A Danish open-set speech corpus for competing-speech studies.

Area of Science:

• Audiology
• Speech processing
• Psychoacoustics

Background:

• Speech-on-speech masking studies often use limited closed-set materials, leading to unrealistic low speech recognition thresholds (SRTs).
• Existing methods do not accurately reflect natural communication scenarios with competing speech.
• A need exists for more ecologically valid speech materials to assess auditory performance.

Purpose of the Study:

• To develop and validate a novel open-set, low-context, multi-talker Danish speech corpus.
• To create speech materials that yield more realistic SRTs in spatially separated masking conditions.
• To assess the corpus's sensitivity to individual differences and test list consistency.

Main Methods:

• Created three sets of 268 unique Danish sentences, recorded by professional female talkers.
• Selected 200 equally intelligible sentences per talker and organized them into 10 test lists.
• Measured intelligibility in speech-shaped noise and assessed test list homogeneity with frontal target and ±50° azimuth maskers.

Main Results:

• Obtained overall SRTs of +1.3 dB for normal-hearing listeners and +6.3 dB for hearing-impaired listeners.
• The corpus demonstrated high sensitivity to inter-individual differences in speech recognition.
• Consistent results were observed across different test lists, indicating reliability.

Conclusions:

• The developed open-set Danish speech corpus provides a more ecologically valid measure of speech recognition in noise.
• The corpus is suitable for research on speech-on-speech masking and aids in understanding real-world listening challenges.
• The publicly available corpus facilitates further research in audiology and speech perception.