wjustus01/dana-voice-dataset3

dana-voice-dataset3是一个包含110个音频样本的TTS数据集，专为Unsloth微调优化。音频采样率为24000 Hz，平均音频时长为10.04秒。所有转录文本使用gpt2 tokenizer，平均token数量为43.1，最长不超过1852个token。所有样本均使用名为Dana的声音。数据集已预过滤，确保转录文本长度适合，避免在微调时因填充长度超出块大小而导致错误。数据集适用于Orpheus等TTS模型，使用Unsloth微调时需确保max_seq_length参数设置为小于等于1900。

dana-voice-dataset3 is a TTS dataset containing 110 audio samples optimized for Unsloth fine-tuning. The audio sample rate is 24000 Hz, with an average audio duration of 10.04 seconds. All transcripts use the gpt2 tokenizer, with an average of 43.1 tokens, and no transcript exceeding 1852 tokens. All samples utilize the voice named Dana. The dataset has been pre-filtered to ensure suitable transcript lengths, preventing AssertionError: Padding is larger than block size during fine-tuning. The dataset is formatted for use with TTS models like Orpheus, and when using Unsloth fine-tuning, ensure the max_seq_length parameter is set to less than or equal to 1900.