Speech Analysis and Synthesis with a Computationally Efficient Adaptive Harmonic Model

Abstract

Harmonic models have to be both precise and fast in order to represent the speech signal adequately and be able to process large amount of data in a reasonable amount of time. For these purposes, the full-band adaptive harmonic model (aHM) used by the adaptive iterative refinement (AIR) algorithm has been proposed in order to accurately model the perceived characteristics of a speech signal. Even though aHM-AIR is precise, it lacks the computational efficiency that would make its use convenient for large databases. The least squares (LS) solution used in the original aHM-AIR accounts for most of the computational load. In a previous paper, we suggested a peak picking (PP) approach as a substitution to the LS solution. In order to integrate the adaptivity scheme of aHM in the PP approach, an adaptive discrete Fourier transform (aDFT), whose frequency basis can fully follow the variations of the curve, was also proposed. In this paper, we complete the previous publication by evaluating the above methods for the whole analysis process of a speech signal. Evaluations have shown an average time reduction by four times using PP and aDFT compared to the LS solution. Additionally, based on formal listening tests, when using PP and aDFT, the quality of the re-synthesis is preserved compared to the original LS-based approach.Index Terms-Fundamental frequency, harmonic model, peak picking (PP), speech analysis/synthesis, voice model.

Individual Identity in Songbirds: Signal Representations and Metric Learning for Locating the Information in Complex Corvid Calls

“…Separately we explored the use of the adaptive Discrete Fourier Transform (aDFT) proposed in [7]. This was originally introduced as an alternative, adaptive method to compute a spectrogram and the sinusoidal parameters used by the "adaptive harmonic model" (aHM) in human speech analysis and synthesis.…”

“…The values of the F 0 curve that was used as the frequency basis for aDFT were then divided by 2, because of the potential importance of sub-and inter-harmonics. Furthermore, we tested this aDFT with and without the adaptive iterative refinement algorithm, also proposed in [7], which iteratively refines the F 0 estimate used to produce the aDFT spectrogram. We therefore tested 'unrefined' and 'refined' versions.…”