Zixchen commited on
Commit
af4690a
·
verified ·
1 Parent(s): 999971d

Clarify benchmark overview, evaluation protocols, and dataset tags

Browse files
Files changed (2) hide show
  1. README.md +47 -1
  2. rubric_qa/README.md +2 -3
README.md CHANGED
@@ -5,8 +5,16 @@ language:
5
  - zh
6
  task_categories:
7
  - audio-classification
 
8
  tags:
9
  - audio
 
 
 
 
 
 
 
10
  configs:
11
  - config_name: rubric_qa
12
  default: true
@@ -41,7 +49,20 @@ configs:
41
 
42
  # FineAudio-Judge Benchmark
43
 
44
- FineAudio-Judge Benchmark evaluates instruction following in complex text-to-audio generation through rubric-level judgments and agreement with human preferences. It contains two evaluation components:
 
 
 
 
 
 
 
 
 
 
 
 
 
45
 
46
  | Component | Main content |
47
  |---|---|
@@ -52,6 +73,31 @@ All published audio is WAV. JSONL files use repository-relative audio paths.
52
  See the README in each component directory for field definitions, label
53
  provenance, subset names, consensus rules, and usage notes.
54
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
55
  ## Dataset structure
56
 
57
  The benchmark is organized into two components, each with its own data files,
 
5
  - zh
6
  task_categories:
7
  - audio-classification
8
+ - audio-text-to-text
9
  tags:
10
  - audio
11
+ - text
12
+ - audio-language-model
13
+ - audio-text-alignment
14
+ - instruction-following
15
+ - rubric-based-evaluation
16
+ - human-preference
17
+ - benchmark
18
  configs:
19
  - config_name: rubric_qa
20
  default: true
 
49
 
50
  # FineAudio-Judge Benchmark
51
 
52
+ [Code](https://github.com/QwenAudio/FunResearch/tree/main/FineAudio-Judge) |
53
+ [Model](https://huggingface.co/Zixchen/FineAudio-Judge)
54
+
55
+ FineAudio-Judge is a fine-grained evaluation framework for assessing
56
+ whether audio matches text descriptions or generation instructions
57
+ for complex audio scenes. It structurally decomposes the text into
58
+ events, attributes, and inter-event relations, then converts them into
59
+ atomic binary rubric questions for audio-based judgments.
60
+
61
+ This repository provides the FineAudio-Judge evaluation benchmark.
62
+ It covers real, composed, and generated audio and supports evaluation
63
+ of rubric-level judgments and agreement with human preferences.
64
+
65
+ The benchmark contains two evaluation components:
66
 
67
  | Component | Main content |
68
  |---|---|
 
73
  See the README in each component directory for field definitions, label
74
  provenance, subset names, consensus rules, and usage notes.
75
 
76
+ The benchmark is organized by evaluation component and subset.
77
+ English and Chinese records are stored together, with the language
78
+ of each record indicated by the `language` field.
79
+
80
+ ## Evaluation overview
81
+
82
+ ### Rubric-QA
83
+
84
+ This component evaluates yes/no judgments at the individual rubric
85
+ level. Each audio–rubric question pair is evaluated independently.
86
+ The reference label is determined by majority vote across five model
87
+ annotations. Predictions are compared with these reference labels
88
+ to compute rubric-level accuracy and yes/no recall.
89
+
90
+ ### Human Preference
91
+
92
+ Each candidate audio is scored by the proportion of rubric questions
93
+ answered yes. Scores are used to compare and rank candidates for the
94
+ same prompt, with accuracy measured against human pairwise
95
+ preferences and three-candidate rankings.
96
+
97
+ See the [Rubric-QA README](rubric_qa/README.md) and
98
+ [Human Preference README](human_preference/README.md)
99
+ for detailed evaluation protocols.
100
+
101
  ## Dataset structure
102
 
103
  The benchmark is organized into two components, each with its own data files,
rubric_qa/README.md CHANGED
@@ -30,9 +30,8 @@ set, such as `a1` and `cross_a1`, or the two generator-OOD subsets.
30
 
31
  ## Consistency coverage
32
 
33
- The benchmark includes controlled prompt-side and audio-side counterfactual
34
- construction to cover varying degrees of prompt–audio consistency. Each of
35
- the seven subsets contains both `yes` and `no` reference labels.
36
 
37
  Across all 33,498 questions, 23,285 reference labels are `yes` (69.51%) and
38
  10,213 are `no` (30.49%). These are majority-vote reference-label statistics.
 
30
 
31
  ## Consistency coverage
32
 
33
+ The benchmark covers varying degrees of audio–text consistency.
34
+ Each of the seven subsets contains both `yes` and `no` reference labels.
 
35
 
36
  Across all 33,498 questions, 23,285 reference labels are `yes` (69.51%) and
37
  10,213 are `no` (30.49%). These are majority-vote reference-label statistics.