Voting-based Pitch Estimation with Temporal and Frequential Alignment and Correlation Aware Selection
Junya Koguchi, Tomoki Koriyama
TL;DR
This work analyzes a voting-based ensemble for fundamental frequency estimation, identifying practical biases from analysis-frame alignment and the lack of theoretical guarantees. It provides a variance-reduction justification for median aggregation and a Condorcet-based argument for majority-vote improvements in V/UV decisions, then proposes two improvements: (i) alignment in time and frequency to correct estimator biases and (ii) a greedy procedure to select a compact, low-correlation subset of estimators. Empirical evaluation across speech, singing, and music shows that the alignment-enhanced voting method outperforms individual state-of-the-art estimators in clean conditions and maintains robust voiced/unvoiced detection under noise, with the greedy selection offering efficient yet effective ensembles. The results offer practical guidance for building robust F0 estimators across domains, particularly in noisy settings, and highlight the value of alignment and diverse, low-correlation estimator pools in ensemble methods.
Abstract
The voting method, an ensemble approach for fundamental frequency estimation, is empirically known for its robustness but lacks thorough investigation. This paper provides a principled analysis and improvement of this technique. First, we offer a theoretical basis for its effectiveness, explaining the error variance reduction for fundamental frequency estimation and invoking Condorcet's jury theorem for voiced/unvoiced detection accuracy. To address its practical limitations, we propose two key improvements: 1) a pre-voting alignment procedure to correct temporal and frequential biases among estimators, and 2) a greedy algorithm to select a compact yet effective subset of estimators based on error correlation. Experiments on a diverse dataset of speech, singing, and music show that our proposed method with alignment outperforms individual state-of-the-art estimators in clean conditions and maintains robust voiced/unvoiced detection in noisy environments.