberries"}, "facts": [{"kind": "allergy", "value": "peanuts", "speaker": 1, "ask": "are allergic to peanuts"}...
fact_attribution-train-00251-q2
fact_attribution-train-00251
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they go cycling?
[ "Kim", "Charlie", "Riley" ]
[ "A", "B", "C" ]
C
2
pocket
mlx-audio 0.5.8 (Apple M5 Pro)
[ "f064cf4902ba261042926b900185871e1a2fdbdc07da9743348b15e35ad4361f", "74c3ac3d5d0c2c87f0aace19c7ab71cdcb993a79c355d76771b54e608cb92776", "48c18d19503981a44c54ad720e03c5e9fd1e975d484c6f3857b86d5fa6907b7d" ]
23.864
0
true
{"names": ["Riley", "Charlie", "Kim"], "speakers": [0, 1, 2, 1, 0, 2, 0, 1, 0], "intro_turns": 3, "target_turn": 6, "fact": {"kind": "sport", "value": "cycling", "speaker": 0, "ask": "go cycling"}, "facts": [{"kind": "allergy", "value": "peanuts", "speaker": 1, "ask": "are allergic to peanuts"}, {"kind": "allergy", "va...
fact_attribution-train-00552-q0
fact_attribution-train-00552
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they read a lot of cookbooks?
[ "Jordan", "Avery" ]
[ "A", "B" ]
B
1
qwen
PyTorch CUDA (RTX 3090)
[ "402d0935b91770b9939a2afe5939426fb34e91917928e65060a906d49e0d1faf", "fc50d0607fd3eeb5a370a8168921567e72e13708533f8be86c50b2b438812790" ]
22.922
0
true
{"names": ["Jordan", "Avery"], "speakers": [0, 1, 0, 1, 0, 1, 0, 1], "intro_turns": 2, "target_turn": 3, "fact": {"kind": "reading", "value": "cookbooks", "speaker": 1, "ask": "read a lot of cookbooks"}, "facts": [{"kind": "reading", "value": "travel guides", "speaker": 0, "ask": "read a lot of travel guides"}, {"kind"...
fact_attribution-train-00552-q1
fact_attribution-train-00552
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they work as a librarian?
[ "Jordan", "Avery" ]
[ "A", "B" ]
A
0
qwen
PyTorch CUDA (RTX 3090)
[ "402d0935b91770b9939a2afe5939426fb34e91917928e65060a906d49e0d1faf", "fc50d0607fd3eeb5a370a8168921567e72e13708533f8be86c50b2b438812790" ]
22.922
0
true
{"names": ["Jordan", "Avery"], "speakers": [0, 1, 0, 1, 0, 1, 0, 1], "intro_turns": 2, "target_turn": 4, "fact": {"kind": "job", "value": "a librarian", "speaker": 0, "ask": "work as a librarian"}, "facts": [{"kind": "reading", "value": "travel guides", "speaker": 0, "ask": "read a lot of travel guides"}, {"kind": "rea...
fact_attribution-train-00552-q2
fact_attribution-train-00552
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they work as a nurse?
[ "Avery", "Jordan" ]
[ "A", "B" ]
A
0
qwen
PyTorch CUDA (RTX 3090)
[ "402d0935b91770b9939a2afe5939426fb34e91917928e65060a906d49e0d1faf", "fc50d0607fd3eeb5a370a8168921567e72e13708533f8be86c50b2b438812790" ]
22.922
0
true
{"names": ["Jordan", "Avery"], "speakers": [0, 1, 0, 1, 0, 1, 0, 1], "intro_turns": 2, "target_turn": 5, "fact": {"kind": "job", "value": "a nurse", "speaker": 1, "ask": "work as a nurse"}, "facts": [{"kind": "reading", "value": "travel guides", "speaker": 0, "ask": "read a lot of travel guides"}, {"kind": "reading", "...
fact_attribution-train-00149-q0
fact_attribution-train-00149
