Preliminary results on speaker-dependent variation in the TIMIT database.

Area of Science:

• Phonetics
• Linguistics
• Speech Processing

Background:

• The TIMIT (Acoustic-Phonetic Continuous Speech) database is a widely used resource for speech research.
• Understanding speaker-dependent variations is crucial for accurate speech recognition and linguistic analysis.

Purpose of the Study:

• To present phonetic studies analyzing the TIMIT speech database.
• To detail new results on speaker-dependent variations in pronunciation.
• To explore the influence of speaker sex and dialect on phonetic features.

Main Methods:

• Database methodological approach using the TIMIT speech database.
• Acoustic-phonetic analysis of speech features.
• Examination of variations in stop release frequency, speaking rate, vowel reduction, flapping, and glottal stop usage.

Main Results:

• Identified significant speaker-dependent variations in phonetic features.
• Demonstrated effects of sex and dialect region on pronunciation characteristics.
• Confirmed TIMIT's utility for acoustic-phonetic knowledge acquisition relevant to speech classification and linguistic description.

Conclusions:

• The TIMIT database provides valuable insights into acoustic-phonetic variability in read English speech.
• Findings are relevant for improving phonetic classification, speech recognition systems, and linguistic studies of pronunciation.
• Speaker characteristics significantly influence phonetic realization, impacting speech processing and linguistic analysis.