soynade-research/Bambara-Speech-Translation-Data

---
license: cc-by-sa-4.0
task_categories:
- automatic-speech-recognition
- translation
language:
- bm
- en
tags:
- bambara
- African-Next-Voices
- ANV
- RobotsMali
- afvoices
- asr
pretty_name: Robots Backtranslated
---

# AfVoices-Translated (Bambara-English)

This is a Bambara speech translation dataset, which is built on the **African Next Voices (AfVoices)** Bambara ASR corpus. It provides English translations for the **human-corrected subset** of the original collection, creating a parallel corpus for Bambara-English machine translation and speech-to-text tasks.

## Methodology

We machine-translated the human-validated transcriptions from AfVoices using the [Oolel-translator](https://github.com/soynade-research/oolel-translator) repository.

- **Inference Engine**: [ms-swift](https://github.com/modelscope/ms-swift) with **vLLM**.
- **Source Data**: `human-corrected` subset (~159 hours / 260k samples).
- **Status**: Machine-translated; human expert validation is the next step.

### Translation Prompt

To ensure technical consistency with the original ASR data, the following prompt was used:
> *"As an expert translator, provide only the natural English translation of the following Bambara text while preserving all tags ([um], [cs], [noise], [?], [pause]) exactly as they appear without any additional commentary"*

## Transcription Tags

This dataset preserves the original acoustic event tags to maintain synchronization with the audio:

- `[um]`: Vocalized pauses/fillers
- `[cs]`: Code-switching or foreign words
- `[noise]`: Background noise
- `[?]`: Inaudible or overlapped speech
- `[pause]`: Long silence (>3-5 seconds)

## Credits \& Acknowledgments

We would like to credit **[RobotsMali](https://huggingface.co/RobotsMali)** and the **African Next Voices (ANV)** project for the original data collection and human-corrected transcriptions. This work would not be possible without their efforts to build open-source resources for underrepresented African languages.

### Original Citation

If you use this dataset, please cite the original AfVoices paper:

```bibtex
@misc{diarra2025dealinghardfactslowresource,
title={Dealing with the Hard Facts of Low-Resource African NLP},
author={Yacouba Diarra and Nouhoum Souleymane Coulibaly and Panga Azazia Kamaté and Madani Amadou Tall and Emmanuel Élisé Koné and Aymane Dembélé and Michael Leventhal},
year={2025},
eprint={2511.18557},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2511.18557},
}
```

## Disclaimer

These translations are currently **machine-generated**. Users should be aware that while the Bambara source is human-corrected, the English "translation" column may contain model-specific errors or hallucinations until the final human validation phase is complete.

---
license: 知识共享署名-相同方式共享4.0（CC BY-SA 4.0）
task_categories:
- 自动语音识别（Automatic Speech Recognition, ASR）
- 机器翻译
language:
- 班巴拉语（bm）
- 英语（en）
tags:
- 班巴拉语（Bambara）
- African-Next-Voices
- ANV
- RobotsMali
- afvoices
- 自动语音识别（ASR）
pretty_name: 机器人回译（Robots Backtranslated）
---

# 阿福语音翻译数据集（班巴拉语-英语）

本数据集为班巴拉语语音翻译数据集，基于**阿福语音（African Next Voices, AfVoices）**班巴拉语自动语音识别语料库构建，为原始语料库的**人工校正子集**提供英语译文，从而构建出可用于班巴拉语-英语机器翻译以及语音转文字任务的平行语料库。

## 构建方法

研究团队借助[Oolel翻译器（Oolel-translator）](https://github.com/soynade-research/oolel-translator)代码库，对阿福语音语料库中的人工校验转录文本进行机器翻译。

- **推理引擎**：搭载**vLLM**的[ms-swift](https://github.com/modelscope/ms-swift)框架。
- **源数据**：`人工校正`子集（约159小时/26万条样本）。
- **当前状态**：已完成机器翻译；下一步将开展人工专家校验工作。

### 翻译提示词

为确保与原始自动语音识别数据的技术一致性，本次翻译采用如下提示词：
> *"作为专业翻译人员，请仅对下述班巴拉语文本生成自然流畅的英语译文，且需严格保留所有标签（[um]、[cs]、[noise]、[?]、[pause]）的原始格式，不得添加任何额外注释。"*

## 转录标签

本数据集保留原始声学事件标签，以确保与音频文件的时序同步：
- `[um]`：带声停顿/填充词
- `[cs]`：语码转换或外来词汇
- `[noise]`：背景噪音
- `[?]`：无法听清或重叠语音
- `[pause]`：长静音（时长超过3-5秒）

## 致谢与贡献声明

衷心感谢**[RobotsMali](https://huggingface.co/RobotsMali)**与**阿福语音（African Next Voices, ANV）**项目团队完成原始数据采集与人工校正转录工作。若没有他们为弱势非洲语言构建开源资源的努力，本研究无法顺利开展。

### 原始文献引用

若您使用本数据集，请引用阿福语音项目的原始论文：
bibtex
@misc{diarra2025dealinghardfactslowresource,
title={Dealing with the Hard Facts of Low-Resource African NLP},
author={Yacouba Diarra and Nouhoum Souleymane Coulibaly and Panga Azazia Kamaté and Madani Amadou Tall and Emmanuel Élisé Koné and Aymane Dembélé and Michael Leventhal},
year={2025},
eprint={2511.18557},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2511.18557},
}

## 免责声明

当前生成的译文均为**机器自动产出**。请注意：尽管班巴拉语源文本已完成人工校正，但英语译文列仍可能存在模型特定的错误或幻觉内容，直至最终人工校验阶段完成后方可修正。