Upload folder using huggingface_hub
Browse filesThis view is limited to 50 files because it contains too many changes.
See raw diff
- .gitattributes +18 -0
- examples/2AC/AudioCaps-Test/dataset_info.json +160 -0
- examples/2AC/AudioCaps-Test/sample_0.wav +0 -0
- examples/2AC/AudioCaps-Test/sample_1.wav +0 -0
- examples/2AC/AudioCaps-Test/sample_2.wav +0 -0
- examples/2AC/AudioCaps-Test/state.json +24 -0
- examples/2AC/WavCaps-Test/dataset_info.json +156 -0
- examples/2AC/WavCaps-Test/sample_0.wav +0 -0
- examples/2AC/WavCaps-Test/sample_1.wav +0 -0
- examples/2AC/WavCaps-Test/sample_2.wav +0 -0
- examples/2AC/WavCaps-Test/state.json +24 -0
- examples/2AQA/AudioCaps-QA-Test/dataset_info.json +164 -0
- examples/2AQA/AudioCaps-QA-Test/sample_0.wav +0 -0
- examples/2AQA/AudioCaps-QA-Test/sample_1.wav +0 -0
- examples/2AQA/AudioCaps-QA-Test/sample_2.wav +0 -0
- examples/2AQA/AudioCaps-QA-Test/state.json +24 -0
- examples/2AQA/Clotho-AQA-Test/dataset_info.json +147 -0
- examples/2AQA/Clotho-AQA-Test/sample_0.wav +0 -0
- examples/2AQA/Clotho-AQA-Test/sample_1.wav +0 -0
- examples/2AQA/Clotho-AQA-Test/sample_2.wav +0 -0
- examples/2AQA/Clotho-AQA-Test/state.json +24 -0
- examples/2AQA/WavCaps-QA-Test/dataset_info.json +160 -0
- examples/2AQA/WavCaps-QA-Test/sample_0.wav +0 -0
- examples/2AQA/WavCaps-QA-Test/sample_1.wav +0 -0
- examples/2AQA/WavCaps-QA-Test/sample_2.wav +0 -0
- examples/2AQA/WavCaps-QA-Test/state.json +24 -0
- examples/2AR/VoxCeleb-Accent-Test/dataset_info.json +168 -0
- examples/2AR/VoxCeleb-Accent-Test/sample_0.wav +0 -0
- examples/2AR/VoxCeleb-Accent-Test/sample_1.wav +0 -0
- examples/2AR/VoxCeleb-Accent-Test/sample_2.wav +0 -0
- examples/2AR/VoxCeleb-Accent-Test/state.json +24 -0
- examples/2ASR/Common-Voice-15-En-Test/dataset_info.json +188 -0
- examples/2ASR/Common-Voice-15-En-Test/sample_0.wav +0 -0
- examples/2ASR/Common-Voice-15-En-Test/sample_1.wav +0 -0
- examples/2ASR/Common-Voice-15-En-Test/sample_2.wav +0 -0
- examples/2ASR/Common-Voice-15-En-Test/state.json +24 -0
- examples/2ASR/Earnings21-Test/dataset_info.json +152 -0
- examples/2ASR/Earnings21-Test/state.json +24 -0
- examples/2ASR/Earnings22-Test/dataset_info.json +152 -0
- examples/2ASR/Earnings22-Test/state.json +24 -0
- examples/2ASR/GigaSpeech-Test2/dataset_info.json +180 -0
- examples/2ASR/GigaSpeech-Test2/sample_0.wav +0 -0
- examples/2ASR/GigaSpeech-Test2/sample_1.wav +0 -0
- examples/2ASR/GigaSpeech-Test2/sample_2.wav +0 -0
- examples/2ASR/GigaSpeech-Test2/state.json +24 -0
- examples/2ASR/IMDA-Part1-ASR-Test/dataset_info.json +200 -0
- examples/2ASR/IMDA-Part1-ASR-Test/sample_0.wav +0 -0
- examples/2ASR/IMDA-Part1-ASR-Test/sample_1.wav +0 -0
- examples/2ASR/IMDA-Part1-ASR-Test/sample_2.wav +0 -0
- examples/2ASR/IMDA-Part1-ASR-Test/state.json +24 -0
.gitattributes
CHANGED
|
@@ -60,3 +60,21 @@ examples/SQA/CN-College-Listen-MCQ-Test/sample_0.wav filter=lfs diff=lfs merge=l
|
|
| 60 |
examples/SQA/Spoken-Squad-Test/sample_0.wav filter=lfs diff=lfs merge=lfs -text
|
| 61 |
examples/SQA/Spoken-Squad-Test/sample_1.wav filter=lfs diff=lfs merge=lfs -text
|
| 62 |
examples/SQA/Spoken-Squad-Test/sample_2.wav filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 60 |
examples/SQA/Spoken-Squad-Test/sample_0.wav filter=lfs diff=lfs merge=lfs -text
|
| 61 |
examples/SQA/Spoken-Squad-Test/sample_1.wav filter=lfs diff=lfs merge=lfs -text
|
| 62 |
examples/SQA/Spoken-Squad-Test/sample_2.wav filter=lfs diff=lfs merge=lfs -text
|
| 63 |
+
examples/2ASR/Earnings21-Test/sample_0.wav filter=lfs diff=lfs merge=lfs -text
|
| 64 |
+
examples/2ASR/Earnings21-Test/sample_1.wav filter=lfs diff=lfs merge=lfs -text
|
| 65 |
+
examples/2ASR/Earnings21-Test/sample_2.wav filter=lfs diff=lfs merge=lfs -text
|
| 66 |
+
examples/2ASR/Earnings22-Test/sample_0.wav filter=lfs diff=lfs merge=lfs -text
|
| 67 |
+
examples/2ASR/Earnings22-Test/sample_1.wav filter=lfs diff=lfs merge=lfs -text
|
| 68 |
+
examples/2ASR/Earnings22-Test/sample_2.wav filter=lfs diff=lfs merge=lfs -text
|
| 69 |
+
examples/2ASR/Tedlium3-Long-form-Test/sample_0.wav filter=lfs diff=lfs merge=lfs -text
|
| 70 |
+
examples/2ASR/Tedlium3-Long-form-Test/sample_1.wav filter=lfs diff=lfs merge=lfs -text
|
| 71 |
+
examples/2ASR/Tedlium3-Long-form-Test/sample_2.wav filter=lfs diff=lfs merge=lfs -text
|
| 72 |
+
examples/2SQA/CN-College-Listen-MCQ-Test/sample_0.wav filter=lfs diff=lfs merge=lfs -text
|
| 73 |
+
examples/2SQA/Public-SG-Speech-QA-Test/sample_0.wav filter=lfs diff=lfs merge=lfs -text
|
| 74 |
+
examples/2SQA/SLUE-P2-SQA5-Test/sample_0.wav filter=lfs diff=lfs merge=lfs -text
|
| 75 |
+
examples/2SQA/SLUE-P2-SQA5-Test/sample_1.wav filter=lfs diff=lfs merge=lfs -text
|
| 76 |
+
examples/2SQA/SLUE-P2-SQA5-Test/sample_2.wav filter=lfs diff=lfs merge=lfs -text
|
| 77 |
+
examples/2SQA/Spoken-Squad-Test/sample_0.wav filter=lfs diff=lfs merge=lfs -text
|
| 78 |
+
examples/2SQA/Spoken-Squad-Test/sample_1.wav filter=lfs diff=lfs merge=lfs -text
|
| 79 |
+
examples/2SQA/Spoken-Squad-Test/sample_2.wav filter=lfs diff=lfs merge=lfs -text
|
| 80 |
+
examples/SQA/DREAM-TTS-MCQ-Test/sample_1.wav filter=lfs diff=lfs merge=lfs -text
|
examples/2AC/AudioCaps-Test/dataset_info.json
ADDED
|
@@ -0,0 +1,160 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"citation": "",
|
| 3 |
+
"description": "",
|
| 4 |
+
"features": {
|
| 5 |
+
"context": {
|
| 6 |
+
"text": {
|
| 7 |
+
"dtype": "string",
|
| 8 |
+
"_type": "Value"
|
| 9 |
+
},
|
| 10 |
+
"audio": {
|
| 11 |
+
"sampling_rate": 16000,
|
| 12 |
+
"_type": "Audio"
|
| 13 |
+
}
|
| 14 |
+
},
|
| 15 |
+
"instruction": {
|
| 16 |
+
"audio": {
|
| 17 |
+
"dtype": "null",
|
| 18 |
+
"_type": "Value"
|
| 19 |
+
},
|
| 20 |
+
"text": {
|
| 21 |
+
"dtype": "string",
|
| 22 |
+
"_type": "Value"
|
| 23 |
+
}
|
| 24 |
+
},
|
| 25 |
+
"answer": {
|
| 26 |
+
"audio": {
|
| 27 |
+
"dtype": "null",
|
| 28 |
+
"_type": "Value"
|
| 29 |
+
},
|
| 30 |
+
"text": {
|
| 31 |
+
"dtype": "string",
|
| 32 |
+
"_type": "Value"
|
| 33 |
+
}
|
| 34 |
+
},
|
| 35 |
+
"other_attributes": {
|
| 36 |
+
"audiocap_id": {
|
| 37 |
+
"dtype": "string",
|
| 38 |
+
"_type": "Value"
|
| 39 |
+
},
|
| 40 |
+
"start_time": {
|
| 41 |
+
"dtype": "string",
|
| 42 |
+
"_type": "Value"
|
| 43 |
+
},
|
| 44 |
+
"youtube_id": {
|
| 45 |
+
"dtype": "string",
|
| 46 |
+
"_type": "Value"
|
| 47 |
+
}
|
| 48 |
+
},
|
| 49 |
+
"salmonn_7b": {
|
| 50 |
+
"answer": {
|
| 51 |
+
"dtype": "string",
|
| 52 |
+
"_type": "Value"
|
| 53 |
+
},
|
| 54 |
+
"model_prediction": {
|
| 55 |
+
"dtype": "string",
|
| 56 |
+
"_type": "Value"
|
| 57 |
+
},
|
| 58 |
+
"task_type": {
|
| 59 |
+
"dtype": "string",
|
| 60 |
+
"_type": "Value"
|
| 61 |
+
},
|
| 62 |
+
"text": {
|
| 63 |
+
"dtype": "string",
|
| 64 |
+
"_type": "Value"
|
| 65 |
+
}
|
| 66 |
+
},
|
| 67 |
+
"wavllm_fairseq": {
|
| 68 |
+
"answer": {
|
| 69 |
+
"dtype": "string",
|
| 70 |
+
"_type": "Value"
|
| 71 |
+
},
|
| 72 |
+
"model_prediction": {
|
| 73 |
+
"dtype": "string",
|
| 74 |
+
"_type": "Value"
|
| 75 |
+
},
|
| 76 |
+
"task_type": {
|
| 77 |
+
"dtype": "string",
|
| 78 |
+
"_type": "Value"
|
| 79 |
+
},
|
| 80 |
+
"text": {
|
| 81 |
+
"dtype": "string",
|
| 82 |
+
"_type": "Value"
|
| 83 |
+
}
|
| 84 |
+
},
|
| 85 |
+
"Qwen2-Audio-7B-Instruct": {
|
| 86 |
+
"answer": {
|
| 87 |
+
"dtype": "string",
|
| 88 |
+
"_type": "Value"
|
| 89 |
+
},
|
| 90 |
+
"model_prediction": {
|
| 91 |
+
"dtype": "string",
|
| 92 |
+
"_type": "Value"
|
| 93 |
+
},
|
| 94 |
+
"task_type": {
|
| 95 |
+
"dtype": "string",
|
| 96 |
+
"_type": "Value"
|
| 97 |
+
},
|
| 98 |
+
"text": {
|
| 99 |
+
"dtype": "string",
|
| 100 |
+
"_type": "Value"
|
| 101 |
+
}
|
