HARPERVALLEYBANK

HARPERVALLEYBANK是一个公开的特定领域口语对话语料库，由斯坦福大学开发。该数据集模拟了简单的消费者银行业务交互，包含约23小时的音频，来自1446次人-人对话，涉及59个独特的发言者。数据集不仅提供音频数据，还附带了转录和注释，包括发言者身份、呼叫者意图、对话动作和情感极性。数据集规模适中，适合快速进行现代端到端神经方法的转录实验。此外，数据集还提供了表示学习的基准，适用于多种下游预测任务。该数据集主要用于教育和研究，旨在解决语音识别和其他口语语言任务的结合问题。

HARPERVALLEYBANK is a public domain-specific spoken dialogue corpus developed by Stanford University. This corpus simulates simple consumer banking interactions, containing approximately 23 hours of audio from 1,446 human-human conversations involving 59 unique speakers. In addition to audio data, the dataset provides transcripts and annotations including speaker identity, caller intent, dialogue acts, and sentiment polarity. With a moderate scale, the dataset is suitable for rapid transcription experiments using modern end-to-end neural methods. Furthermore, it offers benchmarks for representation learning applicable to various downstream prediction tasks. Primarily intended for educational and research purposes, this dataset aims to address the integration of speech recognition and other spoken language tasks.