Pitch period detection and chaining: method and evaluation.

Area of Science:

• Speech processing
• Signal analysis
• Acoustics

Background:

• Accurate fundamental frequency (f0) estimation is crucial for speech analysis.
• Existing methods for quasi-periodic detection in speech signals have limitations.

Purpose of the Study:

• To present a new algorithm for detecting quasi-periodic parts in speech signals.
• To provide the fundamental frequency for these detected parts.
• To evaluate the performance of the proposed algorithm.

Main Methods:

• The algorithm analyzes the time-amplitude representation of speech signals.
• It identifies potential 'period-twins' within the signal.
• It verifies these twins by checking for membership in a 'period-chain'.

Main Results:

• The algorithm successfully detects quasi-periodic segments in speech.
• Fundamental frequency is accurately provided for these segments.
• Performance evaluation includes absolute and relative figures, with comparisons to other methods.

Conclusions:

• The developed algorithm offers a robust method for quasi-periodic detection in speech.
• It provides reliable fundamental frequency estimation for these segments.
• The findings suggest potential improvements over existing speech analysis techniques.