GECKO — A Tool for Effective Annotation of Human Conversations

Golan Levy, Raquel Sitman, Ido Amir, Eduard Golshtein, Ran Mochary, Eilon Reshef, Roi Reichart, Omri Allouche


With the dramatic improvement in automated speech recognition (ASR) accuracy, a variety of machine learning (ML) and natural language processing (NLP) algorithms are designed for human conversation data. Supervised machine learning and particularly deep neural networks (DNNs) require large annotated datasets in order to train high quality models. In this paper we describe Gecko, a tool for annotation of speech and language features of conversations. Gecko allows efficient and effective segmentation of the voice signal by speaker as well as annotation of the linguistic content of the conversation. A key feature of Gecko is the presentation of the output of automatic segmentation and transcription systems in an intuitive user interface for editing. Gecko allows annotation of Voice Activity Detection (VAD), Diarization, Speaker Identification and ASR outputs on a large scale. Both annotators and data scientists have reported improvement in the speed and accuracy of work. Gecko is publicly available for the benefit of the community at https://github.com/gong-io/gecko.


Cite as: Levy, G., Sitman, R., Amir, I., Golshtein, E., Mochary, R., Reshef, E., Reichart, R., Allouche, O. (2019) GECKO — A Tool for Effective Annotation of Human Conversations. Proc. Interspeech 2019, 3677-3678.


@inproceedings{Levy2019,
  author={Golan Levy and Raquel Sitman and Ido Amir and Eduard Golshtein and Ran Mochary and Eilon Reshef and Roi Reichart and Omri Allouche},
  title={{GECKO — A Tool for Effective Annotation of Human Conversations}},
  year=2019,
  booktitle={Proc. Interspeech 2019},
  pages={3677--3678}
}