ghananlpcommunity/asante-twi-bible-speech-phonemes
收藏资源简介:
Asante Twi圣经语音—音素数据集是一个用于自动语音识别(ASR)任务的数据集,专门针对Asante Twi语言(一种加纳阿坎语方言)。该数据集基于ghananlpcommunity/asante-twi-bible-speech-text构建,并添加了音素标签,旨在训练基于wav2vec2(CTC)的音素识别模型。每个样本包含音频(16 kHz单声道)、原始Twi转录文本以及音素序列(通过twi-g2p工具生成,以空格分隔的Asante-Twi音素)。数据集已过滤掉包含数字的转录本和生成空音素字符串的示例。数据分为训练集(29,794个样本)、验证集(1,655个样本)和测试集(1,656个样本),总大小约为21.2 GB。数据集使用CC-BY-4.0许可证,适用于语音处理研究和模型训练。
The Asante Twi Bible Speech-Phoneme Dataset is a dataset dedicated to automatic speech recognition (ASR) tasks, specifically targeting the Asante Twi language, a dialect of Akan spoken in Ghana. Built upon the ghananlpcommunity/asante-twi-bible-speech-text corpus, this dataset adds phoneme labels, with the goal of training phoneme recognition models based on wav2vec2 (CTC). Each sample contains audio (16 kHz, mono channel), raw Twi transcription text, and a phoneme sequence generated by the twi-g2p tool, formatted as space-separated Asante-Twi phonemes. The dataset has filtered out transcriptions containing numbers and examples that produce empty phoneme strings. The data is split into training set (29,794 samples), validation set (1,655 samples) and test set (1,656 samples), with a total size of approximately 21.2 GB. Released under the CC-BY-4.0 license, this dataset is suitable for speech processing research and model training.