| 102 |
+
},
|
| 103 |
+
"whisper_large_v3_with_llama_3_8b_instruct": {
|
| 104 |
+
"answer": {
|
| 105 |
+
"dtype": "string",
|
| 106 |
+
"_type": "Value"
|
| 107 |
+
},
|
| 108 |
+
"model_prediction": {
|
| 109 |
+
"dtype": "string",
|
| 110 |
+
"_type": "Value"
|
| 111 |
+
},
|
| 112 |
+
"task_type": {
|
| 113 |
+
"dtype": "string",
|
| 114 |
+
"_type": "Value"
|
| 115 |
+
},
|
| 116 |
+
"text": {
|
| 117 |
+
"dtype": "string",
|
| 118 |
+
"_type": "Value"
|
| 119 |
+
}
|
| 120 |
+
},
|
| 121 |
+
"mowe_audio": {
|
| 122 |
+
"answer": {
|
| 123 |
+
"dtype": "string",
|
| 124 |
+
"_type": "Value"
|
| 125 |
+
},
|
| 126 |
+
"model_prediction": {
|
| 127 |
+
"dtype": "string",
|
| 128 |
+
"_type": "Value"
|
| 129 |
+
},
|
| 130 |
+
"task_type": {
|
| 131 |
+
"dtype": "string",
|
| 132 |
+
"_type": "Value"
|
| 133 |
+
},
|
| 134 |
+
"text": {
|
| 135 |
+
"dtype": "string",
|
| 136 |
+
"_type": "Value"
|
| 137 |
+
}
|
| 138 |
+
},
|
| 139 |
+
"qwen_audio_chat": {
|
| 140 |
+
"answer": {
|
| 141 |
+
"dtype": "string",
|
| 142 |
+
"_type": "Value"
|
| 143 |
+
},
|
| 144 |
+
"model_prediction": {
|
| 145 |
+
"dtype": "string",
|
| 146 |
+
"_type": "Value"
|
| 147 |
+
},
|
| 148 |
+
"task_type": {
|
| 149 |
+
"dtype": "string",
|
| 150 |
+
"_type": "Value"
|
| 151 |
+
},
|
| 152 |
+
"text": {
|
| 153 |
+
"dtype": "string",
|
| 154 |
+
"_type": "Value"
|
| 155 |
+
}
|
| 156 |
+
}
|
| 157 |
+
},
|
| 158 |
+
"homepage": "",
|
| 159 |
+
"license": ""
|
| 160 |
+
}
|
examples/2AC/AudioCaps-Test/sample_0.wav
ADDED
|
Binary file (320 kB). View file
|
|
|
examples/2AC/AudioCaps-Test/sample_1.wav
ADDED
|
Binary file (320 kB). View file
|
|
|
examples/2AC/AudioCaps-Test/sample_2.wav
ADDED
|
Binary file (315 kB). View file
|
|
|
examples/2AC/AudioCaps-Test/state.json
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_data_files": [
|
| 3 |
+
{
|
| 4 |
+
"filename": "data-00000-of-00001.arrow"
|
| 5 |
+
}
|
| 6 |
+
],
|
| 7 |
+
"_fingerprint": "e736bf1821a473f3",
|
| 8 |
+
"_format_columns": [
|
| 9 |
+
"context",
|
| 10 |
+
"instruction",
|
| 11 |
+
"answer",
|
| 12 |
+
"other_attributes",
|
| 13 |
+
"salmonn_7b",
|
| 14 |
+
"wavllm_fairseq",
|
| 15 |
+
"Qwen2-Audio-7B-Instruct",
|
| 16 |
+
"whisper_large_v3_with_llama_3_8b_instruct",
|
| 17 |
+
"mowe_audio",
|
| 18 |
+
"qwen_audio_chat"
|
| 19 |
+
],
|
| 20 |
+
"_format_kwargs": {},
|
| 21 |
+
"_format_type": null,
|
| 22 |
+
"_output_all_columns": false,
|
| 23 |
+
"_split": null
|
| 24 |
+
}
|
examples/2AC/WavCaps-Test/dataset_info.json
ADDED
|
@@ -0,0 +1,156 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"citation": "",
|
| 3 |
+
"description": "",
|
| 4 |
+
"features": {
|
| 5 |
+
"context": {
|
| 6 |
+
"text": {
|
| 7 |
+
"dtype": "string",
|
| 8 |
+
"_type": "Value"
|
| 9 |
+
},
|
| 10 |
+
"audio": {
|
| 11 |
+
"sampling_rate": 16000,
|
| 12 |
+
"_type": "Audio"
|
| 13 |
+
}
|
| 14 |
+
},
|
| 15 |
+
"instruction": {
|
| 16 |
+
"audio": {
|
| 17 |
+
"dtype": "null",
|
| 18 |
+
"_type": "Value"
|
| 19 |
+
},
|
| 20 |
+
"text": {
|
| 21 |
+
"dtype": "string",
|
| 22 |
+
"_type": "Value"
|
| 23 |
+
}
|
| 24 |
+
},
|
| 25 |
+
"answer": {
|
| 26 |
+
"audio": {
|
| 27 |
+
"dtype": "null",
|
| 28 |
+
"_type": "Value"
|
| 29 |
+
},
|
| 30 |
+
"text": {
|
| 31 |
+
"dtype": "string",
|
| 32 |
+
"_type": "Value"
|
| 33 |
+
}
|
| 34 |
+
},
|
| 35 |
+
"other_attributes": {
|
| 36 |
+
"audio_path": {
|
| 37 |
+
"dtype": "string",
|
| 38 |
+
"_type": "Value"
|
| 39 |
+
},
|
| 40 |
+
"duration": {
|
| 41 |
+
"dtype": "string",
|
| 42 |
+
"_type": "Value"
|
| 43 |
+
}
|
| 44 |
+
},
|
| 45 |
+
"salmonn_7b": {
|
| 46 |
+
"answer": {
|
| 47 |
+
"dtype": "string",
|
| 48 |
+
"_type": "Value"
|
| 49 |
+
},
|
| 50 |
+
"model_prediction": {
|
| 51 |
+
"dtype": "string",
|
| 52 |
+
"_type": "Value"
|
| 53 |
+
},
|
| 54 |
+
"task_type": {
|
| 55 |
+
"dtype": "string",
|
| 56 |
+
"_type": "Value"
|
| 57 |
+
},
|
| 58 |
+
"text": {
|
| 59 |
+
"dtype": "string",
|
| 60 |
+
"_type": "Value"
|
| 61 |
+
}
|
| 62 |
+
},
|
| 63 |
+
"wavllm_fairseq": {
|
| 64 |
+
"answer": {
|
| 65 |
+
"dtype": "string",
|
| 66 |
+
"_type": "Value"
|
| 67 |
+
},
|
| 68 |
+
"model_prediction": {
|
| 69 |
+
"dtype": "string",
|
| 70 |
+
"_type": "Value"
|
| 71 |
+
},
|
| 72 |
+
"task_type": {
|
| 73 |
+
"dtype": "string",
|
| 74 |
+
"_type": "Value"
|
| 75 |
+
},
|
| 76 |
+
"text": {
|
| 77 |
+
"dtype": "string",
|
| 78 |
+
"_type": "Value"
|
| 79 |
+
}
|
| 80 |
+
},
|
| 81 |
+
"Qwen2-Audio-7B-Instruct": {
|
| 82 |
+
"answer": {
|
| 83 |
+
"dtype": "string",
|
| 84 |
+
"_type": "Value"
|
| 85 |
+
},
|
| 86 |
+
"model_prediction": {
|
| 87 |
+
"dtype": "string",
|
| 88 |
+
"_type": "Value"
|
| 89 |
+
},
|
| 90 |
+
"task_type": {
|
| 91 |
+
"dtype": "string",
|
| 92 |
+
"_type": "Value"
|
| 93 |
+
},
|
| 94 |
+
"text": {
|
| 95 |
+
"dtype": "string",
|
| 96 |
+
"_type": "Value"
|
| 97 |
+
}
|
| 98 |
+
},
|
| 99 |
+
"whisper_large_v3_with_llama_3_8b_instruct": {
|
| 100 |
+
"answer": {
|
| 101 |
+
"dtype": "string",
|
| 102 |
+
"_type": "Value"
|
| 103 |
+
},
|
| 104 |
+
"model_prediction": {
|
| 105 |
+
"dtype": "string",
|
| 106 |
+
"_type": "Value"
|
| 107 |
+
},
|
| 108 |
+
"task_type": {
|
| 109 |
+
"dtype": "string",
|
| 110 |
+
"_type": "Value"
|
| 111 |
+
},
|
| 112 |
+
"text": {
|
| 113 |
+
"dtype": "string",
|
| 114 |
+
"_type": "Value"
|
| 115 |
+
}
|
| 116 |
+
},
|
| 117 |
+
"mowe_audio": {
|
| 118 |
+
"answer": {
|
| 119 |
+
"dtype": "string",
|
| 120 |
+
"_type": "Value"
|
| 121 |
+
},
|
| 122 |
+
"model_prediction": {
|
| 123 |
+
"dtype": "string",
|
| 124 |
+
"_type": "Value"
|
| 125 |
+
},
|
| 126 |
+
"task_type": {
|
| 127 |
+
"dtype": "string",
|
| 128 |
+
"_type": "Value"
|
| 129 |
+
},
|
| 130 |
+
"text": {
|
| 131 |
+
"dtype": "string",
|
| 132 |
+
"_type": "Value"
|
| 133 |
+
}
|
| 134 |
+
},
|
| 135 |
+
"qwen_audio_chat": {
|
| 136 |
+
"answer": {
|
| 137 |
+
"dtype": "string",
|
| 138 |
+
"_type": "Value"
|
| 139 |
+
},
|
| 140 |
+
"model_prediction": {
|
| 141 |
+
"dtype": "string",
|
| 142 |
+
"_type": "Value"
|
| 143 |
+
},
|
| 144 |
+
"task_type": {
|
| 145 |
+
"dtype": "string",
|
| 146 |
+
"_type": "Value"
|
| 147 |
+
},
|
| 148 |
+
"text": {
|
| 149 |
+
"dtype": "string",
|
| 150 |
+
"_type": "Value"
|
| 151 |
+
}
|
| 152 |
+
}
|
| 153 |
+
},
|
| 154 |
+
"homepage": "",
|
| 155 |
+
"license": ""
|
| 156 |
+
}
|
examples/2AC/WavCaps-Test/sample_0.wav
ADDED
|
Binary file (32 kB). View file
|
|
|
examples/2AC/WavCaps-Test/sample_1.wav
ADDED
|
Binary file (268 kB). View file
|
|
|
examples/2AC/WavCaps-Test/sample_2.wav
ADDED
|
Binary file (500 kB). View file
|
|
|
examples/2AC/WavCaps-Test/state.json
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_data_files": [
|
| 3 |
+
{
|
| 4 |
+
"filename": "data-00000-of-00001.arrow"
|
| 5 |
+
}
|
| 6 |
+
],
|
| 7 |
+
"_fingerprint": "742ab313af054565",
|
| 8 |
+
"_format_columns": [
|
| 9 |
+
"context",
|
| 10 |
+
"instruction",
|
| 11 |
+
"answer",
|
| 12 |
+
"other_attributes",
|
| 13 |
+
"salmonn_7b",
|
| 14 |
+
"wavllm_fairseq",
|
| 15 |
+
"Qwen2-Audio-7B-Instruct",
|
| 16 |
+
"whisper_large_v3_with_llama_3_8b_instruct",
|
| 17 |
+
"mowe_audio",
|
| 18 |
+
"qwen_audio_chat"
|
| 19 |
+
],
|
| 20 |
+
"_format_kwargs": {},
|
| 21 |
+
"_format_type": null,
|
| 22 |
+
"_output_all_columns": false,
|
| 23 |
+
"_split": null
|
| 24 |
+
}
|
examples/2AQA/AudioCaps-QA-Test/dataset_info.json
ADDED
|
@@ -0,0 +1,164 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"citation": "",
|
| 3 |
+
"description": "",
|
| 4 |
+
"features": {
|
| 5 |
+
"context": {
|
| 6 |
+
"text": {
|
| 7 |
+
"dtype": "string",
|
| 8 |
+
"_type": "Value"
|
| 9 |
+
},
|
| 10 |
