Regression Tree Approach to Studying Factors Influencing Acoustic Voice Analysis

Abstract

Multiple factors influence voice quality measurements (VQM) obtained during an acoustic voice assessment including: gender, intrasubject variability, microphone, environmental noise (type and level), data acquisition (DA) system, and analysis software. This study used regression trees to investigate the order and relative importance of these factors on VQM including interaction effects of the factors and how the outcome differs when the acoustic environment is controlled for noise. Twenty normophonic participants provided 20 voice samples each, which were recorded synchronously on five DA systems combined with six different microphones. The samples were mixed with five noise types at eight signal-to-noise ratio (SNR) levels. The resulting 80,000 audio samples were analyzed for fundamental frequency (F₀), jitter and shimmer using three software analysis systems: MDVP, PRAAT, and TF32 (CSpeech). Fifteen regression trees and their Variable Importance Measures were utilized to analyze the data. The analyses confirmed that all of the factors listed above were influential. The results suggest that gender, intrasubject variability, and microphone were significant influences on F₀. Software systems and gender were highly influential on measurements of jitter and shimmer. Environmental noise was shown to be the prominent factor that affects VQM when SNR levels are below 30 dB.

Comparison of voice acquisition methodologies in speech research

“…These results replicated or supported a number of studies that have demonstrated the robust and invariant nature of key frequency and timing-based measures (Mundt et al, 2007), and the sensitive character of supplementary acoustic parameters (Carding et al, 2004;Deliyski et al, 2006).…”

“…Deliyski and colleagues have conducted a series of comprehensive studies investigating the influence of data acquisition environments (Deliyski, Shaw, Evans, & Vesselinov, 2006), sampling rate (Deliyski et al, 2005a), and environmental noise (Deliyski, Evans, & Shaw, 2005). The authors have concluded that gender, intrasubject variability, microphone selection, environmental noise (type and level), data acquisition system, and analysis software all play a role in voice quality measurements.…”

Análise espectrográfica da voz: efeito do treinamento visual na confiabilidade da avaliação

“…Espectrograma com traçado regular, intensidade fraca, estável, com ruído ausente, sub-harmônicos ausentes e harmônicos definidos até 3 kHz de normalidade diferem entre eles. Além disso, esses valores também variam conforme os instrumentos de gravação, ruído ambiental, gênero e idade do falante (14,15) o que mostra que a qualidade do equipamento utilizado no registro das vozes, o tipo de programa e as características anatomofuncionais da laringe podem influenciar nos resultados destas medidas a curto prazo.…”

Comparability of Modern Recording Devices for Speech Analysis: Smartphone, Landline, Laptop, and Hard Disc Recorder

“…Thus, individual hardware components such as microphones [15,16] and recording hardware [11,17] may be independent sources of error, but the fidelity of the system with all the components working in synergy will ultimately reveal the error in application.…”

“…Adoption of alternative technologies in voice research has been slow, perhaps due to the requirement for high-quality recording devices needed for some types of analyses (e.g., measures of perturbation) [9,10,11,12]. Historically, only professional-grade data acquisition systems have been accepted as valid and reliable methods for collecting speech data [13].…”