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they studied art history?
[ "Charlie", "Sam", "Casey", "Alex" ]
[ "A", "B", "C", "D" ]
C
2
sopro
PyTorch MPS (Apple M5 Pro)
[ "b8b06172c6b26dc5058e2b9ee84a40322b3018db4d81194a28fa06df386a2ca0", "b2787d582d0114ccb56cb9c666c1537bba8419cd5baa7e4c77aa064885c4d803", "a7346f8da24ae777cb7286cf6d70532f3e408835b47470c03074980026b8b433", "47e948a5b3c3cbddb606aaa4cb92ad2bf2b760f7cd87e9bb3fe94f2caa9c9bc0" ]
22.337
0
true
{"names": ["Charlie", "Sam", "Alex", "Casey"], "speakers": [0, 1, 2, 3, 3, 0, 2, 0], "intro_turns": 4, "target_turn": 4, "fact": {"kind": "studies", "value": "art history", "speaker": 3, "ask": "studied art history"}, "facts": [{"kind": "studies", "value": "art history", "speaker": 3, "ask": "studied art history"}, {"k...
fact_attribution-train-00149-q1
fact_attribution-train-00149
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they studied physics?
[ "Sam", "Alex", "Charlie", "Casey" ]
[ "A", "B", "C", "D" ]
C
2
sopro
PyTorch MPS (Apple M5 Pro)
[ "b8b06172c6b26dc5058e2b9ee84a40322b3018db4d81194a28fa06df386a2ca0", "b2787d582d0114ccb56cb9c666c1537bba8419cd5baa7e4c77aa064885c4d803", "a7346f8da24ae777cb7286cf6d70532f3e408835b47470c03074980026b8b433", "47e948a5b3c3cbddb606aaa4cb92ad2bf2b760f7cd87e9bb3fe94f2caa9c9bc0" ]
22.337
0
true
{"names": ["Charlie", "Sam", "Alex", "Casey"], "speakers": [0, 1, 2, 3, 3, 0, 2, 0], "intro_turns": 4, "target_turn": 5, "fact": {"kind": "studies", "value": "physics", "speaker": 0, "ask": "studied physics"}, "facts": [{"kind": "studies", "value": "art history", "speaker": 3, "ask": "studied art history"}, {"kind": "s...
fact_attribution-train-00149-q2
fact_attribution-train-00149
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they moved here a year ago?
[ "Casey", "Alex", "Charlie", "Sam" ]
[ "A", "B", "C", "D" ]
C
2
sopro
PyTorch MPS (Apple M5 Pro)
[ "b8b06172c6b26dc5058e2b9ee84a40322b3018db4d81194a28fa06df386a2ca0", "b2787d582d0114ccb56cb9c666c1537bba8419cd5baa7e4c77aa064885c4d803", "a7346f8da24ae777cb7286cf6d70532f3e408835b47470c03074980026b8b433", "47e948a5b3c3cbddb606aaa4cb92ad2bf2b760f7cd87e9bb3fe94f2caa9c9bc0" ]
22.337
0
true
{"names": ["Charlie", "Sam", "Alex", "Casey"], "speakers": [0, 1, 2, 3, 3, 0, 2, 0], "intro_turns": 4, "target_turn": 7, "fact": {"kind": "moved", "value": "a year", "speaker": 0, "ask": "moved here a year ago"}, "facts": [{"kind": "studies", "value": "art history", "speaker": 3, "ask": "studied art history"}, {"kind":...
fact_attribution-train-00150-q0
fact_attribution-train-00150
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they grew up in Glasgow?
[ "Charlie", "Jordan", "Alex", "Taylor" ]
[ "A", "B", "C", "D" ]
B
1
qwen
PyTorch CUDA (RTX 3090)
[ "aa809c8fe0053162495e84ed8f16de99cba36560863bff598d89f82e85c67ce2", "6003d581bdb5084048374c74a1c45a60f4f28ac6723b7ac701a504ff47434992", "4dda641da601f66fc44aa4c38f1c346c7e64363f552a911bbe7aa6d86487a62f", "577d14d47fab1140d06c467a58874891494921b69a4cafc93cd334256ee37a63" ]
