Speaker Detection

The Goal: Find all Contributions of a Single Speaker in a Conversation

waveFormIn Speaker Detection, you want to find all contributions (utterances) of a speaker in a conversation or discussion. Typically, you are given an audio sample of the target speaker in advance, which you use for training your algorithms. Then, you use these algorithms to analyze the recording of a conversation, where you want to find all positions where the target speaker is talking. These systems can be used, for instance, to find out who was talking most in a political discussion.

Alternative Terms: Speaker Recognition

Our Contribution

  • New algorithm for speaker detection using temporal characteristics: We use advances temporal characteristics of a speaker to improve existing algorithms for speaker detection.
  • Develop a mobile app that measures who talks how much in a job interview. (“Talkalyzer”).

Team

Publications

  1. Entwicklung einer Android-App zur Erkennung von Redeanteilen im Unterricht. Arash Besadi. Bachelor Thesis ZHAW, 2015.
  2. Talkalyzer: Mobile-App zur automatischen Sprecher-Erkennung. Lukas Kündig. Bachelor Thesis ZHAW, 2014.
  3. Talkalyzer: Neue Algorithmen für automatische Sprechererkennung. Jan Stampfli. Bachelor Thesis ZHAW, 2014.