qwen3-tts-african-blindspots

该数据集名为'Qwen3-TTS非洲语言盲点'，旨在记录Qwen3-TTS-12Hz-1.7B-CustomVoice模型在处理非洲语言输入时的失败模式。数据集覆盖了7种不同的错误类别，包括语言不在训练集中、音调符号处理失败、代码切换失败、命名实体发音错误、点击辅音缺失、非洲英语变体处理不当以及粘着形态学问题。这些错误类别涉及多种非洲语言，如斯瓦希里语、约鲁巴语、祖鲁语等。数据集包含26个训练样本，每个样本包含id、错误类别、语言标签、输入文本、预期输出、模型输出描述、修正建议和音频文件等字段。该数据集的建立基于对非洲语言多样性的重视，旨在为语音合成技术的改进提供参考，特别是在处理非洲语言时的特定挑战。

This dataset, named 'Qwen3-TTS African Language Blind Spot', aims to document the failure modes of the Qwen3-TTS-12Hz-1.7B-CustomVoice model when processing input in African languages. The dataset covers 7 distinct error categories, including language not present in the training corpus, failure to handle tone marks, code-switching failure, mispronunciation of named entities, absence of click consonants, improper handling of African English varieties, and issues related to agglutinative morphology. These error categories involve a variety of African languages, such as Swahili, Yoruba, Zulu, and others. The dataset contains 26 training samples, with each sample including fields such as id, error category, language label, input text, expected output, model output description, correction suggestions, and audio files. This dataset was developed with an emphasis on the diversity of African languages, aiming to provide a reference for the improvement of speech synthesis technologies, particularly regarding the specific challenges encountered when processing African languages.