TL;DR
This paper addresses the issue of outdated recordings in a widely used speech perception in noise test (RSPIN) by re-recording it with modern techniques and new speakers. The resulting R2SPIN test offers improved audio quality and broader applicability, validated through both objective and subjective analyses.
What Problem Does This Paper Address?
Understanding how people hear speech in noisy environments is crucial for both clinical diagnosis of hearing impairments and fundamental research into auditory perception. Speech perception in noise (SPIN) tests are essential tools for this, but many, including the popular Revised SPIN (RSPIN), rely on recordings made decades ago. These older recordings suffer from limitations like limited bandwidth, tape hiss, and a single male speaker with an American accent, which can hinder their effectiveness and applicability in diverse research and clinical settings. This paper tackles the problem of updating these legacy recordings to ensure their continued relevance and usability.
Methods
The authors re-recorded the RSPIN test, creating R2SPIN, by employing several key methodological improvements:
- New Speakers and Accent: The re-recording features both a male and a female speaker, both using British Received Pronunciation. This broadens the accent base compared to the original American English speaker.
- Modern Recording Techniques: High-quality audio was captured using a Neumann TLM 193 microphone in a quiet room, recorded into a Sadie digital audio workstation at a 48kHz sample rate and 32-bit depth, saved as uncompressed WAV files.
- Quality Verification: A rigorous process involving five experienced listeners was used to validate the recordings, checking for mispronunciations, delivery issues, and recording artifacts. Problematic sentences were re-recorded.
- Objective Intelligibility Analysis: The "glimpse proportion" (GP) metric was used to objectively assess the audibility of speech in noise. This metric quantifies time-frequency regions of speech that survive masking, allowing for comparisons between the original and new recordings.
- Subjective Intelligibility Analysis: A pilot study with 6 listeners was conducted to determine an appropriate signal-to-noise ratio (SNR) for the new recordings. This was followed by a larger subjective evaluation with 16 participants (11 male, 5 female) to assess word recognition rates.
- Statistical Analysis: Generalized estimating equations (GEE) were used for the subjective data analysis, a method suitable for dichotomous outcome variables in repeated measures designs.
Datasets
The paper utilizes the original RSPIN sentence lists. The sentences themselves were originally selected from "The teacher's word book of 30,000 words" [14], ensuring word familiarity. The re-recording process involved re-recording all 400 sentences from the original RSPIN test.
Key Insights and Ideas
- The Importance of Modernization: Legacy audio recordings in speech research, even for well-established tests, can become obsolete due to technological advancements and changing linguistic norms. Re-recording is a viable and necessary approach to maintain the utility of such tools.
- Speaker and Accent Diversity Matters: Including a female speaker and a different accent (British Received Pronunciation) not only makes the test more balanced but also increases its applicability to a wider range of research and clinical populations.
- Objective Metrics Complement Subjective Evaluation: The use of the glimpse proportion (GP) provided an objective measure of how well speech segments were preserved in noise, offering insights that complemented the subjective word recognition data.
- Predictability Remains Key: The R2SPIN test, like its predecessor, continues to effectively evaluate the interplay between top-down (predictability) and bottom-up (acoustic signal) processing in speech perception.
Key Results
- Objective Analysis: The original speaker's recordings showed a higher mean glimpse proportion (GP) for both keywords and preceding speech, suggesting less energetic masking compared to the new recordings. However, the new male and female speakers' recordings were significantly different from the original speaker in terms of keyword GP.
- Subjective Analysis: The R2SPIN test successfully maintained the characteristic RSPIN improvement of 30-40% between low and high predictability sentences. The female speaker generally achieved higher word recognition rates (WRR) than the male speaker, particularly for low predictability sentences.
- Statistical Significance: Predictability, speaker gender, and keyword GP were found to be significant factors influencing word recognition. The authors recommend using a larger SNR for the female speaker compared to the male speaker when using the same masker.
Limitations and Future Work
- Normative Data Specificity: The normative data presented in this paper is specific to the exact timing relationships between the speech and noise used in this particular recording. Changes in timing or masker will likely alter results.
- Accent Limitation: While the inclusion of a female speaker and British Received Pronunciation broadens the scope, it still represents only one additional accent, and further diversity could be beneficial.
- Future Work: The authors mention that binaural recordings of the stimuli are planned for subsequent work, which will further expand the application of the R2SPIN test to the evaluation of spatial hearing in noise. They also suggest that variations between lists could be managed by adjusting sentence alignment relative to the noise to achieve more uniform GPs.
Summary updated