Bio-inspired evolutionary oral tract shape modeling for physical modeling vocal synthesis

Area of Science:

• Acoustics and Speech Synthesis
• Computational Intelligence
• Bio-inspired Computing

Background:

• Physical modeling using digital waveguide mesh (DWM) models offers organic sound in music synthesis.
• Current speech synthesis relies on accurate oral tract measurements, often from MRI, which has significant drawbacks.
• MRI limitations include long exposure times, acoustic noise, and subject positioning constraints.

Purpose of the Study:

• To apply physical modeling inspired by music synthesis to speech synthesis.
• To develop an alternative method for determining human oral tract shapes for speech synthesis.
• To overcome the limitations associated with MRI-derived measurements.

Main Methods:

• A physical modeling mesh model of the human oral tract was created.
• A bio-inspired computing technique simulating evolution was used to derive oral tract shapes.
• Acoustic and excitation data, along with electrolaryngograph recordings, were utilized.

Main Results:

• The evolutionary technique successfully produced appropriate oral tract shapes for open vowels.
• The method generated less appropriate shapes for close vowels.
• High-quality synthetic speech was achieved for vowel sounds and diphthongs using the evolved shapes.

Conclusions:

• Bio-inspired computing offers a viable alternative to MRI for obtaining oral tract shapes in speech synthesis.
• The developed method shows promise for synthesizing natural-sounding vowels and dynamic speech sounds.
• Further research is needed to improve the synthesis of close vowels.