SpeechPPL/SALMon_TWIST-350M

该数据集包含了多种配置，每种配置都有特定的特征和文件路径。数据集被分为不同的配置，包括bg_alignment、bg_all_consistency、bg_domain_consistency、gender_consistency、rir_consistency、sentiment_alignment、sentiment_consistency和speaker_consistency。每个配置包括任务、索引、正负音频样本、提示音频、延续音频、词级损失、原始单元、代码帧率、代码深度、偏移量、模型采样率、困惑度检查以及模型生成的延续音频。数据集主要用于训练音频处理模型，专注于在不同音频特征和条件下的一致性和对齐任务。

This dataset includes various configurations, each with specific features and file paths. The dataset is split into different configurations such as bg_alignment, bg_all_consistency, bg_domain_consistency, gender_consistency, rir_consistency, sentiment_alignment, sentiment_consistency, and speaker_consistency. Each configuration includes features like task, indices, positive and negative audio samples, prompt audio, continuation audio, tokenwise losses, raw units, code frame rate, code depth, offset, model sampling rate, perplexity for sanity check, and model generated continuation audio. The dataset is primarily designed for training audio processing models with a focus on consistency and alignment tasks across different audio features and conditions.