Objective speech quality assessment and the RPE-LTP coding algorithm in different noise and language conditions

Area of Science:

• Signal Processing
• Speech Technology
• Acoustics

Background:

• Reliable signal processing algorithms for speech coding and synthesis necessitate performance criteria.
• While coding efficiency and computational demands are important, speech quality is a crucial final performance measure.
• The regular-pulse excitation with long-term prediction (RPE-LTP) algorithm is the standard for the European Digital Mobile Radio system.

Purpose of the Study:

• To evaluate three objective speech quality measures for assessing voice coding performance.
• To determine if objective measures can quantify quality changes across different languages and noise conditions.
• To validate objective measures against known subjective performance levels for the RPE-LTP algorithm.

Main Methods:

• Assessed objective speech quality measures for American English (noise-free).
• Analyzed speech quality variations with three additive background noise sources.
• Evaluated noise-free performance across seven languages: English, Japanese, Finnish, German, Hindi, Spanish, and French.

Main Results:

• Objective measures showed potential in quantifying quality variations due to noise and language.
• The RPE-LTP algorithm's performance was assessed under various conditions.
• Correlation between objective and subjective assessments was implied for algorithm tuning.

Conclusions:

• Objective speech quality measures cannot replace subjective testing but are valuable tools.
• These measures can effectively assess performance changes and identify areas for algorithm improvement.
• Objective measures can augment subjective tests for voice coding in diverse environments and languages.