+
"audio": {
|
| 11 |
+
"sampling_rate": 16000,
|
| 12 |
+
"_type": "Audio"
|
| 13 |
+
}
|
| 14 |
+
},
|
| 15 |
+
"instruction": {
|
| 16 |
+
"audio": {
|
| 17 |
+
"dtype": "null",
|
| 18 |
+
"_type": "Value"
|
| 19 |
+
},
|
| 20 |
+
"text": {
|
| 21 |
+
"dtype": "string",
|
| 22 |
+
"_type": "Value"
|
| 23 |
+
}
|
| 24 |
+
},
|
| 25 |
+
"answer": {
|
| 26 |
+
"audio": {
|
| 27 |
+
"dtype": "null",
|
| 28 |
+
"_type": "Value"
|
| 29 |
+
},
|
| 30 |
+
"text": {
|
| 31 |
+
"dtype": "string",
|
| 32 |
+
"_type": "Value"
|
| 33 |
+
}
|
| 34 |
+
},
|
| 35 |
+
"other_attributes": {
|
| 36 |
+
"audiocap_id": {
|
| 37 |
+
"dtype": "string",
|
| 38 |
+
"_type": "Value"
|
| 39 |
+
},
|
| 40 |
+
"caption": {
|
| 41 |
+
"dtype": "string",
|
| 42 |
+
"_type": "Value"
|
| 43 |
+
},
|
| 44 |
+
"start_time": {
|
| 45 |
+
"dtype": "string",
|
| 46 |
+
"_type": "Value"
|
| 47 |
+
},
|
| 48 |
+
"youtube_id": {
|
| 49 |
+
"dtype": "string",
|
| 50 |
+
"_type": "Value"
|
| 51 |
+
}
|
| 52 |
+
},
|
| 53 |
+
"salmonn_7b": {
|
| 54 |
+
"answer": {
|
| 55 |
+
"dtype": "string",
|
| 56 |
+
"_type": "Value"
|
| 57 |
+
},
|
| 58 |
+
"model_prediction": {
|
| 59 |
+
"dtype": "string",
|
| 60 |
+
"_type": "Value"
|
| 61 |
+
},
|
| 62 |
+
"task_type": {
|
| 63 |
+
"dtype": "string",
|
| 64 |
+
"_type": "Value"
|
| 65 |
+
},
|
| 66 |
+
"text": {
|
| 67 |
+
"dtype": "string",
|
| 68 |
+
"_type": "Value"
|
| 69 |
+
}
|
| 70 |
+
},
|
| 71 |
+
"wavllm_fairseq": {
|
| 72 |
+
"answer": {
|
| 73 |
+
"dtype": "string",
|
| 74 |
+
"_type": "Value"
|
| 75 |
+
},
|
| 76 |
+
"model_prediction": {
|
| 77 |
+
"dtype": "string",
|
| 78 |
+
"_type": "Value"
|
| 79 |
+
},
|
| 80 |
+
"task_type": {
|
| 81 |
+
"dtype": "string",
|
| 82 |
+
"_type": "Value"
|
| 83 |
+
},
|
| 84 |
+
"text": {
|
| 85 |
+
"dtype": "string",
|
| 86 |
+
"_type": "Value"
|
| 87 |
+
}
|
| 88 |
+
},
|
| 89 |
+
"Qwen2-Audio-7B-Instruct": {
|
| 90 |
+
"answer": {
|
| 91 |
+
"dtype": "string",
|
| 92 |
+
"_type": "Value"
|
| 93 |
+
},
|
| 94 |
+
"model_prediction": {
|
| 95 |
+
"dtype": "string",
|
| 96 |
+
"_type": "Value"
|
| 97 |
+
},
|
| 98 |
+
"task_type": {
|
| 99 |
+
"dtype": "string",
|
| 100 |
+
"_type": "Value"
|
| 101 |
+
},
|
| 102 |
+
"text": {
|
| 103 |
+
"dtype": "string",
|
| 104 |
+
"_type": "Value"
|
| 105 |
+
}
|
| 106 |
+
},
|
| 107 |
+
"whisper_large_v3_with_llama_3_8b_instruct": {
|
| 108 |
+
"answer": {
|
| 109 |
+
"dtype": "string",
|
| 110 |
+
"_type": "Value"
|
| 111 |
+
},
|
| 112 |
+
"model_prediction": {
|
| 113 |
+
"dtype": "string",
|
| 114 |
+
"_type": "Value"
|
| 115 |
+
},
|
| 116 |
+
"task_type": {
|
| 117 |
+
"dtype": "string",
|
| 118 |
+
"_type": "Value"
|
| 119 |
+
},
|
| 120 |
+
"text": {
|
| 121 |
+
"dtype": "string",
|
| 122 |
+
"_type": "Value"
|
| 123 |
+
}
|
| 124 |
+
},
|
| 125 |
+
"mowe_audio": {
|
| 126 |
+
"answer": {
|
| 127 |
+
"dtype": "string",
|
| 128 |
+
"_type": "Value"
|
| 129 |
+
},
|
| 130 |
+
"model_prediction": {
|
| 131 |
+
"dtype": "string",
|
| 132 |
+
"_type": "Value"
|
| 133 |
+
},
|
| 134 |
+
"task_type": {
|
| 135 |
+
"dtype": "string",
|
| 136 |
+
"_type": "Value"
|
| 137 |
+
},
|
| 138 |
+
"text": {
|
| 139 |
+
"dtype": "string",
|
| 140 |
+
"_type": "Value"
|
| 141 |
+
}
|
| 142 |
+
},
|
| 143 |
+
"qwen_audio_chat": {
|
| 144 |
+
"answer": {
|
| 145 |
+
"dtype": "string",
|
| 146 |
+
"_type": "Value"
|
| 147 |
+
},
|
| 148 |
+
"model_prediction": {
|
| 149 |
+
"dtype": "string",
|
| 150 |
+
"_type": "Value"
|
| 151 |
+
},
|
| 152 |
+
"task_type": {
|
| 153 |
+
"dtype": "string",
|
| 154 |
+
"_type": "Value"
|
| 155 |
+
},
|
| 156 |
+
"text": {
|
| 157 |
+
"dtype": "string",
|
| 158 |
+
"_type": "Value"
|
| 159 |
+
}
|
| 160 |
+
}
|
| 161 |
+
},
|
| 162 |
+
"homepage": "",
|
| 163 |
+
"license": ""
|
| 164 |
+
}
|
examples/2AQA/AudioCaps-QA-Test/sample_0.wav
ADDED
|
Binary file (320 kB). View file
|
|
|
examples/2AQA/AudioCaps-QA-Test/sample_1.wav
ADDED
|
Binary file (320 kB). View file
|
|
|
examples/2AQA/AudioCaps-QA-Test/sample_2.wav
ADDED
|
Binary file (302 kB). View file
|
|
|
examples/2AQA/AudioCaps-QA-Test/state.json
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_data_files": [
|
| 3 |
+
{
|
| 4 |
+
"filename": "data-00000-of-00001.arrow"
|
| 5 |
+
}
|
| 6 |
+
],
|
| 7 |
+
"_fingerprint": "52bc1dfcaf2a0f4b",
|
| 8 |
+
"_format_columns": [
|
| 9 |
+
"context",
|
| 10 |
+
"instruction",
|
| 11 |
+
"answer",
|
| 12 |
+
"other_attributes",
|
| 13 |
+
"salmonn_7b",
|
| 14 |
+
"wavllm_fairseq",
|
| 15 |
+
"Qwen2-Audio-7B-Instruct",
|
| 16 |
+
"whisper_large_v3_with_llama_3_8b_instruct",
|
| 17 |
+
"mowe_audio",
|
| 18 |
+
"qwen_audio_chat"
|
| 19 |
+
],
|
| 20 |
+
"_format_kwargs": {},
|
| 21 |
+
"_format_type": null,
|
| 22 |
+
"_output_all_columns": false,
|
| 23 |
+
"_split": null
|
| 24 |
+
}
|
examples/2AQA/Clotho-AQA-Test/dataset_info.json
ADDED
|
@@ -0,0 +1,147 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"citation": "",
|
| 3 |
+
"description": "",
|
| 4 |
+
"features": {
|
| 5 |
+
"context": {
|
| 6 |
+
"text": {
|
| 7 |
+
"dtype": "string",
|
| 8 |
+
"_type": "Value"
|
| 9 |
+
},
|
| 10 |
+
"audio": {
|
| 11 |
+
"sampling_rate": 16000,
|
| 12 |
+
"_type": "Audio"
|
| 13 |
+
}
|
| 14 |
+
},
|
| 15 |
+
"instruction": {
|
| 16 |
+
"audio": {
|
| 17 |
+
"dtype": "null",
|
| 18 |
+
"_type": "Value"
|
| 19 |
+
},
|
| 20 |
+
"text": {
|
| 21 |
+
"dtype": "string",
|
| 22 |
+
"_type": "Value"
|
| 23 |
+
}
|
| 24 |
+
},
|
| 25 |
+
"answer": {
|
| 26 |
+
"audio": {
|
| 27 |
+
"dtype": "null",
|
| 28 |
+
"_type": "Value"
|
| 29 |
+
},
|
| 30 |
+
"text": {
|
| 31 |
+
"dtype": "string",
|
| 32 |
+
"_type": "Value"
|
| 33 |
+
}
|
| 34 |
+
},
|
| 35 |
+
"other_attributes": {},
|
| 36 |
+
"salmonn_7b": {
|
| 37 |
+
"answer": {
|
| 38 |
+
"dtype": "string",
|
| 39 |
+
"_type": "Value"
|
| 40 |
+
},
|
| 41 |
+
"model_prediction": {
|
| 42 |
+
"dtype": "string",
|
| 43 |
+
"_type": "Value"
|
| 44 |
+
},
|
| 45 |
+
"task_type": {
|
| 46 |
+
"dtype": "string",
|
| 47 |
+
"_type": "Value"
|
| 48 |
+
},
|
| 49 |
+
"text": {
|
| 50 |
+
"dtype": "string",
|
| 51 |
+
"_type": "Value"
|
| 52 |
+
}
|
| 53 |
+
},
|
| 54 |
+
"wavllm_fairseq": {
|
| 55 |
+
"answer": {
|
| 56 |
+
"dtype": "string",
|
| 57 |
+
"_type": "Value"
|
| 58 |
+
},
|
| 59 |
+
"model_prediction": {
|
| 60 |
+
"dtype": "string",
|
| 61 |
+
"_type": "Value"
|
| 62 |
+
},
|
| 63 |
+
"task_type": {
|
| 64 |
+
"dtype": "string",
|
| 65 |
+
"_type": "Value"
|
| 66 |
+
},
|
| 67 |
+
"text": {
|
| 68 |
+
"dtype": "string",
|
| 69 |
+
"_type": "Value"
|
| 70 |
+
}
|
| 71 |
+
},
|
| 72 |
+
"Qwen2-Audio-7B-Instruct": {
|
| 73 |
+
"answer": {
|
| 74 |
+
"dtype": "string",
|
| 75 |
+
"_type": "Value"
|
| 76 |
+
},
|
| 77 |
+
"model_prediction": {
|
| 78 |
+
"dtype": "string",
|
| 79 |
+
"_type": "Value"
|
| 80 |
+
},
|
| 81 |
+
"task_type": {
|
| 82 |
+
"dtype": "string",
|
| 83 |
+
"_type": "Value"
|
| 84 |
+
},
|
| 85 |
+
"text": {
|
| 86 |
+
"dtype": "string",
|
| 87 |
+
"_type": "Value"
|
| 88 |
+
}
|
| 89 |
+
},
|
| 90 |
+
"whisper_large_v3_with_llama_3_8b_instruct": {
|
| 91 |
+
"answer": {
|
| 92 |
+
"dtype": "string",
|
| 93 |
+
"_type": "Value"
|
| 94 |
+
},
|
| 95 |
+
"model_prediction": {
|
| 96 |
+
"dtype": "string",
|
| 97 |
+
"_type": "Value"
|
| 98 |
+
},
|
| 99 |
+
"task_type": {
|
| 100 |
+
"dtype": "string",
|
| 101 |
+
"_type": "Value"
|
| 102 |
+
},
|
| 103 |
+
"text": {
|
| 104 |
+
"dtype": "string",
|
| 105 |
+
"_type": "Value"
|
| 106 |
+
}
|
| 107 |
+
},
|
| 108 |
+
"mowe_audio": {
|
| 109 |
+
"answer": {
|
| 110 |
+
"dtype": "string",
|
| 111 |
+
"_type": "Value"
|
| 112 |
+
},
|
| 113 |
+
"model_prediction": {
|
| 114 |
+
"dtype": "string",