20.41
0
true
{"names": ["Jordan", "Taylor", "Charlie", "Alex"], "speakers": [0, 1, 2, 3, 0, 1, 3, 1], "intro_turns": 4, "target_turn": 4, "fact": {"kind": "hometown", "value": "Glasgow", "speaker": 0, "ask": "grew up in Glasgow"}, "facts": [{"kind": "hometown", "value": "Glasgow", "speaker": 0, "ask": "grew up in Glasgow"}, {"kind"...
fact_attribution-train-00150-q1
fact_attribution-train-00150
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they drive a tiny electric car?
[ "Taylor", "Charlie", "Jordan", "Alex" ]
[ "A", "B", "C", "D" ]
D
3
qwen
PyTorch CUDA (RTX 3090)
[ "aa809c8fe0053162495e84ed8f16de99cba36560863bff598d89f82e85c67ce2", "6003d581bdb5084048374c74a1c45a60f4f28ac6723b7ac701a504ff47434992", "4dda641da601f66fc44aa4c38f1c346c7e64363f552a911bbe7aa6d86487a62f", "577d14d47fab1140d06c467a58874891494921b69a4cafc93cd334256ee37a63" ]
20.41
0
true
{"names": ["Jordan", "Taylor", "Charlie", "Alex"], "speakers": [0, 1, 2, 3, 0, 1, 3, 1], "intro_turns": 4, "target_turn": 6, "fact": {"kind": "vehicle", "value": "a tiny electric car", "speaker": 3, "ask": "drive a tiny electric car"}, "facts": [{"kind": "hometown", "value": "Glasgow", "speaker": 0, "ask": "grew up in ...
fact_attribution-train-00150-q2
fact_attribution-train-00150
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they drive an old red van?
[ "Charlie", "Alex", "Taylor", "Jordan" ]
[ "A", "B", "C", "D" ]
C
2
qwen
PyTorch CUDA (RTX 3090)
[ "aa809c8fe0053162495e84ed8f16de99cba36560863bff598d89f82e85c67ce2", "6003d581bdb5084048374c74a1c45a60f4f28ac6723b7ac701a504ff47434992", "4dda641da601f66fc44aa4c38f1c346c7e64363f552a911bbe7aa6d86487a62f", "577d14d47fab1140d06c467a58874891494921b69a4cafc93cd334256ee37a63" ]
20.41
0
true
{"names": ["Jordan", "Taylor", "Charlie", "Alex"], "speakers": [0, 1, 2, 3, 0, 1, 3, 1], "intro_turns": 4, "target_turn": 7, "fact": {"kind": "vehicle", "value": "an old red van", "speaker": 1, "ask": "drive an old red van"}, "facts": [{"kind": "hometown", "value": "Glasgow", "speaker": 0, "ask": "grew up in Glasgow"},...
fact_attribution-train-00108-q0
fact_attribution-train-00108
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they moved here a year ago?
[ "Jamie", "Kim" ]
[ "A", "B" ]
A
0
sopro
PyTorch MPS (Apple M5 Pro)
[ "0cc415290bbe581f8a3e8caf31ec7da38d45305bc0a37345d87e0c752b37a8ae", "5325d78382931dd4d303a7feb90f689b789d1d5405835a6aa8662e816dc52423" ]
24.96
0
true
{"names": ["Jamie", "Kim"], "speakers": [0, 1, 1, 0, 1, 0, 1, 0], "intro_turns": 2, "target_turn": 3, "fact": {"kind": "moved", "value": "a year", "speaker": 0, "ask": "moved here a year ago"}, "facts": [{"kind": "moved", "value": "two years", "speaker": 1, "ask": "moved here two years ago"}, {"kind": "moved", "value":...
fact_attribution-train-00108-q1
fact_attribution-train-00108
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they take their scooter to work?
[ "Jamie", "Kim" ]
[ "A", "B" ]
B
1
sopro
PyTorch MPS (Apple M5 Pro)
