ghananlpcommunity/ghana-english-tts-clean2
收藏资源简介:
Ghana English TTS Filtered Clean v2 是一个文本到语音(TTS)数据集,专门针对加纳口音的英语。该数据集是基于 ghananlpcommunity/ghana-english-tts-filtered 数据集的过滤子集,使用 PANNs CNN14 模型进行二次过滤处理。过滤条件包括:音乐概率 ≤ 0.2、掌声概率 ≤ 0.2 且语音概率 ≥ 0.5,以确保音频质量。数据集包含 282,096 个音频片段,每个片段包含 corrected_text(校正后的文本)和 bytes(原始 WAV 字节,16 位 PCM 单声道,原始采样率)字段。该数据集主要用于预计算 VoxCPM2 AudioVAE V2 的潜在表示,以用于加纳口音英语 TTS 的微调任务。
Ghana English TTS Filtered Clean v2 is a text-to-speech (TTS) dataset specifically tailored for Ghanaian-accented English. It is a filtered subset derived from the ghananlpcommunity/ghana-english-tts-filtered dataset, which underwent secondary filtering using the PANNs CNN14 model. The filtering criteria are set as: music probability ≤ 0.2, applause probability ≤ 0.2, and speech probability ≥ 0.5, to guarantee high audio quality. The dataset comprises 282,096 audio clips, each containing the fields corrected_text (corrected text) and bytes (raw WAV bytes, 16-bit PCM mono, original sampling rate). This dataset is primarily utilized to precompute the latent representations of VoxCPM2 AudioVAE V2 for fine-tuning tasks of Ghanaian-accented English TTS.




