Revisiting Input Time-frequency Representations in Multi-pitch Estimation for Vocal Ensembles
arXiv paper shows linear STFT inputs beat harmonic constant-Q transform for vocal-ensemble multi-pitch estimation while cutting feature-extraction cost.
The study compares HCQT-based input representations against a linear short-time Fourier transform (STFT) for multi-pitch estimation in vocal ensembles. The linear STFT outperforms HCQT despite fixed frequency resolution and no pitch-aligned input grid, while substantially reducing feature-extraction cost. Longer analysis windows and broader spectral coverage provided no additional improvement, and restricting inputs to the predicted pitch range reduced the STFT's advantage.
- Linear STFT outperforms HCQT for vocal-ensemble multi-pitch estimation
- Fixed-resolution STFT substantially reduces feature-extraction cost
- Shorter analysis windows are more effective for time-varying vocal pitches
From the original article
1 min read at arxiv.orgMulti-pitch estimation in vocal ensembles is challenging because singers occupy overlapping pitch ranges and often sing at closely spaced fundamental frequencies, causing their harmonics to overlap in time-frequency representations.
By Junyoung Koh · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.