Speech Analysis – Synthesis Based on the Ptdft for Voice Conversion

Voice conversion problem became very popular in the world. It has applications in many fields, for example in systems that make use of prerecorded speech, such as voice mailboxes or text-to-speech synthesizers based on acoustic unit concatenation. In such cases, voice modification would be a simple and efficient way to create a desired variety of voices while avoiding recording of different speakers [1]. Another field of application is old films restoration, where the aim is to reconstruct corrupted voices of old actors. One of the critical tasks in voice conversion framework is speaker parameter estimation. Here the method based on Pitch-Tracking modified Discrete Fourier Transform (PTDFT) is proposed with an idea to decompose speech signal into voiced and noise-like components.