Aim

The scope of this internship project dealt with different ways of rendering spatial audio that could be employed to improve their current research and generate a more immersive and realistic user experience.

Research

Detailed research into binaural audio, signal processing, spatial audio rendering and ear anatomy was carried out to get a background into how people hear, why they hear what they hear and how to recreate realistic audio recordings.

This information was used to develop binaural audio with different acoustic attributes such as room impulse responses, apparent source width and changing decibel level in Python. Different functions were developed for the different audio attributes.

Example of Rendered Audio

This audio created using a Python funciton that allowed the user input an audio signal and the direction the audio is coming from. In this sample the audio is travelling around the back of the listner.

Better listened to with earphones

User Testing

The audio that was developed was used in subjective listening tests which were designed in accordance with International Telecommunication Recommendations, as subjective listening tests are recognised as being one of the most reliable ways of measuring the quality of any audio system.

It is imperative that the correct audio samples are used in listening tests. Harvard sentences were used in these subjective listening tests. These sentences were originally developed during WWII in research to improve communication in combat aircrafts, but now are widely used in the testing of telecommunication devices or by speech scientists and engineers. The sentences are phonetically balanced leading to the frequency of the sentences matching that of the natural language.

Speach

Should be able to hear the voice moving from left to right around your head.

Better listened to with earphones

Distance & Multiple Speakers

If listened to closely you can hear the voices moving and should be able to distinguish which voice is furthest away from you.

Better listened to with earphones

What I learned

  • Taught myself Python. Initially the audio was to be created in MATLAB. However, as Python is more widely used, and is free in comparison to MATLAB’s licence, it was decided that I would learn Python.
  • Taught myself how to use Blender.
  • Formal research, paper formatting & data presentation
  • Developed knowledge of virtual reality audio.
  • Improved presentation & communication skills through regular progress update meetings and presentations for stakeholders with different levels of knowledge into binaurla audio. Here I learned to adapt my presentations for different audiences, depending on their depth of knowledge in the area.