Investigating the Effect of Music and Lyrics on Spoken-Word Recognition

Odette Scharenborg, Martha A. LarsonarXiv 2018 · arXiv

Introduction

Background noise is a ubiquitous challenge in social environments like bars and restaurants. While it is well-established that noise hinders conversation, the specific properties of background music that contribute to this interference are less understood. This study investigates how music—specifically the presence of lyrics and rhythmic complexity—interferes with spoken-word recognition.

The researchers distinguish between two primary types of interference:

  1. Energetic Masking: Occurs when physical sounds overlap and "drown out" speech frequencies.
  2. Informational Masking: Occurs when the brain struggles to separate target speech from background content that contains meaningful information (like lyrics) or complex patterns competing for cognitive resources.

This study specifically examines whether music with lyrics masks speech more than instrumental music and how rhythmic complexity influences this effect.

Methods and Datasets

The researchers conducted a word identification experiment with 20 native Dutch listeners to quantify these interference effects. Participants were presented with 150 Dutch CVC (Consonant-Vowel-Consonant) words, which were carefully controlled for linguistic factors such as frequency of use and "neighborhood density" (the number of phonologically similar words). These target words were embedded within background music tracks sourced from a curated Amsterdam restaurant playlist, reflecting real-world social listening conditions.

Experimental Variables

  • Lyrics vs. No Lyrics: To isolate the effect of lyrics, segments from three songs were used in two versions: one with singing and an instrumental version with a nearly identical backing track.
  • Signal-to-Noise Ratio (SNR): Words were presented at three difficulty levels: +15 dB (quiet music), +5 dB (moderate), and 0 dB (loud music, equal to speech volume).
  • Musical Complexity: Three tracks were selected based on their rhythmic density:
  • E go betta by Dele Sosimi (Complex Afrobeat/Funk, 110 bpm)
    1. Purple by Crustation (Simple Trip Hop, 76 bpm)
    2. Stay away from music by Stephen Colebrook (Simple rhythm, 118 bpm)

Key Results and Insights

The study revealed a clear "lyrics penalty" in spoken-word recognition, particularly as listening conditions became more adverse.

Proportion of correct word identifications across different Signal-to-Noise Ratios (SNR) for songs with and without lyrics.

The Impact of Lyrics

At the easiest SNR (+15 dB), the presence of lyrics had a negligible effect. However, at +5 dB and 0 dB, recognition accuracy dropped significantly when lyrics were present. This interaction between SNR and lyrics was statistically significant (p < .001) at 0 dB SNR. This suggests that as the acoustic signal degrades, the brain's attempt to process the linguistic content of lyrics interferes with decoding target words—a clear instance of informational masking.

Rhythm and Complexity

Interestingly, rhythmic complexity proved more disruptive than tempo. Song 1, the complex Afrobeat track, was the most effective masker despite having a moderate tempo. The authors attribute this to its dense percussion, which fills the acoustic "glimpses" or gaps that listeners typically use to catch speech fragments.

Waveforms and spectrograms illustrating the acoustic density of the musical stimuli, which affects the availability of 'glimpses' for speech recognition.

Language Familiarity

The researchers noted that lyrics in simpler songs were in standard English, familiar to Dutch speakers, while the complex song featured West African Coastal English. They hypothesize that more familiar languages might cause greater informational masking as the brain "tunes in" more automatically.

Discussion and Future Work

The findings highlight that musical complexity and lyrics are independent factors in speech interference. However, the study's use of isolated words may underestimate the cognitive load in natural conversations where context is used to fill gaps.

Future research should aim to quantify the relationship between "glimpses" (time-frequency regions not masked by noise) and musical complexity to better predict disruptive tracks. Additionally, investigating the impact on older listeners or those with hearing impairments is crucial, as they likely face even greater challenges in these complex lyrical soundscapes.

Summary updated

BibTeX
@misc{scharenborg2018investigating,
  author = {Odette Scharenborg and Martha A. Larson},
  title = {{Investigating the Effect of Music and Lyrics on Spoken-Word Recognition}},
  year = {2018},
  eprint = {1803.05058},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1803.05058},
}
Figure