HarperValleyBank: A Domain-Specific Spoken Dialog Corpus

We introduce HarperValleyBank, a free, public domain spoken dialog corpus. The data simulate simple consumer banking interactions, containing about 23 hours of audio from 1,446 human-human conversations between 59 unique speakers. We selected intents and utterance templates to allow realistic variation while controlling overall task complexity and limiting vocabulary size to about 700 unique words. We provide audio data along with transcripts and annotations for speaker ID, caller intent, dialog actions, and emotional valence. The size and domain specificity of this data makes for quick experiments with modern end-to-end neural approaches. Further, we provide baselines for representation learning and transfer tasks. These experiments adapt recent work to embed utterances and use the resulting representations in prediction tasks. Our experiments show that tasks using our annotations are sensitive to both the model choice and corpus size for representation learning approaches.