ukrainian-tts-audiobook-pani-nina-parquet

乌克兰语TTS有声书数据集（Parquet格式）是一个用于训练和评估文本到语音（TTS）模型的分段乌克兰语语音数据集，包含对齐的文本。数据集以Hugging Face兼容的Parquet分片形式发布，支持Hub数据集预览中的`audio`列渲染。数据集通过`whisper`和`ffmpeg`工具处理，将语音切片为2-10秒的片段，并包含转录文本。数据集适用于训练乌克兰语TTS模型，提供简单的表格格式（`audio` + `text` + 元数据），与`datasets`库兼容。每个样本包含`id`、`path`、`audio`、`text`、`duration`和`source`字段。数据集统计信息包括116,575行数据，总时长约114.9小时，来自39个独特录音源。音频格式为单声道、PCM16、16 kHz WAV。

The Ukrainian TTS Audiobook Dataset (Parquet format) is a segmented Ukrainian speech dataset with aligned transcripts for training and evaluating text-to-speech (TTS) models. It is released as Hugging Face-compatible Parquet shards, and supports rendering of the `audio` column in the Hugging Face Hub dataset preview. The dataset is processed using `whisper` and `ffmpeg` tools, with speech sliced into 2-10 second segments and includes transcribed text. It is suitable for training Ukrainian TTS models, provides a simple tabular format (`audio` + `text` + metadata), and is compatible with the `datasets` library. Each sample contains the fields: `id`, `path`, `audio`, `text`, `duration`, and `source`. The dataset statistics include 116,575 rows of data, with a total duration of approximately 114.9 hours, sourced from 39 unique recording sources. The audio format is mono-channel, PCM16, 16 kHz WAV.