|
| 115 |
+
"_type": "Value"
|
| 116 |
+
},
|
| 117 |
+
"task_type": {
|
| 118 |
+
"dtype": "string",
|
| 119 |
+
"_type": "Value"
|
| 120 |
+
},
|
| 121 |
+
"text": {
|
| 122 |
+
"dtype": "string",
|
| 123 |
+
"_type": "Value"
|
| 124 |
+
}
|
| 125 |
+
},
|
| 126 |
+
"qwen_audio_chat": {
|
| 127 |
+
"answer": {
|
| 128 |
+
"dtype": "string",
|
| 129 |
+
"_type": "Value"
|
| 130 |
+
},
|
| 131 |
+
"model_prediction": {
|
| 132 |
+
"dtype": "string",
|
| 133 |
+
"_type": "Value"
|
| 134 |
+
},
|
| 135 |
+
"task_type": {
|
| 136 |
+
"dtype": "string",
|
| 137 |
+
"_type": "Value"
|
| 138 |
+
},
|
| 139 |
+
"text": {
|
| 140 |
+
"dtype": "string",
|
| 141 |
+
"_type": "Value"
|
| 142 |
+
}
|
| 143 |
+
}
|
| 144 |
+
},
|
| 145 |
+
"homepage": "",
|
| 146 |
+
"license": ""
|
| 147 |
+
}
|
examples/2AQA/Clotho-AQA-Test/sample_0.wav
ADDED
|
Binary file (615 kB). View file
|
|
|
examples/2AQA/Clotho-AQA-Test/sample_1.wav
ADDED
|
Binary file (576 kB). View file
|
|
|
examples/2AQA/Clotho-AQA-Test/sample_2.wav
ADDED
|
Binary file (835 kB). View file
|
|
|
examples/2AQA/Clotho-AQA-Test/state.json
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_data_files": [
|
| 3 |
+
{
|
| 4 |
+
"filename": "data-00000-of-00001.arrow"
|
| 5 |
+
}
|
| 6 |
+
],
|
| 7 |
+
"_fingerprint": "e2e76326f448d7c4",
|
| 8 |
+
"_format_columns": [
|
| 9 |
+
"context",
|
| 10 |
+
"instruction",
|
| 11 |
+
"answer",
|
| 12 |
+
"other_attributes",
|
| 13 |
+
"salmonn_7b",
|
| 14 |
+
"wavllm_fairseq",
|
| 15 |
+
"Qwen2-Audio-7B-Instruct",
|
| 16 |
+
"whisper_large_v3_with_llama_3_8b_instruct",
|
| 17 |
+
"mowe_audio",
|
| 18 |
+
"qwen_audio_chat"
|
| 19 |
+
],
|
| 20 |
+
"_format_kwargs": {},
|
| 21 |
+
"_format_type": null,
|
| 22 |
+
"_output_all_columns": false,
|
| 23 |
+
"_split": null
|
| 24 |
+
}
|
examples/2AQA/WavCaps-QA-Test/dataset_info.json
ADDED
|
@@ -0,0 +1,160 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"citation": "",
|
| 3 |
+
"description": "",
|
| 4 |
+
"features": {
|
| 5 |
+
"context": {
|
| 6 |
+
"text": {
|
| 7 |
+
"dtype": "string",
|
| 8 |
+
"_type": "Value"
|
| 9 |
+
},
|
| 10 |
+
"audio": {
|
| 11 |
+
"sampling_rate": 16000,
|
| 12 |
+
"_type": "Audio"
|
| 13 |
+
}
|
| 14 |
+
},
|
| 15 |
+
"instruction": {
|
| 16 |
+
"audio": {
|
| 17 |
+
"dtype": "null",
|
| 18 |
+
"_type": "Value"
|
| 19 |
+
},
|
| 20 |
+
"text": {
|
| 21 |
+
"dtype": "string",
|
| 22 |
+
"_type": "Value"
|
| 23 |
+
}
|
| 24 |
+
},
|
| 25 |
+
"answer": {
|
| 26 |
+
"audio": {
|
| 27 |
+
"dtype": "null",
|
| 28 |
+
"_type": "Value"
|
| 29 |
+
},
|
| 30 |
+
"text": {
|
| 31 |
+
"dtype": "string",
|
| 32 |
+
"_type": "Value"
|
| 33 |
+
}
|
| 34 |
+
},
|
| 35 |
+
"other_attributes": {
|
| 36 |
+
"audio_path": {
|
| 37 |
+
"dtype": "string",
|
| 38 |
+
"_type": "Value"
|
| 39 |
+
},
|
| 40 |
+
"caption": {
|
| 41 |
+
"dtype": "string",
|
| 42 |
+
"_type": "Value"
|
| 43 |
+
},
|
| 44 |
+
"duration": {
|
| 45 |
+
"dtype": "string",
|
| 46 |
+
"_type": "Value"
|
| 47 |
+
}
|
| 48 |
+
},
|
| 49 |
+
"salmonn_7b": {
|
| 50 |
+
"answer": {
|
| 51 |
+
"dtype": "string",
|
| 52 |
+
"_type": "Value"
|
| 53 |
+
},
|
| 54 |
+
"model_prediction": {
|
| 55 |
+
"dtype": "string",
|
| 56 |
+
"_type": "Value"
|
| 57 |
+
},
|
| 58 |
+
"task_type": {
|
| 59 |
+
"dtype": "string",
|
| 60 |
+
"_type": "Value"
|
| 61 |
+
},
|
| 62 |
+
"text": {
|
| 63 |
+
"dtype": "string",
|
| 64 |
+
"_type": "Value"
|
| 65 |
+
}
|
| 66 |
+
},
|
| 67 |
+
"wavllm_fairseq": {
|
| 68 |
+
"answer": {
|
| 69 |
+
"dtype": "string",
|
| 70 |
+
"_type": "Value"
|
| 71 |
+
},
|
| 72 |
+
"model_prediction": {
|
| 73 |
+
"dtype": "string",
|
| 74 |
+
"_type": "Value"
|
| 75 |
+
},
|
| 76 |
+
"task_type": {
|
| 77 |
+
"dtype": "string",
|
| 78 |
+
"_type": "Value"
|
| 79 |
+
},
|
| 80 |
+
"text": {
|
| 81 |
+
"dtype": "string",
|
| 82 |
+
"_type": "Value"
|
| 83 |
+
}
|
| 84 |
+
},
|
| 85 |
+
"Qwen2-Audio-7B-Instruct": {
|
| 86 |
+
"answer": {
|
| 87 |
+
"dtype": "string",
|
| 88 |
+
"_type": "Value"
|
| 89 |
+
},
|
| 90 |
+
"model_prediction": {
|
| 91 |
+
"dtype": "string",
|
| 92 |
+
"_type": "Value"
|
| 93 |
+
},
|
| 94 |
+
"task_type": {
|
| 95 |
+
"dtype": "string",
|
| 96 |
+
"_type": "Value"
|
| 97 |
+
},
|
| 98 |
+
"text": {
|
| 99 |
+
"dtype": "string",
|
| 100 |
+
"_type": "Value"
|
| 101 |
+
}
|
| 102 |
+
},
|
| 103 |
+
"whisper_large_v3_with_llama_3_8b_instruct": {
|
| 104 |
+
"answer": {
|
| 105 |
+
"dtype": "string",
|
| 106 |
+
"_type": "Value"
|
| 107 |
+
},
|
| 108 |
+
"model_prediction": {
|
| 109 |
+
"dtype": "string",
|
| 110 |
+
"_type": "Value"
|
| 111 |
+
},
|
| 112 |
+
"task_type": {
|
| 113 |
+
"dtype": "string",
|
| 114 |
+
"_type": "Value"
|
| 115 |
+
},
|
| 116 |
+
"text": {
|
| 117 |
+
"dtype": "string",
|
| 118 |
+
"_type": "Value"
|
| 119 |
+
}
|
| 120 |
+
},
|
| 121 |
+
"mowe_audio": {
|
| 122 |
+
"answer": {
|
| 123 |
+
"dtype": "string",
|
| 124 |
+
"_type": "Value"
|
| 125 |
+
},
|
| 126 |
+
"model_prediction": {
|
| 127 |
+
"dtype": "string",
|
| 128 |
+
"_type": "Value"
|
| 129 |
+
},
|
| 130 |
+
"task_type": {
|
| 131 |
+
"dtype": "string",
|
| 132 |
+
"_type": "Value"
|
| 133 |
+
},
|
| 134 |
+
"text": {
|
| 135 |
+
"dtype": "string",
|
| 136 |
+
"_type": "Value"
|
| 137 |
+
}
|
| 138 |
+
},
|
| 139 |
+
"qwen_audio_chat": {
|
| 140 |
+
"answer": {
|
| 141 |
+
"dtype": "string",
|
| 142 |
+
"_type": "Value"
|
| 143 |
+
},
|
| 144 |
+
"model_prediction": {
|
| 145 |
+
"dtype": "string",
|
| 146 |
+
"_type": "Value"
|
| 147 |
+
},
|
| 148 |
+
"task_type": {
|
| 149 |
+
"dtype": "string",
|
| 150 |
+
"_type": "Value"
|
| 151 |
+
},
|
| 152 |
+
"text": {
|
| 153 |
+
"dtype": "string",
|
| 154 |
+
"_type": "Value"
|
| 155 |
+
}
|
| 156 |
+
}
|
| 157 |
+
},
|
| 158 |
+
"homepage": "",
|
| 159 |
+
"license": ""
|
| 160 |
+
}
|
examples/2AQA/WavCaps-QA-Test/sample_0.wav
ADDED
|
Binary file (320 kB). View file
|
|
|
examples/2AQA/WavCaps-QA-Test/sample_1.wav
ADDED
|
Binary file (251 kB). View file
|
|
|
examples/2AQA/WavCaps-QA-Test/sample_2.wav
ADDED
|
Binary file (256 kB). View file
|
|
|
examples/2AQA/WavCaps-QA-Test/state.json
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_data_files": [
|
| 3 |
+
{
|
| 4 |
+
"filename": "data-00000-of-00001.arrow"
|
| 5 |
+
}
|
| 6 |
+
],
|
| 7 |
+
"_fingerprint": "40995a6cc1fe3dc7",
|
| 8 |
+
"_format_columns": [
|
| 9 |
+
"context",
|
| 10 |
+
"instruction",
|
| 11 |
+
"answer",
|
| 12 |
+
"other_attributes",
|
| 13 |
+
"salmonn_7b",
|
| 14 |
+
"wavllm_fairseq",
|
| 15 |
+
"Qwen2-Audio-7B-Instruct",
|
| 16 |
+
"whisper_large_v3_with_llama_3_8b_instruct",
|
| 17 |
+
"mowe_audio",
|
| 18 |
+
"qwen_audio_chat"
|
| 19 |
+
],
|
| 20 |
+
"_format_kwargs": {},
|
| 21 |
+
"_format_type": null,
|
| 22 |
+
"_output_all_columns": false,
|
| 23 |
+
"_split": null
|
| 24 |
+
}
|
examples/2AR/VoxCeleb-Accent-Test/dataset_info.json
ADDED
|
@@ -0,0 +1,168 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"citation": "",
|
| 3 |
+
"description": "",
|
| 4 |
+
"features": {
|
| 5 |
+
"context": {
|
| 6 |
+
"text": {
|
| 7 |
+
"dtype": "string",
|
| 8 |
+
"_type": "Value"
|
| 9 |
+
},
|
| 10 |
+
"audio": {
|
| 11 |
+
"sampling_rate": 16000,
|
| 12 |
+
"_type": "Audio"
|
| 13 |
+
}
|
| 14 |
+
},
|
| 15 |
+
"instruction": {
|
| 16 |
+
"audio": {
|
| 17 |
+
"dtype": "null",
|
| 18 |
+
"_type": "Value"
|
| 19 |
+
},
|
| 20 |
+
"text": {
|
| 21 |
+
"dtype": "string",
|
| 22 |
+
"_type": "Value"
|
| 23 |
+
}
|
| 24 |
+
},
|
| 25 |
+
"answer": {
|
| 26 |
+
"audio": {
|
| 27 |
+
"dtype": "null",
|
| 28 |
+
"_type": "Value"
|
| 29 |
+
},
|
| 30 |
+
"text": {
|
| 31 |
+
"dtype": "string",
|
| 32 |
+
"_type": "Value"
|
| 33 |
+
}
|
| 34 |
