Kadria Ezzine1*, Joseph Di Martino2 and Mondher Frikha3
1National Engineering School of Carthage, Carthage University, Tunisia
2Lorraine Laboratory of Research in Computer Science and its Applications, Lorraine University, France
3ATISP - National School of Electronics and Telecommunications of Sfax, Sfax University, Tunisia
Esophageal Speech (ES) can be used as an alternative speaking method for laryngectomes. Compared to laryngeal voice, ES is characterized by poor intelligibility and poor quality due to chaotic fundamental frequency, specific noises that resemble belching, and low intensity. These issues are alleviated by converting ES into more natural speech that is an effective way to improve speech quality and intelligibility. To accomplish this, we propose in this work a novel esophagealto- laryngeal Voice Conversion (VC) system based on a Sequence-to-Sequence (Seq2Seq) technique combined with an attention mechanism. The originality of the proposed method is that it does not require any dynamic time alignment during the training phase, which avoids erroneous mappings and significantly reduces the computing time. In addition, to preserve the identity of the target speaker, excitation and phase coefficients are estimated by querying a binary search tree in the target training space through the coefficients of the vocal tract previously predicted by the proposed Seq2Seq mapping model. In experiments, we compare our approach with baseline methods using numerous measures for objective and subjective evaluations. Perceptual tests confirmed that our proposed method behaves better and achieves better performance even in some difficult cases. In fact, it consistently exceeds conventional methods as acoustic models in terms of speech quality and intelligibility.
Ezzine K, Di Martino J, Frikha M. Esophageal to Laryngeal Voice Conversion Using a Sequence to Sequence Mapping Model Including an Attention Mechanism. Clin Surg. 2022; 7: 3534..