README.md first commit
Browse filesAdding descriptions to the dataset.
README.md
CHANGED
@@ -1,3 +1,143 @@
|
|
1 |
---
|
2 |
license: apache-2.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: apache-2.0
|
3 |
+
|
4 |
+
dataset_info:
|
5 |
+
features:
|
6 |
+
- name: session_id
|
7 |
+
dtype: string
|
8 |
+
- name: start_time
|
9 |
+
dtype: float
|
10 |
+
- name: end_time
|
11 |
+
dtype: float
|
12 |
+
- name: words
|
13 |
+
dtype: string
|
14 |
+
- name: speaker
|
15 |
+
dtype: string
|
16 |
+
splits:
|
17 |
+
- name: dev
|
18 |
+
num_bytes:
|
19 |
+
num_examples: 142
|
20 |
+
- name: eval
|
21 |
+
num_bytes:
|
22 |
+
num_examples: 104
|
23 |
+
download_size:
|
24 |
+
dataset_size:
|
25 |
---
|
26 |
+
|
27 |
+
|
28 |
+
# Dataset Name: Dataset for ASR Speaker-Tagging Corrections (Speaker Diarization)
|
29 |
+
|
30 |
+
|
31 |
+
## Description
|
32 |
+
|
33 |
+
- This dataset is pairs of erroneous ASR output and speaker tagging, which are generated from a ASR system and speaker diarization system.
|
34 |
+
Each source erroneous transcription is paired with human-annotated transcription, which has correct transcription and speaker tagging.
|
35 |
+
- [SEGment-wise Long-form Speech Transcription annotation](#segment-wise-long-form-speech-transcription-annotation-seglst) (`SegLST`), the file format used in the [CHiME challenges](https://www.chimechallenge.org)
|
36 |
+
|
37 |
+
|
38 |
+
Example) `session_ge1nse2c.seglst.json`
|
39 |
+
|
40 |
+
```
|
41 |
+
[
|
42 |
+
...
|
43 |
+
{
|
44 |
+
"session_id": "session_ge1nse2c",
|
45 |
+
"words": "well that is the problem we have erroneous transcript and speaker tagging we want to correct it using large language models",
|
46 |
+
"start_time": 181.88,
|
47 |
+
"end_time": 193.3,
|
48 |
+
"speaker": "speaker1"
|
49 |
+
},
|
50 |
+
{
|
51 |
+
"session_id": "session_ge1nse2c",
|
52 |
+
"words": "it seems like a really interesting problem I feel that we can start with very simple methods",
|
53 |
+
"start_time": 194.48,
|
54 |
+
"end_time": 205.03,
|
55 |
+
"speaker": "speaker2"
|
56 |
+
},
|
57 |
+
...
|
58 |
+
]
|
59 |
+
```
|
60 |
+
|
61 |
+
## Structure
|
62 |
+
|
63 |
+
### Data Split
|
64 |
+
|
65 |
+
The dataset is divided into training and test splits:
|
66 |
+
|
67 |
+
- Development Data: 142 entries
|
68 |
+
- 2 to 4 speakers in each session
|
69 |
+
- Approximately 10 ~ 40 mins of recordings
|
70 |
+
- Evaluation Data: 104 entries
|
71 |
+
- 2 to 4 speakers in each session
|
72 |
+
- Approximately 10 ~ 40 mins of recordings
|
73 |
+
|
74 |
+
### Keys (items)
|
75 |
+
|
76 |
+
`session_id`: "session_ge1nse2c",
|
77 |
+
`words": Transcription corresponding to t
|
78 |
+
`start_time`: Start time in second.
|
79 |
+
`end_time`: End time in second.
|
80 |
+
`speaker`: Speaker tagging in string: "speaker<N>"
|
81 |
+
|
82 |
+
### Source Datasets
|
83 |
+
|
84 |
+
The dataset combines entries from various sources:
|
85 |
+
|
86 |
+
- **Development Sources**:
|
87 |
+
- `dev`: 142 sessions
|
88 |
+
|
89 |
+
- **Evaluation Sources**:
|
90 |
+
- `eval`: 104 Sessions
|
91 |
+
|
92 |
+
|
93 |
+
|
94 |
+
## Access
|
95 |
+
|
96 |
+
The dataset can be accessed and downloaded through the HuggingFace Datasets library.
|
97 |
+
|
98 |
+
|
99 |
+
## Evaluation
|
100 |
+
|
101 |
+
This dataset can be evaluated by [MeetEval Software](https://github.com/fgnt/meeteval)
|
102 |
+
|
103 |
+
### From PyPI
|
104 |
+
```
|
105 |
+
pip install meeteval
|
106 |
+
```
|
107 |
+
|
108 |
+
### From source
|
109 |
+
```
|
110 |
+
git clone https://github.com/fgnt/meeteval
|
111 |
+
pip install -e ./meeteval
|
112 |
+
```
|
113 |
+
|
114 |
+
### Evaluate the corrected segLST files:
|
115 |
+
```
|
116 |
+
python -m meeteval.wer cpwer -h err_source_text/dev/session_ge1nse2c.json -r ref_annotate_text/dev/session_ge1nse2c.json
|
117 |
+
```
|
118 |
+
Or after installation, you can use the following command alternatively.
|
119 |
+
```
|
120 |
+
meeteval-wer cpwer -h err_source_text/dev/session_ge1nse2c.json -r ref_annotate_text/dev/session_ge1nse2c.json
|
121 |
+
```
|
122 |
+
|
123 |
+
### References
|
124 |
+
|
125 |
+
```bib
|
126 |
+
@inproceedings{park2024enhancing,
|
127 |
+
title={Enhancing speaker diarization with large language models: A contextual beam search approach},
|
128 |
+
author={Park, Tae Jin and Dhawan, Kunal and Koluguri, Nithin and Balam, Jagadeesh},
|
129 |
+
booktitle={ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
|
130 |
+
pages={10861--10865},
|
131 |
+
year={2024},
|
132 |
+
organization={IEEE}
|
133 |
+
}
|
134 |
+
```
|
135 |
+
|
136 |
+
```bib
|
137 |
+
@InProceedings{MeetEval23,
|
138 |
+
title={MeetEval: A Toolkit for Computation of Word Error Rates for Meeting Transcription Systems},
|
139 |
+
author={von Neumann, Thilo and Boeddeker, Christoph and Delcroix, Marc and Haeb-Umbach, Reinhold},
|
140 |
+
booktitle={CHiME-2023 Workshop, Dublin, England},
|
141 |
+
year={2023}
|
142 |
+
}
|
143 |
+
```
|