This paper proposes a new deep speech transformation system for converting speech with dysarthric speech into intelligible, clear speech. By combining CycleGAN for MFCC feature mapping, BLSTM for speech recognition, and Tacotron for speech synthesis, the system effectively converts degraded speech into naturally sounding speech. On testing with UASpeech, it reflects significant improvements in intelligibility, in terms of Word Error Rate (WER), over state-of-the-art approaches. It sees its potential for speech improvement for individuals with dysarthria, offering a personalized and effective remedy.

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Cycle-GAN Dysarthric-to-Normal Voice Conversion

  • Salma Chlaikhy,
  • Adil Chakhtouna,
  • Abdellah Adib

摘要

This paper proposes a new deep speech transformation system for converting speech with dysarthric speech into intelligible, clear speech. By combining CycleGAN for MFCC feature mapping, BLSTM for speech recognition, and Tacotron for speech synthesis, the system effectively converts degraded speech into naturally sounding speech. On testing with UASpeech, it reflects significant improvements in intelligibility, in terms of Word Error Rate (WER), over state-of-the-art approaches. It sees its potential for speech improvement for individuals with dysarthria, offering a personalized and effective remedy.