+
},
|
| 35 |
+
"other_attributes": {
|
| 36 |
+
"Gender": {
|
| 37 |
+
"dtype": "string",
|
| 38 |
+
"_type": "Value"
|
| 39 |
+
},
|
| 40 |
+
"Nationality": {
|
| 41 |
+
"dtype": "string",
|
| 42 |
+
"_type": "Value"
|
| 43 |
+
},
|
| 44 |
+
"VGGFace1 ID": {
|
| 45 |
+
"dtype": "string",
|
| 46 |
+
"_type": "Value"
|
| 47 |
+
},
|
| 48 |
+
"VoxCeleb1 ID": {
|
| 49 |
+
"dtype": "string",
|
| 50 |
+
"_type": "Value"
|
| 51 |
+
},
|
| 52 |
+
"index": {
|
| 53 |
+
"dtype": "string",
|
| 54 |
+
"_type": "Value"
|
| 55 |
+
}
|
| 56 |
+
},
|
| 57 |
+
"salmonn_7b": {
|
| 58 |
+
"answer": {
|
| 59 |
+
"dtype": "string",
|
| 60 |
+
"_type": "Value"
|
| 61 |
+
},
|
| 62 |
+
"model_prediction": {
|
| 63 |
+
"dtype": "string",
|
| 64 |
+
"_type": "Value"
|
| 65 |
+
},
|
| 66 |
+
"task_type": {
|
| 67 |
+
"dtype": "string",
|
| 68 |
+
"_type": "Value"
|
| 69 |
+
},
|
| 70 |
+
"text": {
|
| 71 |
+
"dtype": "string",
|
| 72 |
+
"_type": "Value"
|
| 73 |
+
}
|
| 74 |
+
},
|
| 75 |
+
"wavllm_fairseq": {
|
| 76 |
+
"answer": {
|
| 77 |
+
"dtype": "string",
|
| 78 |
+
"_type": "Value"
|
| 79 |
+
},
|
| 80 |
+
"model_prediction": {
|
| 81 |
+
"dtype": "string",
|
| 82 |
+
"_type": "Value"
|
| 83 |
+
},
|
| 84 |
+
"task_type": {
|
| 85 |
+
"dtype": "string",
|
| 86 |
+
"_type": "Value"
|
| 87 |
+
},
|
| 88 |
+
"text": {
|
| 89 |
+
"dtype": "string",
|
| 90 |
+
"_type": "Value"
|
| 91 |
+
}
|
| 92 |
+
},
|
| 93 |
+
"Qwen2-Audio-7B-Instruct": {
|
| 94 |
+
"answer": {
|
| 95 |
+
"dtype": "string",
|
| 96 |
+
"_type": "Value"
|
| 97 |
+
},
|
| 98 |
+
"model_prediction": {
|
| 99 |
+
"dtype": "string",
|
| 100 |
+
"_type": "Value"
|
| 101 |
+
},
|
| 102 |
+
"task_type": {
|
| 103 |
+
"dtype": "string",
|
| 104 |
+
"_type": "Value"
|
| 105 |
+
},
|
| 106 |
+
"text": {
|
| 107 |
+
"dtype": "string",
|
| 108 |
+
"_type": "Value"
|
| 109 |
+
}
|
| 110 |
+
},
|
| 111 |
+
"whisper_large_v3_with_llama_3_8b_instruct": {
|
| 112 |
+
"answer": {
|
| 113 |
+
"dtype": "string",
|
| 114 |
+
"_type": "Value"
|
| 115 |
+
},
|
| 116 |
+
"model_prediction": {
|
| 117 |
+
"dtype": "string",
|
| 118 |
+
"_type": "Value"
|
| 119 |
+
},
|
| 120 |
+
"task_type": {
|
| 121 |
+
"dtype": "string",
|
| 122 |
+
"_type": "Value"
|
| 123 |
+
},
|
| 124 |
+
"text": {
|
| 125 |
+
"dtype": "string",
|
| 126 |
+
"_type": "Value"
|
| 127 |
+
}
|
| 128 |
+
},
|
| 129 |
+
"mowe_audio": {
|
| 130 |
+
"answer": {
|
| 131 |
+
"dtype": "string",
|
| 132 |
+
"_type": "Value"
|
| 133 |
+
},
|
| 134 |
+
"model_prediction": {
|
| 135 |
+
"dtype": "string",
|
| 136 |
+
"_type": "Value"
|
| 137 |
+
},
|
| 138 |
+
"task_type": {
|
| 139 |
+
"dtype": "string",
|
| 140 |
+
"_type": "Value"
|
| 141 |
+
},
|
| 142 |
+
"text": {
|
| 143 |
+
"dtype": "string",
|
| 144 |
+
"_type": "Value"
|
| 145 |
+
}
|
| 146 |
+
},
|
| 147 |
+
"qwen_audio_chat": {
|
| 148 |
+
"answer": {
|
| 149 |
+
"dtype": "string",
|
| 150 |
+
"_type": "Value"
|
| 151 |
+
},
|
| 152 |
+
"model_prediction": {
|
| 153 |
+
"dtype": "string",
|
| 154 |
+
"_type": "Value"
|
| 155 |
+
},
|
| 156 |
+
"task_type": {
|
| 157 |
+
"dtype": "string",
|
| 158 |
+
"_type": "Value"
|
| 159 |
+
},
|
| 160 |
+
"text": {
|
| 161 |
+
"dtype": "string",
|
| 162 |
+
"_type": "Value"
|
| 163 |
+
}
|
| 164 |
+
}
|
| 165 |
+
},
|
| 166 |
+
"homepage": "",
|
| 167 |
+
"license": ""
|
| 168 |
+
}
|
examples/2AR/VoxCeleb-Accent-Test/sample_0.wav
ADDED
|
Binary file (268 kB). View file
|
|
|
examples/2AR/VoxCeleb-Accent-Test/sample_1.wav
ADDED
|
Binary file (152 kB). View file
|
|
|
examples/2AR/VoxCeleb-Accent-Test/sample_2.wav
ADDED
|
Binary file (301 kB). View file
|
|
|
examples/2AR/VoxCeleb-Accent-Test/state.json
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_data_files": [
|
| 3 |
+
{
|
| 4 |
+
"filename": "data-00000-of-00001.arrow"
|
| 5 |
+
}
|
| 6 |
+
],
|
| 7 |
+
"_fingerprint": "fa91a59f90c22c3c",
|
| 8 |
+
"_format_columns": [
|
| 9 |
+
"context",
|
| 10 |
+
"instruction",
|
| 11 |
+
"answer",
|
| 12 |
+
"other_attributes",
|
| 13 |
+
"salmonn_7b",
|
| 14 |
+
"wavllm_fairseq",
|
| 15 |
+
"Qwen2-Audio-7B-Instruct",
|
| 16 |
+
"whisper_large_v3_with_llama_3_8b_instruct",
|
| 17 |
+
"mowe_audio",
|
| 18 |
+
"qwen_audio_chat"
|
| 19 |
+
],
|
| 20 |
+
"_format_kwargs": {},
|
| 21 |
+
"_format_type": null,
|
| 22 |
+
"_output_all_columns": false,
|
| 23 |
+
"_split": null
|
| 24 |
+
}
|
examples/2ASR/Common-Voice-15-En-Test/dataset_info.json
ADDED
|
@@ -0,0 +1,188 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"citation": "",
|
| 3 |
+
"description": "",
|
| 4 |
+
"features": {
|
| 5 |
+
"context": {
|
| 6 |
+
"text": {
|
| 7 |
+
"dtype": "string",
|
| 8 |
+
"_type": "Value"
|
| 9 |
+
},
|
| 10 |
+
"audio": {
|
| 11 |
+
"sampling_rate": 16000,
|
| 12 |
+
"_type": "Audio"
|
| 13 |
+
}
|
| 14 |
+
},
|
| 15 |
+
"instruction": {
|
| 16 |
+
"audio": {
|
| 17 |
+
"dtype": "null",
|
| 18 |
+
"_type": "Value"
|
| 19 |
+
},
|
| 20 |
+
"text": {
|
| 21 |
+
"dtype": "string",
|
| 22 |
+
"_type": "Value"
|
| 23 |
+
}
|
| 24 |
+
},
|
| 25 |
+
"answer": {
|
| 26 |
+
"audio": {
|
| 27 |
+
"dtype": "null",
|
| 28 |
+
"_type": "Value"
|
| 29 |
+
},
|
| 30 |
+
"text": {
|
| 31 |
+
"dtype": "string",
|
| 32 |
+
"_type": "Value"
|
| 33 |
+
}
|
| 34 |
+
},
|
| 35 |
+
"other_attributes": {
|
| 36 |
+
"accents": {
|
| 37 |
+
"dtype": "null",
|
| 38 |
+
"_type": "Value"
|
| 39 |
+
},
|
| 40 |
+
"age": {
|
| 41 |
+
"dtype": "null",
|
| 42 |
+
"_type": "Value"
|
| 43 |
+
},
|
| 44 |
+
"client_id": {
|
| 45 |
+
"dtype": "string",
|
| 46 |
+
"_type": "Value"
|
| 47 |
+
},
|
| 48 |
+
"down_votes": {
|
| 49 |
+
"dtype": "int64",
|
| 50 |
+
"_type": "Value"
|
| 51 |
+
},
|
| 52 |
+
"gender": {
|
| 53 |
+
"dtype": "null",
|
| 54 |
+
"_type": "Value"
|
| 55 |
+
},
|
| 56 |
+
"language": {
|
| 57 |
+
"dtype": "string",
|
| 58 |
+
"_type": "Value"
|
| 59 |
+
},
|
| 60 |
+
"locale": {
|
| 61 |
+
"dtype": "string",
|
| 62 |
+
"_type": "Value"
|
| 63 |
+
},
|
| 64 |
+
"segment": {
|
| 65 |
+
"dtype": "null",
|
| 66 |
+
"_type": "Value"
|
| 67 |
+
},
|
| 68 |
+
"up_votes": {
|
| 69 |
+
"dtype": "int64",
|
| 70 |
+
"_type": "Value"
|
| 71 |
+
},
|
| 72 |
+
"variant": {
|
| 73 |
+
"dtype": "null",
|
| 74 |
+
"_type": "Value"
|
| 75 |
+
}
|
| 76 |
+
},
|
| 77 |
+
"salmonn_7b": {
|
| 78 |
+
"answer": {
|
| 79 |
+
"dtype": "string",
|
| 80 |
+
"_type": "Value"
|
| 81 |
+
},
|
| 82 |
+
"model_prediction": {
|
| 83 |
+
"dtype": "string",
|
| 84 |
+
"_type": "Value"
|
| 85 |
+
},
|
| 86 |
+
"task_type": {
|
| 87 |
+
"dtype": "string",
|
| 88 |
+
"_type": "Value"
|
| 89 |
+
},
|
| 90 |
+
"text": {
|
| 91 |
+
"dtype": "string",
|
| 92 |
+
"_type": "Value"
|
| 93 |
+
}
|
| 94 |
+
},
|
| 95 |
+
"wavllm_fairseq": {
|
| 96 |
+
"answer": {
|
| 97 |
+
"dtype": "string",
|
| 98 |
+
"_type": "Value"
|
| 99 |
+
},
|
| 100 |
+
"model_prediction": {
|
| 101 |
+
"dtype": "string",
|
| 102 |
+
"_type": "Value"
|
| 103 |
+
},
|
| 104 |
+
"task_type": {
|
| 105 |
+
"dtype": "string",
|
| 106 |
+
"_type": "Value"
|
| 107 |
+
},
|
| 108 |
+
"text": {
|
| 109 |
+
"dtype": "string",
|
| 110 |
+
"_type": "Value"
|
| 111 |
+
}
|
| 112 |
+
},
|
| 113 |
+
"Qwen2-Audio-7B-Instruct": {
|
| 114 |
+
"answer": {
|
| 115 |
+
"dtype": "string",
|
| 116 |
+
"_type": "Value"
|
| 117 |
+
},
|
| 118 |
+
"model_prediction": {
|
| 119 |
+
"dtype": "string",
|
| 120 |
+
"_type": "Value"
|
| 121 |
+
},
|
| 122 |
+
"task_type": {
|
| 123 |
+
"dtype": "string",
|
| 124 |
+
"_type": "Value"
|
| 125 |
+
},
|
| 126 |
+
"text": {
|
| 127 |
+
"dtype": "string",
|
| 128 |
+
"_type": "Value"
|
| 129 |
+
}
|
| 130 |
+
},
|
| 131 |
+
"whisper_large_v3_with_llama_3_8b_instruct": {
|
| 132 |
+
"answer": {
|
| 133 |
