Title: Speech Sentiment Annotations Authors: Eric Y. Chen, Zhiyun Lu, Hao Xu, Liangliang Cao, Yu Zhang, James Fan Language: en-US Recommended use: Sentiment Analysis TABLE OF CONTENTS 1. Summary Abstract 2. File Content 3. Annotation Task 4. Annotation Guideline 1. Summary Abstract This corpus extends the Switchboard-1 Telephone Speech Corpus (LDC97S62) by adding sentiment labels from at least 3 different human annotators for every transcript segment. Each sentiment label can be one of three options: positive, negative, and neutral. Annotators are recruited using Google Cloud's data labeling service and the labeling task was conducted over the internet. The corpus contains roughly 49500 labeled utterances covering 140 hours of audio. 2. File Content The corpus is stored in sentiment_labels.tsv as a tab-delimited TSV file. Below are two sample rows of this file: sw02005_57 400.271875 404.732375 Negative-{Scared}#Neutral-{Mixed emotions}#Negative-{Scared} sw02006_14 320.189375 324.952125 Positive-{Laughter}#Positive-{Laughter}#Positive-{Laughter} Each row is arranged in a fixed 4 columns format. These columns contain: 1) An audio marker. The substring before the "_" character directly maps to a Switchboard audio file name. e.g., The first example above "sw02005_57" refers to "sw02005.wav". 2) Estimated start time in seconds of the segment of recording that was played to annotators. 3) Estimated end time in seconds of the segment of recording that was played to annotators. 4) Sentiment annotations from at least 3 annotators. The fourth column contains a serialized string that represents the sentiment annotation. This string has the following grammar: