CHiME-9 Task 2 - ECHI

Pre-announcement - Launching in July

Enhancing Conversations to address Hearing Impairment.

Hearing impairment is a growing global issue, with the WHO projecting 2.5 billion people will have hearing loss by 2050, including 700 million needing rehabilitation [1]. Even mild hearing loss can make conversations in noisy settings, like bars or restaurants, difficult and tiring. Many adults from the age of around 40 onwards will experience this. Modern hearing aids still struggle to consistently enhance signals in these dynamic settings. However, the advent of new low power DNN-chips and other advances is opening the door to more advanced signal processing algorithms which have great potential.

The Challenge

We will provide a unique set of recordings of real conversations taking place in a noisy environment simulating a cafeteria, with a mix of surrounding distractor conversations and diffuse cafeteria noise. The conversations will be recorded over both multichannel hearing aid devices and smart glasses worn by the participants. Entrants will be asked to process the signals using real-time, low-latency algorithms in order to produce enhanced stereo output signals. The goal is to remove all background noise while leaving the speech of the conversational partners intact, i.e., with little or no distortion.

Evaluation

Challenge entrants will be evaluated using a held-out evaluation set with conversations and talkers unseen during training or development. The enhanced signals will be evaluated through a range of listening tests designed to assess the quality and intelligibility of the conversation. We will also provide a baseline system and a number of signal-level metrics to help entrants gauge their performance during development.

Further details of the data and the task design will be announced at the end of March. The challenge with the full data, baseline and metrics will be launched on July 1st.

Organizers

  • Jon Barker (University of Sheffield, UK),
  • Stefan Goetze (University of Sheffield, UK),
  • Robert Sutherland (University of Sheffield, UK),
  • Marko Lugger (WSAudiology),
  • Thomas Kuebert (WSAudiology),
  • Stefan Petrausch (WSAudiology)