+
"dtype": "string",
|
| 134 |
+
"_type": "Value"
|
| 135 |
+
},
|
| 136 |
+
"model_prediction": {
|
| 137 |
+
"dtype": "string",
|
| 138 |
+
"_type": "Value"
|
| 139 |
+
},
|
| 140 |
+
"task_type": {
|
| 141 |
+
"dtype": "string",
|
| 142 |
+
"_type": "Value"
|
| 143 |
+
},
|
| 144 |
+
"text": {
|
| 145 |
+
"dtype": "string",
|
| 146 |
+
"_type": "Value"
|
| 147 |
+
}
|
| 148 |
+
},
|
| 149 |
+
"mowe_audio": {
|
| 150 |
+
"answer": {
|
| 151 |
+
"dtype": "string",
|
| 152 |
+
"_type": "Value"
|
| 153 |
+
},
|
| 154 |
+
"model_prediction": {
|
| 155 |
+
"dtype": "string",
|
| 156 |
+
"_type": "Value"
|
| 157 |
+
},
|
| 158 |
+
"task_type": {
|
| 159 |
+
"dtype": "string",
|
| 160 |
+
"_type": "Value"
|
| 161 |
+
},
|
| 162 |
+
"text": {
|
| 163 |
+
"dtype": "string",
|
| 164 |
+
"_type": "Value"
|
| 165 |
+
}
|
| 166 |
+
},
|
| 167 |
+
"qwen_audio_chat": {
|
| 168 |
+
"answer": {
|
| 169 |
+
"dtype": "string",
|
| 170 |
+
"_type": "Value"
|
| 171 |
+
},
|
| 172 |
+
"model_prediction": {
|
| 173 |
+
"dtype": "string",
|
| 174 |
+
"_type": "Value"
|
| 175 |
+
},
|
| 176 |
+
"task_type": {
|
| 177 |
+
"dtype": "string",
|
| 178 |
+
"_type": "Value"
|
| 179 |
+
},
|
| 180 |
+
"text": {
|
| 181 |
+
"dtype": "string",
|
| 182 |
+
"_type": "Value"
|
| 183 |
+
}
|
| 184 |
+
}
|
| 185 |
+
},
|
| 186 |
+
"homepage": "",
|
| 187 |
+
"license": ""
|
| 188 |
+
}
|
examples/2ASR/Common-Voice-15-En-Test/sample_0.wav
ADDED
|
Binary file (158 kB). View file
|
|
|
examples/2ASR/Common-Voice-15-En-Test/sample_1.wav
ADDED
|
Binary file (172 kB). View file
|
|
|
examples/2ASR/Common-Voice-15-En-Test/sample_2.wav
ADDED
|
Binary file (148 kB). View file
|
|
|
examples/2ASR/Common-Voice-15-En-Test/state.json
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_data_files": [
|
| 3 |
+
{
|
| 4 |
+
"filename": "data-00000-of-00001.arrow"
|
| 5 |
+
}
|
| 6 |
+
],
|
| 7 |
+
"_fingerprint": "468db91ad949e4d4",
|
| 8 |
+
"_format_columns": [
|
| 9 |
+
"context",
|
| 10 |
+
"instruction",
|
| 11 |
+
"answer",
|
| 12 |
+
"other_attributes",
|
| 13 |
+
"salmonn_7b",
|
| 14 |
+
"wavllm_fairseq",
|
| 15 |
+
"Qwen2-Audio-7B-Instruct",
|
| 16 |
+
"whisper_large_v3_with_llama_3_8b_instruct",
|
| 17 |
+
"mowe_audio",
|
| 18 |
+
"qwen_audio_chat"
|
| 19 |
+
],
|
| 20 |
+
"_format_kwargs": {},
|
| 21 |
+
"_format_type": null,
|
| 22 |
+
"_output_all_columns": false,
|
| 23 |
+
"_split": null
|
| 24 |
+
}
|
examples/2ASR/Earnings21-Test/dataset_info.json
ADDED
|
@@ -0,0 +1,152 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"citation": "",
|
| 3 |
+
"description": "",
|
| 4 |
+
"features": {
|
| 5 |
+
"context": {
|
| 6 |
+
"text": {
|
| 7 |
+
"dtype": "string",
|
| 8 |
+
"_type": "Value"
|
| 9 |
+
},
|
| 10 |
+
"audio": {
|
| 11 |
+
"sampling_rate": 16000,
|
| 12 |
+
"_type": "Audio"
|
| 13 |
+
}
|
| 14 |
+
},
|
| 15 |
+
"instruction": {
|
| 16 |
+
"audio": {
|
| 17 |
+
"dtype": "null",
|
| 18 |
+
"_type": "Value"
|
| 19 |
+
},
|
| 20 |
+
"text": {
|
| 21 |
+
"dtype": "string",
|
| 22 |
+
"_type": "Value"
|
| 23 |
+
}
|
| 24 |
+
},
|
| 25 |
+
"answer": {
|
| 26 |
+
"audio": {
|
| 27 |
+
"dtype": "null",
|
| 28 |
+
"_type": "Value"
|
| 29 |
+
},
|
| 30 |
+
"text": {
|
| 31 |
+
"dtype": "string",
|
| 32 |
+
"_type": "Value"
|
| 33 |
+
}
|
| 34 |
+
},
|
| 35 |
+
"other_attributes": {
|
| 36 |
+
"id": {
|
| 37 |
+
"dtype": "string",
|
| 38 |
+
"_type": "Value"
|
| 39 |
+
}
|
| 40 |
+
},
|
| 41 |
+
"salmonn_7b": {
|
| 42 |
+
"answer": {
|
| 43 |
+
"dtype": "string",
|
| 44 |
+
"_type": "Value"
|
| 45 |
+
},
|
| 46 |
+
"model_prediction": {
|
| 47 |
+
"dtype": "string",
|
| 48 |
+
"_type": "Value"
|
| 49 |
+
},
|
| 50 |
+
"task_type": {
|
| 51 |
+
"dtype": "string",
|
| 52 |
+
"_type": "Value"
|
| 53 |
+
},
|
| 54 |
+
"text": {
|
| 55 |
+
"dtype": "string",
|
| 56 |
+
"_type": "Value"
|
| 57 |
+
}
|
| 58 |
+
},
|
| 59 |
+
"wavllm_fairseq": {
|
| 60 |
+
"answer": {
|
| 61 |
+
"dtype": "string",
|
| 62 |
+
"_type": "Value"
|
| 63 |
+
},
|
| 64 |
+
"model_prediction": {
|
| 65 |
+
"dtype": "string",
|
| 66 |
+
"_type": "Value"
|
| 67 |
+
},
|
| 68 |
+
"task_type": {
|
| 69 |
+
"dtype": "string",
|
| 70 |
+
"_type": "Value"
|
| 71 |
+
},
|
| 72 |
+
"text": {
|
| 73 |
+
"dtype": "string",
|
| 74 |
+
"_type": "Value"
|
| 75 |
+
}
|
| 76 |
+
},
|
| 77 |
+
"Qwen2-Audio-7B-Instruct": {
|
| 78 |
+
"answer": {
|
| 79 |
+
"dtype": "string",
|
| 80 |
+
"_type": "Value"
|
| 81 |
+
},
|
| 82 |
+
"model_prediction": {
|
| 83 |
+
"dtype": "string",
|
| 84 |
+
"_type": "Value"
|
| 85 |
+
},
|
| 86 |
+
"task_type": {
|
| 87 |
+
"dtype": "string",
|
| 88 |
+
"_type": "Value"
|
| 89 |
+
},
|
| 90 |
+
"text": {
|
| 91 |
+
"dtype": "string",
|
| 92 |
+
"_type": "Value"
|
| 93 |
+
}
|
| 94 |
+
},
|
| 95 |
+
"whisper_large_v3_with_llama_3_8b_instruct": {
|
| 96 |
+
"answer": {
|
| 97 |
+
"dtype": "string",
|
| 98 |
+
"_type": "Value"
|
| 99 |
+
},
|
| 100 |
+
"model_prediction": {
|
| 101 |
+
"dtype": "string",
|
| 102 |
+
"_type": "Value"
|
| 103 |
+
},
|
| 104 |
+
"task_type": {
|
| 105 |
+
"dtype": "string",
|
| 106 |
+
"_type": "Value"
|
| 107 |
+
},
|
| 108 |
+
"text": {
|
| 109 |
+
"dtype": "string",
|
| 110 |
+
"_type": "Value"
|
| 111 |
+
}
|
| 112 |
+
},
|
| 113 |
+
"mowe_audio": {
|
| 114 |
+
"answer": {
|
| 115 |
+
"dtype": "string",
|
| 116 |
+
"_type": "Value"
|
| 117 |
+
},
|
| 118 |
+
"model_prediction": {
|
| 119 |
+
"dtype": "string",
|
| 120 |
+
"_type": "Value"
|
| 121 |
+
},
|
| 122 |
+
"task_type": {
|
| 123 |
+
"dtype": "string",
|
| 124 |
+
"_type": "Value"
|
| 125 |
+
},
|
| 126 |
+
"text": {
|
| 127 |
+
"dtype": "string",
|
| 128 |
+
"_type": "Value"
|
| 129 |
+
}
|
| 130 |
+
},
|
| 131 |
+
"qwen_audio_chat": {
|
| 132 |
+
"answer": {
|
| 133 |
+
"dtype": "string",
|
| 134 |
+
"_type": "Value"
|
| 135 |
+
},
|
| 136 |
+
"model_prediction": {
|
| 137 |
+
"dtype": "string",
|
| 138 |
+
"_type": "Value"
|
| 139 |
+
},
|
| 140 |
+
"task_type": {
|
| 141 |
+
"dtype": "string",
|
| 142 |
+
"_type": "Value"
|
| 143 |
+
},
|
| 144 |
+
"text": {
|
| 145 |
+
"dtype": "string",
|
| 146 |
+
"_type": "Value"
|
| 147 |
+
}
|
| 148 |
+
}
|
| 149 |
+
},
|
| 150 |
+
"homepage": "",
|
| 151 |
+
"license": ""
|
| 152 |
+
}
|
examples/2ASR/Earnings21-Test/state.json
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_data_files": [
|
| 3 |
+
{
|
| 4 |
+
"filename": "data-00000-of-00001.arrow"
|
| 5 |
+
}
|
| 6 |
+
],
|
| 7 |
+
"_fingerprint": "8cc0ad99446f1aba",
|
| 8 |
+
"_format_columns": [
|
| 9 |
+
"context",
|
| 10 |
+
"instruction",
|
| 11 |
+
"answer",
|
| 12 |
+
"other_attributes",
|
| 13 |
+
"salmonn_7b",
|
| 14 |
+
"wavllm_fairseq",
|
| 15 |
+
"Qwen2-Audio-7B-Instruct",
|
| 16 |
+
"whisper_large_v3_with_llama_3_8b_instruct",
|
| 17 |
+
"mowe_audio",
|
| 18 |
+
"qwen_audio_chat"
|
| 19 |
+
],
|
| 20 |
+
"_format_kwargs": {},
|
| 21 |
+
"_format_type": null,
|
| 22 |
+
"_output_all_columns": false,
|
| 23 |
+
"_split": null
|
| 24 |
+
}
|
examples/2ASR/Earnings22-Test/dataset_info.json
ADDED
|
@@ -0,0 +1,152 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"citation": "",
|
| 3 |
+
"description": "",
|
| 4 |
+
"features": {
|
| 5 |
+
"context": {
|
| 6 |
+
"text": {
|
| 7 |
+
"dtype": "string",
|
| 8 |
+
"_type": "Value"
|
| 9 |
+
},
|
| 10 |
+
"audio": {
|
| 11 |
+
"sampling_rate": 16000,
|
| 12 |
+
"_type": "Audio"
|
| 13 |
+
}
|
| 14 |
+
},
|
| 15 |
+
"instruction": {
|
| 16 |
+
"audio": {
|
| 17 |
+
"dtype": "null",
|
| 18 |
+
"_type": "Value"
|
| 19 |
+
},
|
| 20 |
+
"text": {
|
| 21 |
+
"dtype": "string",
|
| 22 |
+
"_type": "Value"
|
| 23 |
+
}
|
| 24 |
+
},
|
| 25 |
+
"answer": {
|
| 26 |
+
"audio": {
|
| 27 |
+
"dtype": "null",
|
| 28 |
+