[ "0cc415290bbe581f8a3e8caf31ec7da38d45305bc0a37345d87e0c752b37a8ae", "5325d78382931dd4d303a7feb90f689b789d1d5405835a6aa8662e816dc52423" ]
24.96
0
true
{"names": ["Jamie", "Kim"], "speakers": [0, 1, 1, 0, 1, 0, 1, 0], "intro_turns": 2, "target_turn": 4, "fact": {"kind": "commute", "value": "my scooter", "speaker": 1, "ask": "take their scooter to work"}, "facts": [{"kind": "moved", "value": "two years", "speaker": 1, "ask": "moved here two years ago"}, {"kind": "moved...
fact_attribution-train-00108-q2
fact_attribution-train-00108
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they are afraid of clowns?
[ "Kim", "Jamie" ]
[ "A", "B" ]
B
1
sopro
PyTorch MPS (Apple M5 Pro)
[ "0cc415290bbe581f8a3e8caf31ec7da38d45305bc0a37345d87e0c752b37a8ae", "5325d78382931dd4d303a7feb90f689b789d1d5405835a6aa8662e816dc52423" ]
24.96
0
true
{"names": ["Jamie", "Kim"], "speakers": [0, 1, 1, 0, 1, 0, 1, 0], "intro_turns": 2, "target_turn": 7, "fact": {"kind": "fear", "value": "clowns", "speaker": 0, "ask": "are afraid of clowns"}, "facts": [{"kind": "moved", "value": "two years", "speaker": 1, "ask": "moved here two years ago"}, {"kind": "moved", "value": "...
fact_attribution-train-00783-q0
fact_attribution-train-00783
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they start the morning with orange juice?
[ "Avery", "Riley", "Robin", "Kim" ]
[ "A", "B", "C", "D" ]
A
0
sopro
PyTorch MPS (Apple M5 Pro)
[ "5b1662f2226a7e6763aeda046320a10c39bc9467b65e6e06ebc8f026ca5a4b68", "53acc0b07cf4ebbccfa1e11a9e7900eb500f816e90fb467c1f29863be4e5a11b", "155767ed4cb55fa495799e4afb65bf5c5a07f9e702b565b51d9f459ca3fa0310", "2720503ebf27ef4fc779c804602f6566c0e56ab9805fd66e46da7d9e87fe5f0a" ]
24.123
0
true
{"names": ["Kim", "Riley", "Avery", "Robin"], "speakers": [0, 1, 2, 3, 2, 1, 2, 0], "intro_turns": 4, "target_turn": 4, "fact": {"kind": "drink", "value": "orange juice", "speaker": 2, "ask": "start the morning with orange juice"}, "facts": [{"kind": "drink", "value": "orange juice", "speaker": 2, "ask": "start the mor...
fact_attribution-train-00783-q1
fact_attribution-train-00783
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they start the morning with an espresso?
[ "Robin", "Riley", "Avery", "Kim" ]
[ "A", "B", "C", "D" ]
B
1
sopro
PyTorch MPS (Apple M5 Pro)
[ "5b1662f2226a7e6763aeda046320a10c39bc9467b65e6e06ebc8f026ca5a4b68", "53acc0b07cf4ebbccfa1e11a9e7900eb500f816e90fb467c1f29863be4e5a11b", "155767ed4cb55fa495799e4afb65bf5c5a07f9e702b565b51d9f459ca3fa0310", "2720503ebf27ef4fc779c804602f6566c0e56ab9805fd66e46da7d9e87fe5f0a" ]
24.123
0
true
{"names": ["Kim", "Riley", "Avery", "Robin"], "speakers": [0, 1, 2, 3, 2, 1, 2, 0], "intro_turns": 4, "target_turn": 5, "fact": {"kind": "drink", "value": "an espresso", "speaker": 1, "ask": "start the morning with an espresso"}, "facts": [{"kind": "drink", "value": "orange juice", "speaker": 2, "ask": "start the morni...
fact_attribution-train-00783-q2
fact_attribution-train-00783
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they like to collect old maps in their free time?
