nilc-nlp/CORAA-NURC-SP-Audio-Corpus

# NURC-SP Corpus

NURC-SP Corpus CORAA ASR is a publicly available dataset for Automatic Speech Recognition (ASR) in the Brazilian Portuguese language containing 239.68 hours of audios ( 239.30 when filtered ) and their respective transcriptions (170k+ segmented audios).

The audios were either validated by annotators or transcripted for the first time aiming at the ASR task.

## How to Use

The datasets library allows easy loading of the dataset with the load_dataset() function.

To load the dataset, you need to pass the name "original" (full list of audio segments) or "filtered" (audio segments with empty transcription removed)
```python
from datasets import load_dataset

dataset_original = load_dataset("nilc-nlp/CORAA-NURC-SP-Audio-Corpus", name="original")
dataset_filtered = load_dataset("nilc-nlp/CORAA-NURC-SP-Audio-Corpus", name="filtered")
```

## Metadata

- audio_name: The name given to the audio in the database. All audios extracted from the same source have the same name.
- file_path: The path to the audio file.
- text: The human-verified trancription for the given audio.
- start_time: The time the audio segment starts in the original source in seconds.
- end_time: The time the audio segment ends in the original source in seconds.
- duration: The duration of the audio segment in seconds.
- quality: Whether or not the audio had parts that could not be transcribed properly. Audios without this characteristic are rated 'high' and audios with it are rated 'low'.
- speech_genre: The speech genre of the original source of the segment. Divided into 'dialogue', 'interview' or 'lecture and talks'.
- speech_style: The speech style of the original source of the segment. All segments are categorized as 'spontaneous speech'.
- variety: The audio language. All segments are categorized as 'pt-br'.
- accent: The speaker's accent. All segments are categorized as 'sp-city'. Note that some audio sources have more than one speaker, so in that case the accent refers to the main speaker or speakers.
- sex: The speaker's sex. Divided into 'F', 'M', 'F e F', 'F e M' and 'M e M' ('F' stands for female and 'M' stands for male). Note that some audio sources have more than one speaker, so in that case the sex refers to the main speaker or speakers.
- age_range: The speaker's age range. Divided into 'I' (25 to 35), 'II' (36 to 55) and 'III' (over 55). Note that some audio sources have more than one speaker, so in that case the age range refers to the main speaker or speakers.
- num_speakers: The number of speakers in the original source of the segment. This field was automatically writter by WhisperX, so it might not be accurate.
- speaker_id: The speaker in the segment (each different speaker in the original source was given an integer id). This field was automatically writter by WhisperX, so it might not be accurate.

## Downloads
Currently, download is only available in the Files section
| Hugging Face |
| ------------ |
| [Download Link]() |

## Citation
Lima, R., Leal, S.E., Candido Junior, A., Aluísio, S.M. A Large Dataset of Spontaneous Speech with the Accent Spoken in São Paulo for Automatic Speech Recognition Evaluation. Proceedings of the 34th Brazilian Conference on Intelligent Systems (BRACIS) (2024).

````
@InProceedings{nurc-sp-audio-corpus-2024,
author = {Rodrigo Lima and Sidney Evaldo Leal and Arnaldo Candido Junior and Sandra Maria Aluisio},
title = {A Large Dataset of Spontaneous Speech with the Accent Spoken in São Paulo for Automatic Speech Recognition Evaluation},
booktitle = {Proceedings of 34th Brazilian Conference on Intelligent Systems (BRACIS)},
year = {2024}
}
````

## Sponsors / Funding

This work was carried out at the Center for Artificial Intelligence (C4AI-USP), with support by the São Paulo Research Foundation (FAPESP grant \#2019/07665-4) and by the IBM Corporation. This project was also supported by the Ministry of Science, Technology and Innovation, with resources of Law No. 8.248, of October 23, 1991, within the scope of PPI-SOFTEX, coordinated by Softex and published Residence in TIC 13, DOU 01245.010222/2022-44.