"_type": "Value"
|
| 29 |
+
},
|
| 30 |
+
"text": {
|
| 31 |
+
"dtype": "string",
|
| 32 |
+
"_type": "Value"
|
| 33 |
+
}
|
| 34 |
+
},
|
| 35 |
+
"other_attributes": {
|
| 36 |
+
"id": {
|
| 37 |
+
"dtype": "string",
|
| 38 |
+
"_type": "Value"
|
| 39 |
+
}
|
| 40 |
+
},
|
| 41 |
+
"salmonn_7b": {
|
| 42 |
+
"answer": {
|
| 43 |
+
"dtype": "string",
|
| 44 |
+
"_type": "Value"
|
| 45 |
+
},
|
| 46 |
+
"model_prediction": {
|
| 47 |
+
"dtype": "string",
|
| 48 |
+
"_type": "Value"
|
| 49 |
+
},
|
| 50 |
+
"task_type": {
|
| 51 |
+
"dtype": "string",
|
| 52 |
+
"_type": "Value"
|
| 53 |
+
},
|
| 54 |
+
"text": {
|
| 55 |
+
"dtype": "string",
|
| 56 |
+
"_type": "Value"
|
| 57 |
+
}
|
| 58 |
+
},
|
| 59 |
+
"wavllm_fairseq": {
|
| 60 |
+
"answer": {
|
| 61 |
+
"dtype": "string",
|
| 62 |
+
"_type": "Value"
|
| 63 |
+
},
|
| 64 |
+
"model_prediction": {
|
| 65 |
+
"dtype": "string",
|
| 66 |
+
"_type": "Value"
|
| 67 |
+
},
|
| 68 |
+
"task_type": {
|
| 69 |
+
"dtype": "string",
|
| 70 |
+
"_type": "Value"
|
| 71 |
+
},
|
| 72 |
+
"text": {
|
| 73 |
+
"dtype": "string",
|
| 74 |
+
"_type": "Value"
|
| 75 |
+
}
|
| 76 |
+
},
|
| 77 |
+
"Qwen2-Audio-7B-Instruct": {
|
| 78 |
+
"answer": {
|
| 79 |
+
"dtype": "string",
|
| 80 |
+
"_type": "Value"
|
| 81 |
+
},
|
| 82 |
+
"model_prediction": {
|
| 83 |
+
"dtype": "string",
|
| 84 |
+
"_type": "Value"
|
| 85 |
+
},
|
| 86 |
+
"task_type": {
|
| 87 |
+
"dtype": "string",
|
| 88 |
+
"_type": "Value"
|
| 89 |
+
},
|
| 90 |
+
"text": {
|
| 91 |
+
"dtype": "string",
|
| 92 |
+
"_type": "Value"
|
| 93 |
+
}
|
| 94 |
+
},
|
| 95 |
+
"whisper_large_v3_with_llama_3_8b_instruct": {
|
| 96 |
+
"answer": {
|
| 97 |
+
"dtype": "string",
|
| 98 |
+
"_type": "Value"
|
| 99 |
+
},
|
| 100 |
+
"model_prediction": {
|
| 101 |
+
"dtype": "string",
|
| 102 |
+
"_type": "Value"
|
| 103 |
+
},
|
| 104 |
+
"task_type": {
|
| 105 |
+
"dtype": "string",
|
| 106 |
+
"_type": "Value"
|
| 107 |
+
},
|
| 108 |
+
"text": {
|
| 109 |
+
"dtype": "string",
|
| 110 |
+
"_type": "Value"
|
| 111 |
+
}
|
| 112 |
+
},
|
| 113 |
+
"mowe_audio": {
|
| 114 |
+
"answer": {
|
| 115 |
+
"dtype": "string",
|
| 116 |
+
"_type": "Value"
|
| 117 |
+
},
|
| 118 |
+
"model_prediction": {
|
| 119 |
+
"dtype": "string",
|
| 120 |
+
"_type": "Value"
|
| 121 |
+
},
|
| 122 |
+
"task_type": {
|
| 123 |
+
"dtype": "string",
|
| 124 |
+
"_type": "Value"
|
| 125 |
+
},
|
| 126 |
+
"text": {
|
| 127 |
+
"dtype": "string",
|
| 128 |
+
"_type": "Value"
|
| 129 |
+
}
|
| 130 |
+
},
|
| 131 |
+
"qwen_audio_chat": {
|
| 132 |
+
"answer": {
|
| 133 |
+
"dtype": "string",
|
| 134 |
+
"_type": "Value"
|
| 135 |
+
},
|
| 136 |
+
"model_prediction": {
|
| 137 |
+
"dtype": "string",
|
| 138 |
+
"_type": "Value"
|
| 139 |
+
},
|
| 140 |
+
"task_type": {
|
| 141 |
+
"dtype": "string",
|
| 142 |
+
"_type": "Value"
|
| 143 |
+
},
|
| 144 |
+
"text": {
|
| 145 |
+
"dtype": "string",
|
| 146 |
+
"_type": "Value"
|
| 147 |
+
}
|
| 148 |
+
}
|
| 149 |
+
},
|
| 150 |
+
"homepage": "",
|
| 151 |
+
"license": ""
|
| 152 |
+
}
|
examples/2ASR/Earnings22-Test/state.json
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_data_files": [
|
| 3 |
+
{
|
| 4 |
+
"filename": "data-00000-of-00001.arrow"
|
| 5 |
+
}
|
| 6 |
+
],
|
| 7 |
+
"_fingerprint": "331c061bce6e651c",
|
| 8 |
+
"_format_columns": [
|
| 9 |
+
"context",
|
| 10 |
+
"instruction",
|
| 11 |
+
"answer",
|
| 12 |
+
"other_attributes",
|
| 13 |
+
"salmonn_7b",
|
| 14 |
+
"wavllm_fairseq",
|
| 15 |
+
"Qwen2-Audio-7B-Instruct",
|
| 16 |
+
"whisper_large_v3_with_llama_3_8b_instruct",
|
| 17 |
+
"mowe_audio",
|
| 18 |
+
"qwen_audio_chat"
|
| 19 |
+
],
|
| 20 |
+
"_format_kwargs": {},
|
| 21 |
+
"_format_type": null,
|
| 22 |
+
"_output_all_columns": false,
|
| 23 |
+
"_split": null
|
| 24 |
+
}
|
examples/2ASR/GigaSpeech-Test2/dataset_info.json
ADDED
|
@@ -0,0 +1,180 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"citation": "",
|
| 3 |
+
"description": "",
|
| 4 |
+
"features": {
|
| 5 |
+
"context": {
|
| 6 |
+
"text": {
|
| 7 |
+
"dtype": "string",
|
| 8 |
+
"_type": "Value"
|
| 9 |
+
},
|
| 10 |
+
"audio": {
|
| 11 |
+
"sampling_rate": 16000,
|
| 12 |
+
"_type": "Audio"
|
| 13 |
+
}
|
| 14 |
+
},
|
| 15 |
+
"instruction": {
|
| 16 |
+
"audio": {
|
| 17 |
+
"dtype": "null",
|
| 18 |
+
"_type": "Value"
|
| 19 |
+
},
|
| 20 |
+
"text": {
|
| 21 |
+
"dtype": "string",
|
| 22 |
+
"_type": "Value"
|
| 23 |
+
}
|
| 24 |
+
},
|
| 25 |
+
"answer": {
|
| 26 |
+
"audio": {
|
| 27 |
+
"dtype": "null",
|
| 28 |
+
"_type": "Value"
|
| 29 |
+
},
|
| 30 |
+
"text": {
|
| 31 |
+
"dtype": "string",
|
| 32 |
+
"_type": "Value"
|
| 33 |
+
}
|
| 34 |
+
},
|
| 35 |
+
"other_attributes": {
|
| 36 |
+
"audio_id": {
|
| 37 |
+
"dtype": "string",
|
| 38 |
+
"_type": "Value"
|
| 39 |
+
},
|
| 40 |
+
"begin_time": {
|
| 41 |
+
"dtype": "float64",
|
| 42 |
+
"_type": "Value"
|
| 43 |
+
},
|
| 44 |
+
"category": {
|
| 45 |
+
"dtype": "int64",
|
| 46 |
+
"_type": "Value"
|
| 47 |
+
},
|
| 48 |
+
"end_time": {
|
| 49 |
+
"dtype": "float64",
|
| 50 |
+
"_type": "Value"
|
| 51 |
+
},
|
| 52 |
+
"segment_id": {
|
| 53 |
+
"dtype": "string",
|
| 54 |
+
"_type": "Value"
|
| 55 |
+
},
|
| 56 |
+
"source": {
|
| 57 |
+
"dtype": "int64",
|
| 58 |
+
"_type": "Value"
|
| 59 |
+
},
|
| 60 |
+
"speaker": {
|
| 61 |
+
"dtype": "string",
|
| 62 |
+
"_type": "Value"
|
| 63 |
+
},
|
| 64 |
+
"url": {
|
| 65 |
+
"dtype": "string",
|
| 66 |
+
"_type": "Value"
|
| 67 |
+
}
|
| 68 |
+
},
|
| 69 |
+
"salmonn_7b": {
|
| 70 |
+
"answer": {
|
| 71 |
+
"dtype": "string",
|
| 72 |
+
"_type": "Value"
|
| 73 |
+
},
|
| 74 |
+
"model_prediction": {
|
| 75 |
+
"dtype": "string",
|
| 76 |
+
"_type": "Value"
|
| 77 |
+
},
|
| 78 |
+
"task_type": {
|
| 79 |
+
"dtype": "string",
|
| 80 |
+
"_type": "Value"
|
| 81 |
+
},
|
| 82 |
+
"text": {
|
| 83 |
+
"dtype": "string",
|
| 84 |
+
"_type": "Value"
|
| 85 |
+
}
|
| 86 |
+
},
|
| 87 |
+
"wavllm_fairseq": {
|
| 88 |
+
"answer": {
|
| 89 |
+
"dtype": "string",
|
| 90 |
+
"_type": "Value"
|
| 91 |
+
},
|
| 92 |
+
"model_prediction": {
|
| 93 |
+
"dtype": "string",
|
| 94 |
+
"_type": "Value"
|
| 95 |
+
},
|
| 96 |
+
"task_type": {
|
| 97 |
+
"dtype": "string",
|
| 98 |
+
"_type": "Value"
|
| 99 |
+
},
|
| 100 |
+
"text": {
|
| 101 |
+
"dtype": "string",
|
| 102 |
+
"_type": "Value"
|
| 103 |
+
}
|
| 104 |
+
},
|
| 105 |
+
"Qwen2-Audio-7B-Instruct": {
|
| 106 |
+
"answer": {
|
| 107 |
+
"dtype": "string",
|
| 108 |
+
"_type": "Value"
|
| 109 |
+
},
|
| 110 |
+
"model_prediction": {
|
| 111 |
+
"dtype": "string",
|
| 112 |
+
"_type": "Value"
|
| 113 |
+
},
|
| 114 |
+
"task_type": {
|
| 115 |
+
"dtype": "string",
|
| 116 |
+
"_type": "Value"
|
| 117 |
+
},
|
| 118 |
+
"text": {
|
| 119 |
+
"dtype": "string",
|
| 120 |
+
"_type": "Value"
|
| 121 |
+
}
|
| 122 |
+
},
|
| 123 |
+
"whisper_large_v3_with_llama_3_8b_instruct": {
|
| 124 |
+
"answer": {
|
| 125 |
+
"dtype": "string",
|
| 126 |
+
"_type": "Value"
|
| 127 |
+
},
|
| 128 |
+
"model_prediction": {
|
| 129 |
+
"dtype": "string",
|
| 130 |
+
"_type": "Value"
|
| 131 |
+
},
|
| 132 |
+
"task_type": {
|
| 133 |
+
"dtype": "string",
|
| 134 |
+
"_type": "Value"
|
| 135 |
+
},
|
| 136 |
+
"text": {
|
| 137 |
+
"dtype": "string",
|
| 138 |
+
"_type": "Value"
|
| 139 |
+
}
|
| 140 |
+
},
|
| 141 |
+
"mowe_audio": {
|
| 142 |
+
"answer": {
|
| 143 |
+
"dtype": "string",
|
| 144 |
+
"_type": "Value"
|
| 145 |
+
},
|
| 146 |
+
"model_prediction": {
|
| 147 |
+
"dtype": "string",
|
| 148 |
+
"_type": "Value"
|
| 149 |
+
},
|
| 150 |
+
"task_type": {
|
| 151 |
+
"dtype": "string",
|
| 152 |
+
"_type": "Value"
|
| 153 |
+
},
|
| 154 |
+
"text": {
|
| 155 |
+
