Electrolaryngeal speech enhancement based on statistical voice conversion-猫眼课题宝

DOI：

10.21437/interspeech.2009-439

发表时间：

2009-09

期刊：

影响因子：

通讯作者：

中科院分区：

其他

文献类型：

作者：

Keigo Nakamura;T. Toda;H. Saruwatari;K. Shikano

摘要：本文提出了一种基于gmm的语音转换的喉切除术辅助语音系统，该系统将喉电语音（EL语音）转换为正常语音。因为有效的f0信息不能从elspeech中获得，我们已经将elspeech转换为低语。本文利用从EL语音的频谱信息估计的f0计数器将EL语音转换为正常语音。在本文中，我们从几个角度对我们的助说系统的这两类输出语音进行了实验评价。实验结果表明，转换后的正常语音优于转换后的耳语语音。检索词：电喉，喉切除术，语音转换，助听器。导读：电喉是一种医疗设备，可以使接受喉切除术的患者通过下颚皮肤给予外部声音刺激来恢复声音。尽管喉切除术患者可以恢复声音，但使用电喉（EL）语音时仍会出现一些问题。其中一个最主要的问题是语音的自然度。因为一个基本的电喉以相同的周期驱动振动器，所以听到的EL语音是单调的。这种不像人的声音不仅给说话者，也给说话者周围的人在口头交流中带来不舒服的感觉。日本已经开发出一些电喉，为视频用户提供更自然的声音[1,2]。一种名为“yourtone”的电喉试图通过使用可操作的气压传感器[1]来实现自然的声音。使用者可以通过气管造口来控制呼吸的音调，气管造口是一个连接肺部的孔。然而，双手被占用来握住电喉和空气压力传感器，因此，带有空气压力传感器的电喉限制了用户可以使用它说话的情况。另一种名为“我的声音”的电喉有一个预先定义的

Abstract This paper proposes a speaking-aid system for laryngectomeesusing GMM-based voice conversion that converts electrolaryn-geal speech (EL speech) to normal speech. Because valid F 0 informationcannotbeobtainedfromtheELspeech, wehavesofar converted the EL speech to whispering. This paper conductsthe EL speech conversion to normal speech using F 0 countersestimated from the spectral information of the EL speech. Inthis paper, we experimentally evaluate these two types of out-put speech of our speaking-aid system from several points ofview. The experimental results demonstrate that the convertednormal speech is preferred to the converted whisper. Index Terms : Electrolarynx, Laryngectomee, Voice conver-sion, Speaking-aid 1. Introduction An electrolarynx is a medical device to enable laryngectomeeswhohaveundergonelaryngectomytorehabilitatetheirvoicebygivingexternalsoundexcitationsthroughskinonthelowerjaw.Some problems are included in the speech using an electrolar-ynx (EL speech), although laryngectomees can get their voicesback. One of the most dominant problems is the naturalness ofthe EL speech. Because a basic electrolarynx drives vibratorswith the same cycle, the EL speech is heard with a monotonepitch. The non-human-like sound of the voice gives not onlythe speaker but also people around the speaker uncomfortablefeelings in their verbal communications.Some electrolarynxes have been developed in Japan to pro-videusersamorenaturalvoice[1,2]. Oneelectrolarynxnamed’yourtone’ tries to achieve the natural voice by using an op-tional air-pressure sensor [1]. Users can control the intonationwith their breath expired from the tracheostoma which is a holeconnecting to their lungs. Both hands are, however, occupiedto hold the electrolarynx and the air-pressure sensor, and there-fore, the electrolarynx with the air-pressure sensor limits thesituations in which users can speak with it. Another electro-larynx named ’myvoice’ has a pre-deﬁned

Electrolaryngeal speech enhancement based on statistical voice conversion

文献摘要

被引文献

相似文献