ghananlpcommunity/twi-agriculture-speech

Twi农业语音数据集是一个针对加纳最广泛使用的语言Twi的领域特定语音识别数据集，数据来源于公开的农业相关视频内容。该数据集由Mich-Seth Owusu创建，并由GhanaNLP社区发布，旨在支持加纳语言（特别是农业词汇和话语）的开源ASR工具开发。音频格式为WAV、16 kHz单声道，转录是通过Google Speech Recognition API（免费版）自动生成的，但转录质量可能较低，因为Twi是一种低资源语言，且农业术语可能增加错误率。数据集包含15,174个训练样本，每个样本包括音频波形和对应的Twi文本转录。它主要用于训练和微调Twi的ASR模型，特别是在农业领域，也可用于引导加纳语言的转录流程改进、低资源语音识别系统基准测试以及支持加纳的农业技术和语言保护工作。

The Twi Agriculture Speech Dataset is a domain-specific speech recognition dataset for Twi, one of Ghanas most widely spoken languages, sourced from publicly available video content on agriculture. Created by Mich-Seth Owusu and published under the GhanaNLP Community, it aims to support the development of open-source ASR tools for Ghanaian languages, starting with agricultural vocabulary and discourse. The audio format is WAV, 16 kHz mono, and transcriptions were generated automatically using the Google Speech Recognition API (free tier), though the quality may be lower due to Twi being a low-resource language and agricultural terminology potentially increasing error rates. The dataset contains 15,174 training examples, each consisting of an audio waveform and corresponding Twi text transcription. It is intended for training and fine-tuning ASR models for Twi, especially in the agriculture domain, bootstrapping better transcription pipelines for Ghanaian languages, benchmarking low-resource speech recognition systems, and supporting agri-tech and language preservation efforts in Ghana.