[ "Kim", "Riley", "Avery", "Robin" ]
[ "A", "B", "C", "D" ]
A
0
sopro
PyTorch MPS (Apple M5 Pro)
[ "5b1662f2226a7e6763aeda046320a10c39bc9467b65e6e06ebc8f026ca5a4b68", "53acc0b07cf4ebbccfa1e11a9e7900eb500f816e90fb467c1f29863be4e5a11b", "155767ed4cb55fa495799e4afb65bf5c5a07f9e702b565b51d9f459ca3fa0310", "2720503ebf27ef4fc779c804602f6566c0e56ab9805fd66e46da7d9e87fe5f0a" ]
24.123
0
true
{"names": ["Kim", "Riley", "Avery", "Robin"], "speakers": [0, 1, 2, 3, 2, 1, 2, 0], "intro_turns": 4, "target_turn": 7, "fact": {"kind": "hobby", "value": "collect old maps", "speaker": 0, "ask": "like to collect old maps in their free time"}, "facts": [{"kind": "drink", "value": "orange juice", "speaker": 2, "ask": "s...
fact_attribution-train-00469-q0
fact_attribution-train-00469
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they grew up in Toronto?
[ "Sam", "Morgan", "Taylor", "Robin" ]
[ "A", "B", "C", "D" ]
D
3
pocket
mlx-audio 0.5.8 (Apple M5 Pro)
[ "4f5149c12a4263270fe8b3e4e13415d30eff5b763311563523ad077e3d9cb59c", "1677cdecf516ec8f2af54afb320e1f4419173552241cdce87e70d967098a54cf", "3efe1856feafdffa86bd058b8f9d3851056ccc25e43bf07fe16983a52d4bbd7a", "2a6dc1f5b9978d486d3b353e6d67e29b31b6f91a72d0f5968417354f202614ff" ]
20.128
0
true
{"names": ["Morgan", "Sam", "Taylor", "Robin"], "speakers": [0, 1, 2, 3, 3, 0, 1, 2], "intro_turns": 4, "target_turn": 4, "fact": {"kind": "hometown", "value": "Toronto", "speaker": 3, "ask": "grew up in Toronto"}, "facts": [{"kind": "hometown", "value": "Toronto", "speaker": 3, "ask": "grew up in Toronto"}, {"kind": "...
fact_attribution-train-00469-q1
fact_attribution-train-00469
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they grew up in Vienna?
[ "Taylor", "Sam", "Morgan", "Robin" ]
[ "A", "B", "C", "D" ]
C
2
pocket
mlx-audio 0.5.8 (Apple M5 Pro)
[ "4f5149c12a4263270fe8b3e4e13415d30eff5b763311563523ad077e3d9cb59c", "1677cdecf516ec8f2af54afb320e1f4419173552241cdce87e70d967098a54cf", "3efe1856feafdffa86bd058b8f9d3851056ccc25e43bf07fe16983a52d4bbd7a", "2a6dc1f5b9978d486d3b353e6d67e29b31b6f91a72d0f5968417354f202614ff" ]
20.128
0
true
{"names": ["Morgan", "Sam", "Taylor", "Robin"], "speakers": [0, 1, 2, 3, 3, 0, 1, 2], "intro_turns": 4, "target_turn": 5, "fact": {"kind": "hometown", "value": "Vienna", "speaker": 0, "ask": "grew up in Vienna"}, "facts": [{"kind": "hometown", "value": "Toronto", "speaker": 3, "ask": "grew up in Toronto"}, {"kind": "ho...
fact_attribution-train-00469-q2
fact_attribution-train-00469
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they recently traveled to Chile?
[ "Taylor", "Sam", "Morgan", "Robin" ]
[ "A", "B", "C", "D" ]
A
0
pocket
mlx-audio 0.5.8 (Apple M5 Pro)
[ "4f5149c12a4263270fe8b3e4e13415d30eff5b763311563523ad077e3d9cb59c", "1677cdecf516ec8f2af54afb320e1f4419173552241cdce87e70d967098a54cf", "3efe1856feafdffa86bd058b8f9d3851056ccc25e43bf07fe16983a52d4bbd7a", "2a6dc1f5b9978d486d3b353e6d67e29b31b6f91a72d0f5968417354f202614ff" ]
