Harmonic Differences Method for Robust Fundamental Frequency Detection in Wideband and Narrowband Speech Signals

In this article, a novel pitch determination algorithm based on harmonic differences method (HDM) is proposed. Most of the algorithms today rely autocorrelation, cepstrum, and lastly convolutional neural networks, they have some limitations (small datasets, wideband or narrowband, musical sounds, temporal smoothing, etc.), accuracy, speed problems. There are very rare works exploiting spacing between harmonics. HDM designed for both exclusively narrowband (telephone) speech tries to find most repeating difference harmonics signal. We use three vowel databases in our experiments, namely, Hillenbrand Vowel Database, Texas Vowels from TIMIT corpus. compare with YIN, YAAPT, CREPE, FCN algorithms. Results show that reliable fast choice robust detection. Also, it superior others cases.