An edit-distance model for the approximate matching of timed strings

Area of Science:

• Computer Science
• Computational Linguistics
• Speech Processing

Background:

• Approximate string matching is crucial for analyzing sequential data.
• Existing edit-distance models often do not account for temporal information.
• Accurate classification of speech recognition errors requires robust sequence alignment.

Purpose of the Study:

• To present a novel edit-distance model for timed strings.
• To extend weighted string-edit distance with time-dependent costs and operations.
• To demonstrate the model's utility in classifying phone-recognition errors.

Main Methods:

• Developed an edit-distance model incorporating timed edit operations.
• Introduced time-dependent costs for edit operations.
• Focused on timed null symbols for insertions and deletions.
• Applied the model to the TIMIT speech database for error classification.

Main Results:

• The proposed model effectively handles approximate matching of timed strings.
• Time-dependent costs improve the accuracy of sequence comparison.
• The model successfully classified phone-recognition errors in the TIMIT dataset.

Conclusions:

• The novel timed edit-distance model offers a powerful tool for sequential data analysis.
• This approach enhances the accuracy of speech recognition error classification.
• The model's flexibility makes it applicable to various time-series data processing tasks.