20.128
0
true
{"names": ["Morgan", "Sam", "Taylor", "Robin"], "speakers": [0, 1, 2, 3, 3, 0, 1, 2], "intro_turns": 4, "target_turn": 7, "fact": {"kind": "travel", "value": "Chile", "speaker": 2, "ask": "recently traveled to Chile"}, "facts": [{"kind": "hometown", "value": "Toronto", "speaker": 3, "ask": "grew up in Toronto"}, {"kind...
fact_attribution-train-00780-q0
fact_attribution-train-00780
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they grow herbs on their balcony?
[ "Jamie", "Sam", "Robin", "Alex" ]
[ "A", "B", "C", "D" ]
B
1
omni
PyTorch CUDA (RTX 3090)
[ "63014c9514cf1d7ae17407a5b3f3a1e8ce515ed362a562471f521c49528631c9", "e1e0b0aaee9fe0f26aed294f3e7d9d387b425b290da7e34464d599ddf10622bf", "e7a2a7e521e3454d36bb408919e46ffa819e3d98dd8a309d21db444c836b5958", "c22f64ff6c3a8e535acb9cb8a29ab1debc8989be73621772e2c96ba3f8988439" ]
19.564
0
true
{"names": ["Jamie", "Sam", "Robin", "Alex"], "speakers": [0, 1, 2, 3, 1, 0, 2, 3], "intro_turns": 4, "target_turn": 4, "fact": {"kind": "garden", "value": "herbs", "speaker": 1, "ask": "grow herbs on their balcony"}, "facts": [{"kind": "garden", "value": "herbs", "speaker": 1, "ask": "grow herbs on their balcony"}, {"k...
fact_attribution-train-00780-q1
fact_attribution-train-00780
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they grow lavender on their balcony?
[ "Sam", "Jamie", "Alex", "Robin" ]
[ "A", "B", "C", "D" ]
B
1
omni
PyTorch CUDA (RTX 3090)
[ "63014c9514cf1d7ae17407a5b3f3a1e8ce515ed362a562471f521c49528631c9", "e1e0b0aaee9fe0f26aed294f3e7d9d387b425b290da7e34464d599ddf10622bf", "e7a2a7e521e3454d36bb408919e46ffa819e3d98dd8a309d21db444c836b5958", "c22f64ff6c3a8e535acb9cb8a29ab1debc8989be73621772e2c96ba3f8988439" ]
19.564
0
true
{"names": ["Jamie", "Sam", "Robin", "Alex"], "speakers": [0, 1, 2, 3, 1, 0, 2, 3], "intro_turns": 4, "target_turn": 5, "fact": {"kind": "garden", "value": "lavender", "speaker": 0, "ask": "grow lavender on their balcony"}, "facts": [{"kind": "garden", "value": "herbs", "speaker": 1, "ask": "grow herbs on their balcony"...
fact_attribution-train-00780-q2
fact_attribution-train-00780
fact_attribution
Listen to the conversation. At the start, each speaker introduces themselves by name. Who said they drive an old red van?
[ "Alex", "Robin", "Jamie", "Sam" ]
[ "A", "B", "C", "D" ]
B
1
omni
PyTorch CUDA (RTX 3090)
[ "63014c9514cf1d7ae17407a5b3f3a1e8ce515ed362a562471f521c49528631c9", "e1e0b0aaee9fe0f26aed294f3e7d9d387b425b290da7e34464d599ddf10622bf", "e7a2a7e521e3454d36bb408919e46ffa819e3d98dd8a309d21db444c836b5958", "c22f64ff6c3a8e535acb9cb8a29ab1debc8989be73621772e2c96ba3f8988439" ]
19.564
0
true
{"names": ["Jamie", "Sam", "Robin", "Alex"], "speakers": [0, 1, 2, 3, 1, 0, 2, 3], "intro_turns": 4, "target_turn": 6, "fact": {"kind": "vehicle", "value": "an old red van", "speaker": 2, "ask": "drive an old red van"}, "facts": [{"kind": "garden", "value": "herbs", "speaker": 1, "ask": "grow herbs on their balcony"}, ...
End of preview. Expand in Data Studio

