FAU speech-DPOAE data set for auditory attention

Dataset Description: Speech-DPOAE Experiment

If you use this dataset, please cite our paper: doi: https://doi.org/10.1101/2025.08.15.670505

This dataset accompanies an experiment measuring distortion product otoacoustic emissions (DPOAEs) evoked by human running speech (speech-DPOAEs).  
Stimuli were designed to follow harmonic overtones n and m (n < m) of the speech signal’s fundamental frequency (f₀), with the expected lower-sideband DPOAE arising at harmonic 2n–m.

The experiment comprised:

1. Single-speaker measurements – either male or female voice presented in isolation (two trials per participant).  
2. Competing-speaker measurements – both voices presented simultaneously, with participants instructed to attend to one voice while ignoring the other (18 trials per participant).

Speech Signals

Two audiobooks were synthesized using a text-to-speech engine (ElevenLabs, U.S.A.) [1], with the voice “Matilda” for the female speaker and the voice “Brian” for the male speaker.

Source texts (in german language):  

• Eine Frau erlebt die Polarnacht [4]   
• Darum [2]  

Each text was synthesized with both voices, yielding:  `Darum_female`, `Darum_male`, `Polarnacht_female`, and `Polarnacht_male`.

Format and organization:   

• Audiobooks are provided chapter-wise for single-speaker presentation. 
• Mixed speech files for competing-speaker measurements are provided in `MixedBook_` folders.  
• Each segment is ~2 minutes long.  
• For single-speaker trials: 20 chapters per voice in total (2 chapters intended for single-speaker measurements, 18 chapters intended for subsequent mixing for competing-speaker measurements).
• For competing-speaker trials: 18 chapters per audiobook presented simultaneously.

Stimuli for Speech-DPOAEs

Speech-DPOAE stimuli were computed using our recently developed method [5], pairing resolved resp. unresolved harmonic overtones n and m.  The lower-sideband cubic DPOAE arises at harmonic 2n–m.  

The table below lists n and m for the four stimulus types $F_{res}$ , $F_{unres}$ , $M_{res}$ , $M_{unres}$ :    

Presentation protocol:

• Single-speaker:  two stimulus pairs (resolved + unresolved, matching the presented voice).  
• Competing-speaker:  four stimulus pairs (resolved + unresolved for both voices).

File organization:
Stimuli are stored in `Stimuli/`, sorted by book and voice.  

• `SingleTwobandStimuli_xxx` – single-speaker two-band stimuli. 
• `MixedTwobandStimuli_xxx` – competing-speaker two-band stimuli.
• `Book_voice` - unmixed stimuli for chapters 1 to 18 of competing-speaker trials and DPOAE stimuli needed for evaluation.

Recordings

Stimuli were presented via an extended-bandwidth otoacoustic measurement system (ER10X, Etymotics, U.S.A.).  Audiobooks were delivered to the right ear, while stimulus presentation and DPOAE recordings were made in the left ear.  Recordings contain both the presented stimuli and the evoked DPOAEs.