Estimation of voice-onset time in continuous speech using temporal measures

Area of Science:

• Speech processing
• Acoustic phonetics
• Computational linguistics

Background:

• Accurate estimation of voice-onset time (VOT) is crucial for phonetic analysis and speech technology.
• Existing methods often rely on manual transcriptions or require classifier training, limiting their scalability and applicability.
• There is a need for automatic, transcription-free methods for VOT estimation.

Purpose of the Study:

• To propose an automatic acoustic-phonetic method for estimating voice-onset time (VOT) of stop consonants.
• To develop a method that does not require manual utterance transcriptions or classifier training.
• To validate the proposed method's performance against state-of-the-art techniques.

Main Methods:

• Utilizes the plosion index for automatic detection of stop burst onsets.
• Employs epochal information and a maximum weighted inner product measure for detecting voicing onset post-burst.
• Applies the method to the TIMIT database and CMU Arctic corpora for comprehensive validation.

Main Results:

• The proposed automatic method successfully estimates voice-onset time for stops.
• The method demonstrates robustness across different speech corpora.
• Performance is comparable to established state-of-the-art techniques in VOT estimation.

Conclusions:

• The developed acoustic-phonetic method offers an effective, automatic, and transcription-free approach to VOT estimation.
• This method has the potential to advance phonetic research and improve speech processing applications.
• The plosion index and maximum weighted inner product provide reliable features for automatic phonetic analysis.