SoccerNet-Echoes / README.md
SushantGautam's picture
Convert dataset to Parquet (#6)
6d1fa9e verified
|
raw
history blame
4.35 kB
---
language:
- en
- de
license: apache-2.0
task_categories:
- text-classification
- summarization
dataset_info:
- config_name: whisper_v1
features:
- name: segment_index
dtype: string
- name: start_time
dtype: float32
- name: end_time
dtype: float32
- name: transcribed_text
dtype: string
- name: game
dtype: string
splits:
- name: train
num_bytes: 110289348
num_examples: 780160
download_size: 34176839
dataset_size: 110289348
- config_name: whisper_v1_en
features:
- name: segment_index
dtype: int32
- name: start_time
dtype: float32
- name: end_time
dtype: float32
- name: transcribed_text
dtype: string
splits:
- name: train
num_bytes: 31843296
num_examples: 563064
download_size: 96617459
dataset_size: 31843296
- config_name: whisper_v2
features:
- name: segment_index
dtype: int32
- name: start_time
dtype: float32
- name: end_time
dtype: float32
- name: transcribed_text
dtype: string
splits:
- name: train
num_bytes: 47409793
num_examples: 761240
download_size: 96617459
dataset_size: 47409793
- config_name: whisper_v2_en
features:
- name: segment_index
dtype: string
- name: start_time
dtype: float32
- name: end_time
dtype: float32
- name: transcribed_text
dtype: string
splits:
- name: train
num_bytes: 32198972
num_examples: 538990
download_size: 96617459
dataset_size: 32198972
- config_name: whisper_v3
features:
- name: segment_index
dtype: string
- name: start_time
dtype: float32
- name: end_time
dtype: float32
- name: transcribed_text
dtype: string
splits:
- name: train
num_bytes: 52528392
num_examples: 923221
download_size: 96617459
dataset_size: 52528392
configs:
- config_name: whisper_v1
data_files:
- split: train
path: whisper_v1/train-*
default: true
---
# SoccerNet-Echoes
Official repo for the paper: [SoccerNet-Echoes: A Soccer Game Audio Commentary Dataset](https://arxiv.org/abs/2405.07354).
## Dataset
Each folder inside the **Dataset** directory is categorized by league, season, and game. Within these folders, JSON files contain the transcribed and translated game commentary.
```python
πŸ“‚ Dataset
β”œβ”€β”€ πŸ“ whisper_v1
β”‚ β”œβ”€β”€ πŸ† england_epl
β”‚ β”‚ β”œβ”€β”€ πŸ“… 2014-2015
β”‚ β”‚ β”‚ └── ⚽ 2016-03-02 - 23-00 Liverpool 3 - 0 Manchester City
β”‚ β”‚ β”‚ β”œβ”€β”€ ☁️ 1_asr.json
β”‚ β”‚ β”‚ └── ☁️ 2_asr.json
β”‚ β”‚ β”œβ”€β”€ πŸ“… 2015-2016
β”‚ β”‚ └── ...
β”‚ β”œβ”€β”€ πŸ† europe_uefa-champions-league
β”‚ └── ...
β”œβ”€β”€ πŸ“ whisper_v1_en
β”‚ └── ...
β”œβ”€β”€ πŸ“ whisper_v2
β”‚ └── ...
β”œβ”€β”€ πŸ“ whisper_v2_en
β”‚ └── ...
β”œβ”€β”€ πŸ“ whisper_v3
β”‚ └── ...
whisper_v1: Contains ASR from Whisper v1.
whisper_v1_en: English-translated datasets from Whisper v1.
whisper_v2: Contains ASR from Whisper v2.
whisper_v2_en: English-translated datasets from Whisper v2.
whisper_v3: Contains ASR from Whisper v3.
```
Each JSON file has the following format:
```python
{
"segments": {
segment index (int):[
start time in second (float),
end time in second (float),
transcribed text from ASR
]
....
}
}
```
The top-level object is named segments.
It contains an object where each key represents a unique segment index (e.g., "0", "1", "2", etc.).
Each segment index object has the following properties:
```python
start_time: A number representing the starting time of the segment in seconds.
end_time: A number representing the ending time of the segment in seconds.
text: A string containing the textual content of the commentary segment.
```
## Citation
Please cite our work if you use the SoccerNet-Echoes dataset:
<pre><code>
@misc{gautam2024soccernetechoes,
title={SoccerNet-Echoes: A Soccer Game Audio Commentary Dataset},
author={Sushant Gautam and Mehdi Houshmand Sarkhoosh and Jan Held and Cise Midoglu and Anthony Cioppa and Silvio Giancola and Vajira Thambawita and Michael A. Riegler and PΓ₯l Halvorsen and Mubarak Shah},
year={2024},
eprint={2405.07354},
archivePrefix={arXiv},
primaryClass={cs.SD},
doi={10.48550/arXiv.2405.07354}
}
</code></pre>