Understanding the Lombard Effect for Mandarin: Relation Between Speech Recognition Thresholds and Acoustic Parameters

Area of Science:

• Acoustic Phonetics
• Speech Perception
• Psychoacoustics

Background:

• The Lombard effect, a change in speech production when speaking in noise, is crucial for maintaining communication.
• Understanding the acoustic correlates of the Lombard effect in different languages is essential for speech technology and audiology.
• Mandarin Chinese, a tonal language, presents unique characteristics for studying speech production modifications.

Purpose of the Study:

• To quantify the Lombard effect in native Mandarin Chinese speakers.
• To investigate the influence of speaker gender and acoustic parameters on speech intelligibility in noise.
• To identify acoustic features contributing to intelligibility gains in Lombard speech.

Main Methods:

• Utilized the Mandarin Chinese Matrix sentence test to measure speech recognition thresholds (SRTs) in stationary noise.
• Recorded plain and Lombard speech from 11 native Mandarin speakers.
• Analyzed acoustic features including fundamental frequency (F0), formant frequencies (F1, F2), segment duration/rate, alpha ratio, and speech-weighted signal-to-noise ratio (swSNR).

Main Results:

• Mandarin Lombard speech was significantly more intelligible than plain speech for both genders.
• Female speakers demonstrated a larger Lombard gain in intelligibility compared to male speakers.
• Speech-weighted signal-to-noise ratio (swSNR) and alpha ratio were strongly correlated with Lombard gain, indicating the importance of high-frequency energy.

Conclusions:

• The Lombard effect enhances speech intelligibility in Mandarin Chinese, with a notable gender difference.
• Increased high-frequency energy, reflected in swSNR and alpha ratio, is a key factor driving Lombard gain.
• While F0 and formant frequencies show some correlation, their role in Lombard gain appears secondary to spectral energy changes.