Decider Audio – fact attribution

Part of the Decider Audio collection.

Who said which fact? 2–4 speakers introduce themselves by name, then take turns stating personal facts. Each question asks for one fact, for example Who said they grew up in Denver?, and the choices are the speakers' names. Only the voice links a fact to the name its speaker introduced, so a transcript without speaker labels cannot answer it.

1,300 conversations, 3,900 questions (rows), 8.3 hours of audio, 10,833 rendered turns, 1,591 reference voices.

Split Conversations Questions (rows) Hours Rows failing checks
train 800 2,400 5.1 0
validation 100 300 0.6 0
test 400 1,200 2.6 0
from datasets import load_dataset

ds = load_dataset("SebastianBodza/decider-audio-fact-attribution", split="test")

A conversation looks like this (speaker labels are not part of the audio).

[Taylor] Hey, I'm Taylor. Nice to meet you all.
[Jamie]  Hi everyone, I'm Jamie.
[Riley]  Hi there, I'm Riley.
[Avery]  Hi, this is Avery.
[Taylor] I've been working as a carpenter for a few years now.
[Jamie]  For me it's different. These days I work as a chef.
[Avery]  So, I'm originally from Cape Town.
[Taylor] Oh nice. I grew up in Toronto.

Who said they work as a chef?          Riley / Jamie / Taylor / Avery
Who said they grew up in Cape Town?    Jamie / Taylor / Avery / Riley
Who said they grew up in Toronto?      Jamie / Riley / Avery / Taylor

How it is made

  • Script. Each conversation follows a fixed shape.
    • Every speaker first says a greeting with their name ("Hello everyone, Robin here.") and no fact. The greetings within a conversation are worded differently.
    • Then come 6 fact turns (2 or 3 speakers) or 4 fact turns (4 speakers), in random order with no speaker twice in a row. The order does not follow the introduction order.
    • Facts come in pairs of the same kind with different values, as a natural exchange. For example "I moved here ten years ago." is followed by "Oh nice. I moved here five years ago." So the question's topic alone does not identify the turn.
    • Facts come from 21 kinds (home town, job, pet, instrument, sport, food, travel, language, commute, hobby, home, allergy, fear, morning drink, vehicle, move, season, studies, balcony garden, reading, weekend plans). Each kind has 2–3 phrasings and 4–16 values. No turn names a speaker.
    • The whole script stays within 330 characters, so a conversation fits into 30 seconds (median 23.0 s, max 32.8 s, and 14 conversations are longer).
  • Questions. Three fact turns per conversation are asked, each as its own row. The choices are all speakers of the conversation, including those without a fact turn, shuffled per question.
  • Voices. The voices come from a private collection of English reference recordings, the same pool as decider-audio-speaker-count. No speaker appears in more than one split. The voices within one conversation do not sound too similar (ECAPA similarity below 0.6).
  • One model per conversation. All turns of a conversation are rendered by the same TTS model. The four models are balanced within every split and speaker count.
  • Levels. Every turn is trimmed and set to -23.0 dB RMS, so loudness is no cue. Turns are separated by 0.25–0.6 s of silence.

Checks

Every turn was transcribed with Systran/faster-whisper-base.en. A turn fails if any of these holds.

  • WER is above 0.5.
  • It is spoken faster than 30 or slower than 6 characters per second (runaway renders that add silence or noise).
  • It adds words.
  • It is an introduction and the name is not recognized.

