Distance estimation of bird-calls using direct to reverberant energy ratio

Abstract

Kiwi (Apterygidae.) are a vocally cryptic species and are of great national importance to New Zealand. Their endangered conservation status requires active annual monitoring. To assist kiwi conservation efforts, this thesis explores existing kiwi acoustic monitoring techniques using distance estimation. Distance information can be extracted from bird-call recordings using human perception estimates, signal power models and triangulation. However, these existing distance estimation methods are too time-consuming, unreliable and logistically difficult to use. Thus, this thesis investigates the potential of an indoor acoustic parameter - direct-to-reverberant energy ratio (DRR), to model the distance from a kiwi call. There are two primary research questions: 1. Can DRR replace signal power in distance estimation models of bird-calls? 2. Can existing blind DRR estimation methods perform well in outdoor environments? Both research questions are explored using impulse responses (IR) collected from three testing sites: Nihotupu Dam and Lover’s Walk in Auckland, and Ponui Island in the Hauraki Gulf. 400 measurements were collected at each site with 20 IR measurements collected at each distance from 1-20 m in Ponui + Lover’s, and 10 IR measurements collected at each distance from 1-40 m at Nihotupu. The MiniDSP UMA-8SP equilateral triangle microphone array was used to collect the IR data. The kiwi call was simulated through convolving a Department of Conservation North Island brown kiwi recording with the impulse responses collected at each testing site. The first research question compared the DRR calculated from impulse response (IR DRR) against impulse response power (IR power) at each distance. The key findings were that IR DRR functioned well in low SNR environments and generated similar accuracies across all distances. The second research question explored the performance of an existing state-of-the-art blind DRR estimation technique developed for indoor environments, on the simulated kiwi call. The findings showed that the blindly estimated DRR significantly overestimated both the direct and reverberant energy of the kiwi call and therefore resulted in inaccurate DRR estimates. The future recommendations of this research are to explore modifications to the existing blind DRR estimation algorithm to improve the feasibility of using DRR for modelling distance from bird-calls.