# NURC-SP语料库

NURC-SP语料库（CORAA ASR）是面向巴西葡萄牙语自动语音识别（Automatic Speech Recognition，ASR）的公开可用数据集，包含239.68小时音频（过滤后为239.30小时）及对应转录文本，共计超过17万条分段音频。

这些音频要么经标注人员校验，要么专为自动语音识别任务首次转录。

## 使用方法

datasets库可通过`load_dataset()`函数轻松加载该数据集。加载时需传入参数`"original"`（完整音频分段列表）或`"filtered"`（已移除空转录文本的音频分段）：
python
from datasets import load_dataset

dataset_original = load_dataset("nilc-nlp/CORAA-NURC-SP-Audio-Corpus", name="original")
dataset_filtered = load_dataset("nilc-nlp/CORAA-NURC-SP-Audio-Corpus", name="filtered")

## 元数据

- audio_name: 数据库中为音频分配的名称。同一来源提取的所有音频具有相同名称。
- file_path: 音频文件的存储路径。
- text: 经人工校验的音频转录文本。
- start_time: 音频分段在原始素材中的起始时间，单位为秒。
- end_time: 音频分段在原始素材中的结束时间，单位为秒。
- duration: 音频分段的时长，单位为秒。
- quality: 音频是否存在无法正常转录的片段。无此类问题的音频评级为「高」，存在问题的评级为「低」。
- speech_genre: 音频分段所属原始素材的语音体裁，分为「对话」「访谈」或「讲座与演讲」三类。
- speech_style: 音频分段所属原始素材的语音风格，所有分段均归类为「自发口语」。
- variety: 音频所用语言，所有分段均归类为「pt-br」。
- accent: 说话者的口音，所有分段均归类为「sp-city」。需注意，部分音频来源包含多位说话者，此时口音指代主要说话者或多数说话者的口音。
- sex: 说话者的性别，分为「F」（女性）、「M」（男性）、「F e F」（女性与女性）、「F e M」（女性与男性）及「M e M」（男性与男性）五类。需注意，部分音频来源包含多位说话者，此时性别指代主要说话者或多数说话者的性别。
- age_range: 说话者的年龄区间，分为「I」（25至35岁）、「II」（36至55岁）及「III」（55岁以上）三类。需注意，部分音频来源包含多位说话者，此时年龄区间指代主要说话者或多数说话者的年龄区间。
- num_speakers: 音频分段所属原始素材中的说话者数量。该字段由WhisperX自动生成，可能存在误差。
- speaker_id: 音频分段中的说话者标识（原始素材中的每位不同说话者均被分配一个整数ID）。该字段由WhisperX自动生成，可能存在误差。

## 下载方式

目前仅可在Files板块下载：
| Hugging Face |
| ------------ |
| [下载链接]() |

## 引用格式

Lima, R., Leal, S.E., Candido Junior, A., Aluísio, S.M. 面向自动语音识别评估的圣保罗口音自发口语大型数据集. 第34届巴西智能系统大会（BRACIS）论文集（2024）。

bibtex
@InProceedings{nurc-sp-audio-corpus-2024,
author = {Rodrigo Lima and Sidney Evaldo Leal and Arnaldo Candido Junior and Sandra Maria Aluisio},
title = {A Large Dataset of Spontaneous Speech with the Accent Spoken in São Paulo for Automatic Speech Recognition Evaluation},
booktitle = {Proceedings of 34th Brazilian Conference on Intelligent Systems (BRACIS)},
year = {2024}
}

## 资助与支持

本研究在人工智能中心（C4AI-USP）开展，得到圣保罗研究基金会（FAPESP，资助号#2019/07665-4）及IBM公司的支持。本项目同时得到巴西科学、技术与创新部资助，资金来源于1991年10月23日第8.248号法律，隶属于由Softex协调的PPI-SOFTEX框架，相关公告发布于《巴西官方公报》（DOU）2022年第13期TIC驻场栏目，编号01245.010222/2022-44。