Every turn is also checked with ECAPA embeddings (speechbrain/spkrec-ecapa-voxceleb). A turn fails if any of these holds.

  • Its similarity to the same speaker's other turns is below 0.4.
  • That similarity is not at least 0.1 above its similarity to another speaker.
  • Its similarity to another speaker reaches 0.55.

A failing turn is rendered again with the same voice and text and a new seed, up to 5 renders. A second repair pass rendered 43 turns again with a new seed, and with a new voice for their speaker where that still failed. A turn now also passes when its only problem is the word error rate or extra words and its transcript sounds like the text. This check compares phonemes after numbers and units are spelled out, so "topup" and "top up" or "1960" and "nineteen sixty" match. A turn that leaves out part of its text does not pass this way.

  • Re-rendered turns: 407.
  • Turns still failing: 0. Rows containing them have passes_checks = false.
  • Mean WER: 0.009.

The slow-render check (below 6 characters per second) exists because runaway renders took up to 38 s for a 27-character introduction. Whisper still transcribes these correctly because it skips the silence and noise. 19 conversations are longer than 30 s. 5 of them are long because of a failing runaway clip, and the rest last 30–32 s.

TTS model Checkpoint Conversations Clips Re-rendered clips
pocket mlx-community/pocket-tts 330 2,750 184
sopro samuel-vitorino/sopro-v2-turbo 326 2,716 64
qwen Qwen/Qwen3-TTS-12Hz-0.6B-Base 323 2,692 147
omni k2-fsa/OmniVoice 321 2,675 12

Answer balance

Split Answer keys (count)
train A 823, B 879, C 490, D 208
validation A 110, B 100, C 60, D 30
test A 464, B 406, C 237, D 93

The next table shows the introduction position of the correct speaker. A uniform spread means the order carries no information.

Speakers Questions Answer is the 1st / 2nd / 3rd / 4th introduced speaker
2 1,305 661 / 644
3 1,299 435 / 420 / 444
4 1,296 331 / 329 / 321 / 315

Asked fact kinds: allergy 233, commute 177, drink 118, fear 175, food 203, garden 214, hobby 166, home 198, hometown 214, instrument 226, job 161, language 186, moved 210, pet 189, reading 156, season 188, sport 190, studies 204, travel 171, vehicle 200, weekend 121.

Limitations

  • The texts are templates. They are natural but repetitive, and the same phrasings and values appear in every split. The task tests voice attribution, not language understanding.
  • All speech is synthetic (voice clones). Clones can sound less distinct than the real speakers.
  • With 4 speakers there are only 4 fact turns, so some speakers state no fact. They remain valid choices.
  • Speakers alternate strictly with 2 speakers, so the two turns of a pair always belong to different people. A transcript still cannot tell which name is which.

Fields

Field Meaning
audio The whole conversation as one FLAC (24 kHz, mono). All questions of a conversation share it.
conversation_id Groups the rows of one conversation. Split by it, never by row.
instruction, choices, choice_keys The question and the speakers' names (A, B, …), shuffled per question.
answer, answer_index The correct key and its index.
tts_model, tts_runtime The TTS model of every turn, and its software runtime (28 late re-renders are labelled with the runtime of the rest of their conversation).
voice_ids Reference voice ids, in introduction order.
passes_checks The introductions and the asked turn passed the text and speaker checks.
details JSON with names, speaker of each turn, all facts with their speaker, the asked fact, and per turn its text, span in seconds, transcript, WER and speaker similarities.

License and attribution

  • Speech. Rendered with Pocket TTS (CC BY 4.0, attribution), Sopro (Apache-2.0), Qwen3-TTS (Apache-2.0) and OmniVoice (reported as CC BY-NC 4.0, verify on k2-fsa/OmniVoice).
  • Reference voices. Private collection, not redistributed. Voice ids are anonymous hashes.
  • Speaker names in the conversations are invented.

The dataset is marked license: other. Do not use it commercially.

Downloads last month
19

Collection including SebastianBodza/decider-audio-fact-attribution