โ† Alexander Wang

MARingBA: Music-Adaptive Ringtones for Blended Audio Notification Delivery

Alexander Wang, Yi Fei Cheng, David Lindlbauer

CHI 2024 โ€” ACM Conference on Human Factors in Computing Systems

๐Ÿ† Best Paper Honorable Mention (top 5%)

MARingBA preview

We explore a design space for adapting audio notifications and present MARingBA, a system that adapts audio notifications to be seamlessly blended into music with controllable levels of saliency. In a study with 12 users, we demonstrate that our technique successfully modulates user response time as well as their perceived level of noticeability, immediacy, and distraction.

Audio Examples

Hotline Bling โ€” Drake

iPhone ringtone (Classic)

What Is Love โ€” Haddaway

Google Hangouts notification

Happy โ€” Pharrell Williams

Discord notification

Somebody That I Used to Know โ€” Gotye

iPhone ringtone

Drake meme: rejecting jarring interruptions, preferring harmonically blended notifications

Modulating Saliency

Today's audio notifications are largely one-size-fits-all. In contrast, visual notifications span a range of saliency, from subtle indicators like Slack's red dot, to banners, to full pop-up alerts. MARingBA brings this idea to audio by introducing multiple ways to deliver notifications, allowing less important alerts to blend into the music instead of causing abrupt interruptions.

Our Approach

MARingBA takes the music you're already listening to and an incoming notification, analyzes the music, and reshapes the notification to blend in. Here is each step, with audio.

1

Inputs

The music the listener is playing, and an incoming notification โ€” here, the Google Hangouts ringtone.

Music
Notification
2

Analyze musical features

We extract the music's beat, key, and structure โ€” the scaffolding used to reshape and place the notification in the steps below.

Music information retrieval pipeline: downbeat, key and tempo, stereo balance
3

Beat matching

Overlaid naively, the ringtone fights the groove. Aligning it to the beat lets it lock in with the music.

Notification onsets misaligned, then time-stretched to match the tempo
Without beat matching
With beat matching
4

Key matching

Beat-aligned but still clashing harmonically. Pitch-shifting the ringtone into the song's key makes it consonant.

Notes out of key, then pitched up to match the song's key
Without key matching
With key matching
5

Schedule to song structure

We place the notification at a natural moment in the song's structure rather than cutting in arbitrarily.

Notification received, delayed to the start of the next section
Scheduled
6

Adjust timbre

Finally, we shape the ringtone's timbre so it sits inside the mix โ€” controlling just how much it stands out.

Timbre adjusted

Results

Our evaluation showed that MARingBA successfully modulated the saliency of audio notifications. Lower-priority notifications were noticed later and perceived as less distracting without compromising users' confidence that they had heard every notification. Participants also preferred our integrated adaptation over conventional notification delivery.

Box plots of participants' subjective ratings across Low, Medium, High, and Standard delivery methods
Participants' subjective ratings across delivery methods on a scale from 1 (low) to 7 (high), and ranking of preference from 1st (most preferred) to 4th (least preferred). Horizontal bars indicate statistically significant differences (p < .05) between conditions.

Citation

@inproceedings{wang2024maringba,
  title     = {MARingBA: Music-Adaptive Ringtones for Blended Audio Notification Delivery},
  author    = {Wang, Alexander and Cheng, Yi Fei and Lindlbauer, David},
  booktitle = {Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems},
  year      = {2024},
  doi       = {10.1145/3613904.3642376}
}