Impact of perturbation and value of fundamental frequency on the sound quality of electrolaryngeal speech.

Electrolarynx is a device used to replace the glottal source in cases of irreversible loss of natural voice. However, the sound is considered robotic, which in the literature is related to the absence of cycle-by-cycle perturbation. Furthermore, low values of fundamental frequency (f0) represent an additional challenge, especially for women. Thus, was developed the present research aiming to verify the impact of perturbation and f0 value on the sound quality of electrolaryngeal speech. In view of this, three studies were developed: Study 1, cross-sectional, to obtain human perturbation and acoustic measurements of source (f0) and filter (first to fourth formants - F1- F4 and bandwidths - B1- B4) necessary for the synthesis experiment. For this, was analyzed recordings of sustained vowels / a / emitted by 162 laryngeal speakers, 78 women and 84 men, young (18-44 years old), middle-aged (45-59 years old) and elderly (60-80 years old). Study 2, quasi-experimental, with synthesis of 24 vowels / a / sustained to verify, through blind auditory-perceptual judgment, the degree of naturalness provided by two mathematical patterns of perturbation of f0 - (I) random with uniform distribution and (II) second order plus randomness - having as control the natural perturbation extracted from recorded human voices and as a placebo the absence of perturbation. Study 3, quasi-experimental, with blind comparative analysis of electrolaryngeal emissions of 10 total laryngectomized patients (1 woman and 9 men) with Conventional equipment (available in the national market) and Modified (with random perturbation and f0 referring to the sample mean as a function of sex and age group) performed by the patients themselves, in addition to auditory-perceptual judgment performed by Speech-Language Pathologists. As a result, the comparative analysis of the vowels synthesized from the means of f0, F1, F2, F3, F4, B1, B2, B3, B4 of women and men of different age groups, with different patterns of perturbation, indicated that the second-order model provided naturalness degree similar to human perturbation. Despite the random pattern not having the same performance, it showed proximity to the natural perturbation in the voice synthesized with of elderly parameters. Due to the characteristic of the Eletrolaringe circuits, it was not possible to insert the second order pattern. Therefore, the equipment was adjusted with random perturbation. In view of the sonorities, the laryngectomized woman preferred the modified; about the men, 4 preferred the modified, 1 the conventional and 4 said there was no difference. In the auditory-perceptual judgment performed by Speech Pathologist, in 4 the modified was better, among them for women, in 4 the conventional was better and in 2 there was no difference. Given the findings, it was concluded that there were no relevant improvements in electrolaryngeal sound quality, possibly due to other aberrant sound characteristics of the equipment.

电子喉（Electrolarynx）是一类用于在患者出现不可逆自然嗓音丧失时替代声门声源的设备。然而其发出的声音常被认为具有机器人化音质，现有研究表明该问题与缺乏逐周期语音扰动有关。此外，基频（fundamental frequency, f0）取值偏低也是一项额外挑战，尤其对女性群体而言。因此，本研究旨在探究扰动与基频取值对电子喉语音音质的影响。
为此，本研究开展了三项子研究：
研究1为横断面研究，旨在获取合成实验所需的人类语音扰动及声源（基频f0）、声道滤波器相关参数，包括第一至第四共振峰（formants，F1-F4）与带宽（bandwidths，B1-B4）的声学测量数据。本研究分析了162名喉部发声者（78名女性、84名男性，涵盖青年18-44岁、中年45-59岁及老年60-80岁三个年龄段）发出的持续元音/a/的录音。
研究2为准实验研究，合成了24个持续元音/a/，旨在通过盲法听觉-感知评估，验证两种f0扰动数学模型——（I）均匀分布随机扰动、（II）二阶扰动加随机扰动——的自然度表现，以从人类语音录音中提取的自然扰动为参照，以无扰动作为安慰剂对照。
研究3为准实验研究，对10名全喉切除患者（1名女性、9名男性）分别使用常规市售电子喉设备与改良型设备（搭载随机扰动，且基频f0根据性别与年龄组的样本均值设置）发出的语音进行盲法对比分析，语音录制由患者自行完成，同时由言语语言病理学家（Speech-Language Pathologists）开展听觉-感知评估。
结果显示，针对不同年龄段男女的f0、F1、F2、F3、F4、B1、B2、B3、B4均值合成的元音，结合不同扰动模式开展的对比分析表明，二阶扰动模型可提供与人类自然扰动相近的自然度。尽管随机扰动模型的表现未达同等水平，但在采用老年群体参数合成的语音中，其与自然扰动的效果较为接近。受电子喉电路特性限制，无法嵌入二阶扰动模型，因此最终设备采用了随机扰动方案。
针对语音听感的患者调研显示，该女性全喉切除患者更偏好改良型设备；男性患者中，4人偏好改良型设备、1人偏好常规设备，另有4人认为二者无差异。在言语语言病理学家开展的听觉-感知评估中，4例样本显示改良型设备效果更优（其中1例为女性患者），4例显示常规设备效果更优，剩余2例认为二者无差异。
基于上述研究结果，本研究得出结论：本次设备调整未对电子喉的语音音质带来显著提升，其原因可能在于设备本身存在其他异常声学特征。