"dtype": "string",
|
| 156 |
+
"_type": "Value"
|
| 157 |
+
}
|
| 158 |
+
},
|
| 159 |
+
"qwen_audio_chat": {
|
| 160 |
+
"answer": {
|
| 161 |
+
"dtype": "string",
|
| 162 |
+
"_type": "Value"
|
| 163 |
+
},
|
| 164 |
+
"model_prediction": {
|
| 165 |
+
"dtype": "string",
|
| 166 |
+
"_type": "Value"
|
| 167 |
+
},
|
| 168 |
+
"task_type": {
|
| 169 |
+
"dtype": "string",
|
| 170 |
+
"_type": "Value"
|
| 171 |
+
},
|
| 172 |
+
"text": {
|
| 173 |
+
"dtype": "string",
|
| 174 |
+
"_type": "Value"
|
| 175 |
+
}
|
| 176 |
+
}
|
| 177 |
+
},
|
| 178 |
+
"homepage": "",
|
| 179 |
+
"license": ""
|
| 180 |
+
}
|
examples/2ASR/GigaSpeech-Test2/sample_0.wav
ADDED
|
Binary file (285 kB). View file
|
|
|
examples/2ASR/GigaSpeech-Test2/sample_1.wav
ADDED
|
Binary file (148 kB). View file
|
|
|
examples/2ASR/GigaSpeech-Test2/sample_2.wav
ADDED
|
Binary file (126 kB). View file
|
|
|
examples/2ASR/GigaSpeech-Test2/state.json
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_data_files": [
|
| 3 |
+
{
|
| 4 |
+
"filename": "data-00000-of-00001.arrow"
|
| 5 |
+
}
|
| 6 |
+
],
|
| 7 |
+
"_fingerprint": "68d371cc267ff1d2",
|
| 8 |
+
"_format_columns": [
|
| 9 |
+
"context",
|
| 10 |
+
"instruction",
|
| 11 |
+
"answer",
|
| 12 |
+
"other_attributes",
|
| 13 |
+
"salmonn_7b",
|
| 14 |
+
"wavllm_fairseq",
|
| 15 |
+
"Qwen2-Audio-7B-Instruct",
|
| 16 |
+
"whisper_large_v3_with_llama_3_8b_instruct",
|
| 17 |
+
"mowe_audio",
|
| 18 |
+
"qwen_audio_chat"
|
| 19 |
+
],
|
| 20 |
+
"_format_kwargs": {},
|
| 21 |
+
"_format_type": null,
|
| 22 |
+
"_output_all_columns": false,
|
| 23 |
+
"_split": null
|
| 24 |
+
}
|
examples/2ASR/IMDA-Part1-ASR-Test/dataset_info.json
ADDED
|
@@ -0,0 +1,200 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"citation": "",
|
| 3 |
+
"description": "",
|
| 4 |
+
"features": {
|
| 5 |
+
"context": {
|
| 6 |
+
"text": {
|
| 7 |
+
"dtype": "string",
|
| 8 |
+
"_type": "Value"
|
| 9 |
+
},
|
| 10 |
+
"audio": {
|
| 11 |
+
"sampling_rate": 16000,
|
| 12 |
+
"_type": "Audio"
|
| 13 |
+
}
|
| 14 |
+
},
|
| 15 |
+
"instruction": {
|
| 16 |
+
"audio": {
|
| 17 |
+
"dtype": "null",
|
| 18 |
+
"_type": "Value"
|
| 19 |
+
},
|
| 20 |
+
"text": {
|
| 21 |
+
"dtype": "string",
|
| 22 |
+
"_type": "Value"
|
| 23 |
+
}
|
| 24 |
+
},
|
| 25 |
+
"answer": {
|
| 26 |
+
"audio": {
|
| 27 |
+
"dtype": "null",
|
| 28 |
+
"_type": "Value"
|
| 29 |
+
},
|
| 30 |
+
"text": {
|
| 31 |
+
"dtype": "string",
|
| 32 |
+
"_type": "Value"
|
| 33 |
+
}
|
| 34 |
+
},
|
| 35 |
+
"other_attributes": {
|
| 36 |
+
"conversation_id": {
|
| 37 |
+
"dtype": "string",
|
| 38 |
+
"_type": "Value"
|
| 39 |
+
},
|
| 40 |
+
"partition": {
|
| 41 |
+
"dtype": "string",
|
| 42 |
+
"_type": "Value"
|
| 43 |
+
},
|
| 44 |
+
"settings": {
|
| 45 |
+
"channel": {
|
| 46 |
+
"dtype": "string",
|
| 47 |
+
"_type": "Value"
|
| 48 |
+
},
|
| 49 |
+
"session": {
|
| 50 |
+
"dtype": "string",
|
| 51 |
+
"_type": "Value"
|
| 52 |
+
}
|
| 53 |
+
},
|
| 54 |
+
"speaker": {
|
| 55 |
+
"device_c0": {
|
| 56 |
+
"dtype": "string",
|
| 57 |
+
"_type": "Value"
|
| 58 |
+
},
|
| 59 |
+
"device_c1": {
|
| 60 |
+
"dtype": "string",
|
| 61 |
+
"_type": "Value"
|
| 62 |
+
},
|
| 63 |
+
"device_c2": {
|
| 64 |
+
"dtype": "string",
|
| 65 |
+
"_type": "Value"
|
| 66 |
+
},
|
| 67 |
+
"ethnic_group": {
|
| 68 |
+
"dtype": "string",
|
| 69 |
+
"_type": "Value"
|
| 70 |
+
},
|
| 71 |
+
"gender": {
|
| 72 |
+
"dtype": "string",
|
| 73 |
+
"_type": "Value"
|
| 74 |
+
},
|
| 75 |
+
"part1_id": {
|
| 76 |
+
"dtype": "string",
|
| 77 |
+
"_type": "Value"
|
| 78 |
+
},
|
| 79 |
+
"part2_id": {
|
| 80 |
+
"dtype": "string",
|
| 81 |
+
"_type": "Value"
|
| 82 |
+
},
|
| 83 |
+
"speaker_id": {
|
| 84 |
+
"dtype": "string",
|
| 85 |
+
"_type": "Value"
|
| 86 |
+
}
|
| 87 |
+
}
|
| 88 |
+
},
|
| 89 |
+
"salmonn_7b": {
|
| 90 |
+
"answer": {
|
| 91 |
+
"dtype": "string",
|
| 92 |
+
"_type": "Value"
|
| 93 |
+
},
|
| 94 |
+
"model_prediction": {
|
| 95 |
+
"dtype": "string",
|
| 96 |
+
"_type": "Value"
|
| 97 |
+
},
|
| 98 |
+
"task_type": {
|
| 99 |
+
"dtype": "string",
|
| 100 |
+
"_type": "Value"
|
| 101 |
+
},
|
| 102 |
+
"text": {
|
| 103 |
+
"dtype": "string",
|
| 104 |
+
"_type": "Value"
|
| 105 |
+
}
|
| 106 |
+
},
|
| 107 |
+
"wavllm_fairseq": {
|
| 108 |
+
"answer": {
|
| 109 |
+
"dtype": "string",
|
| 110 |
+
"_type": "Value"
|
| 111 |
+
},
|
| 112 |
+
"model_prediction": {
|
| 113 |
+
"dtype": "string",
|
| 114 |
+
"_type": "Value"
|
| 115 |
+
},
|
| 116 |
+
"task_type": {
|
| 117 |
+
"dtype": "string",
|
| 118 |
+
"_type": "Value"
|
| 119 |
+
},
|
| 120 |
+
"text": {
|
| 121 |
+
"dtype": "string",
|
| 122 |
+
"_type": "Value"
|
| 123 |
+
}
|
| 124 |
+
},
|
| 125 |
+
"Qwen2-Audio-7B-Instruct": {
|
| 126 |
+
"answer": {
|
| 127 |
+
"dtype": "string",
|
| 128 |
+
"_type": "Value"
|
| 129 |
+
},
|
| 130 |
+
"model_prediction": {
|
| 131 |
+
"dtype": "string",
|
| 132 |
+
"_type": "Value"
|
| 133 |
+
},
|
| 134 |
+
"task_type": {
|
| 135 |
+
"dtype": "string",
|
| 136 |
+
"_type": "Value"
|
| 137 |
+
},
|
| 138 |
+
"text": {
|
| 139 |
+
"dtype": "string",
|
| 140 |
+
"_type": "Value"
|
| 141 |
+
}
|
| 142 |
+
},
|
| 143 |
+
"meralion_audiollm_v1_lora": {
|
| 144 |
+
"answer": {
|
| 145 |
+
"dtype": "string",
|
| 146 |
+
"_type": "Value"
|
| 147 |
+
},
|
| 148 |
+
"model_prediction": {
|
| 149 |
+
"dtype": "string",
|
| 150 |
+
"_type": "Value"
|
| 151 |
+
},
|
| 152 |
+
"task_type": {
|
| 153 |
+
"dtype": "string",
|
| 154 |
+
"_type": "Value"
|
| 155 |
+
},
|
| 156 |
+
"text": {
|
| 157 |
+
"dtype": "string",
|
| 158 |
+
"_type": "Value"
|
| 159 |
+
}
|
| 160 |
+
},
|
| 161 |
+
"whisper_large_v3_with_llama_3_8b_instruct": {
|
| 162 |
+
"answer": {
|
| 163 |
+
"dtype": "string",
|
| 164 |
+
"_type": "Value"
|
| 165 |
+
},
|
| 166 |
+
"model_prediction": {
|
| 167 |
+
"dtype": "string",
|
| 168 |
+
"_type": "Value"
|
| 169 |
+
},
|
| 170 |
+
"task_type": {
|
| 171 |
+
"dtype": "string",
|
| 172 |
+
"_type": "Value"
|
| 173 |
+
},
|
| 174 |
+
"text": {
|
| 175 |
+
"dtype": "string",
|
| 176 |
+
"_type": "Value"
|
| 177 |
+
}
|
| 178 |
+
},
|
| 179 |
+
"qwen_audio_chat": {
|
| 180 |
+
"answer": {
|
| 181 |
+
"dtype": "string",
|
| 182 |
+
"_type": "Value"
|
| 183 |
+
},
|
| 184 |
+
"model_prediction": {
|
| 185 |
+
"dtype": "string",
|
| 186 |
+
"_type": "Value"
|
| 187 |
+
},
|
| 188 |
+
"task_type": {
|
| 189 |
+
"dtype": "string",
|
| 190 |
+
"_type": "Value"
|
| 191 |
+
},
|
| 192 |
+
"text": {
|
| 193 |
+
"dtype": "string",
|
| 194 |
+
"_type": "Value"
|
| 195 |
+
}
|
| 196 |
+
}
|
| 197 |
+
},
|
| 198 |
+
"homepage": "",
|
| 199 |
+
"license": ""
|
| 200 |
+
}
|
examples/2ASR/IMDA-Part1-ASR-Test/sample_0.wav
ADDED
|
Binary file (197 kB). View file
|
|
|
examples/2ASR/IMDA-Part1-ASR-Test/sample_1.wav
ADDED
|
Binary file (123 kB). View file
|
|
|
examples/2ASR/IMDA-Part1-ASR-Test/sample_2.wav
ADDED
|
Binary file (94.9 kB). View file
|
|
|
examples/2ASR/IMDA-Part1-ASR-Test/state.json
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_data_files": [
|
| 3 |
+
{
|
| 4 |
+
"filename": "data-00000-of-00001.arrow"
|
| 5 |
+
}
|
| 6 |
+
],
|
| 7 |
+
"_fingerprint": "ddfb47abed13c356",
|
| 8 |
+
"_format_columns": [
|
| 9 |
+
"context",
|
| 10 |
+
"instruction",
|
| 11 |
+
"answer",
|
| 12 |
+
"other_attributes",
|
| 13 |
+
"salmonn_7b",
|
| 14 |
+
"wavllm_fairseq",
|
| 15 |
+
"Qwen2-Audio-7B-Instruct",
|
| 16 |
+
"meralion_audiollm_v1_lora",
|
| 17 |
+
"whisper_large_v3_with_llama_3_8b_instruct",
|
| 18 |
+
"qwen_audio_chat"
|
| 19 |
+
],
|
| 20 |
+
"_format_kwargs": {},
|
| 21 |
+
"_format_type": null,
|
| 22 |
+
"_output_all_columns": false,
|
| 23 |
+
"_split": null
|
| 24 |
+
}
|