Datasets:

Modalities:
Text
Formats:
parquet
Libraries:
Datasets
pandas
License:
yentinglin commited on
Commit
a0c69cd
·
verified ·
1 Parent(s): d15e91e

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +49 -0
README.md CHANGED
@@ -31,4 +31,53 @@ configs:
31
  path: data/train-*
32
  - split: test
33
  path: data/test-*
 
34
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
31
  path: data/train-*
32
  - split: test
33
  path: data/test-*
34
+ license: apache-2.0
35
  ---
36
+
37
+ # Dataset Name: Pilot dataset for Multi-domain ASR corrections
38
+
39
+ ## Description
40
+
41
+ Consolidated from [PeacefulData/HyPoradise-v0](https://huggingface.co/datasets/PeacefulData/HyPoradise-v0)
42
+
43
+ ## Structure
44
+
45
+ ### Data Split
46
+
47
+ - **Training Data**: 168,460 entries
48
+ - **Test Data**: 6,992 entries
49
+
50
+ ### Columns
51
+
52
+ - `hypothesis`: N-best hypothesis from beam search.
53
+ - `transcription`: Corrected asr transcription.
54
+ - `hypothesis_concatenated`: An alternative version of the text output.
55
+ - `source`: The source of the text entry, indicating the origin dataset.
56
+ - `score`: An acoustic model score (not all entries have this).
57
+
58
+ ### Source Distribution
59
+
60
+ - **Training Sources**:
61
+ - `train_cv`: 47,293 entries
62
+ - `train_wsj`: 37,514 entries
63
+ - `train_swbd`: 36,539 entries
64
+ - `train_chime4`: 9,600 entries
65
+
66
+ - **Test Sources**:
67
+ - `test_swbd`: 2,000 entries
68
+ - `test_cv`: 2,000 entries
69
+ - `test_chime4`: 1,320 entries
70
+ - `test_wsj`: 836 entries
71
+
72
+ ## Access
73
+
74
+ The dataset can be accessed and downloaded through the HuggingFace Datasets library. Use the following command to load the dataset:
75
+
76
+ ```python
77
+ from datasets import load_dataset
78
+ dataset = load_dataset("PeacefulData/HyPoradise-pilot")
79
+ ```
80
+
81
+ ## Acknowledgments
82
+
83
+ Thanks https://huggingface.co/datasets/PeacefulData/HyPoradise-v0 for sharing this dataset.