Artificial ASMR: A Cyber-Psychological Approach

Zexin Fang,Bin Han,C. Clark Cao,Hans. D. Schotten

2023-07-05

Abstract:The popularity of Autonomous Sensory Meridian Response (ASMR) has skyrockted over the past decade, but scientific studies on what exactly triggered ASMR effect remain few and immature, one most commonly acknowledged trigger is that ASMR clips typically provide rich semantic information. With our attention caught by the common acoustic patterns in ASMR audios, we investigate the correlation between the cyclic features of audio signals and their effectiveness in triggering ASMR effects. A cyber-psychological approach that combines signal processing, artificial intelligence, and experimental psychology is taken, with which we are able to quantize ASMR-related acoustic features, and therewith synthesize ASMR clips with random cyclic patterns but not delivering identifiably scenarios to the audience, which were proven to be effective in triggering ASMR effects.

Audio and Speech Processing,Artificial Intelligence,Multimedia,Sound,Signal Processing

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to explore and verify the correlation between the autonomous sensory meridian response (ASMR) triggering effect and the cyclic characteristics of audio signals. Specifically, existing research has identified some semantic elements that can trigger the ASMR effect, but there is insufficient understanding of the acoustic characteristics of audio triggers. By combining methods in signal processing, artificial intelligence, and experimental psychology, this paper quantifies the acoustic characteristics related to ASMR and synthesizes ASMR audio clips with random cyclic patterns, which can still effectively trigger the ASMR effect without providing an explicit scenario.

### Main research content:
1. **Problem background**:
- The ASMR phenomenon has become very popular in the past decade, but scientific research on its triggering mechanism is still scarce and immature.
- Existing research shows that low - frequency sounds can usually effectively trigger ASMR, but their repetitive characteristics have not been captured yet.

2. **Research methods**:
- **Data collection**: Real - sound recordings from four different sources were obtained from non - commercial online resources, including breathing, mixing soft cream, spraying, and keyboard clicking.
- **Feature extraction**: Short - time Fourier transform (STFT) and cyclic spectral analysis were used to extract the acoustic features of the audio.
- **Generative model**: A deep convolutional generative adversarial network (DCGAN) was used to generate ASMR audio clips with random cyclic patterns.
- **Psychological survey**: A psychological survey was designed to evaluate the effectiveness of the recorded and synthesized ASMR audio in triggering the human ASMR effect.

3. **Main findings**:
- **Correlation between acoustic features and ASMR effect**: Through analysis and experiments, it was proven that there is a significant correlation between the cyclic characteristics of audio signals and their ability to trigger the ASMR effect.
- **Effectiveness of synthesized audio**: The generated random ASMR audio clips can still effectively trigger the ASMR effect without providing an explicit scenario, although they are slightly less effective in triggering a strong ASMR experience than the recorded audio.

### Formula analysis:
- **Spectral correlation density (SCD)**:
\[
S_{XX}(f, \alpha)=\mathbb{E}\left\{X(f + \frac{\alpha}{2})X^*(f-\frac{\alpha}{2})\right\}
\]
where \(\alpha\) is the cyclic frequency and \(X(f)\) is the complex spectrum of the signal \(x\) at frequency \(f\).

- **Cyclic coherence function (CCF)**:
\[
C_{XX}(f, \alpha)=\frac{S_{XX}(f, \alpha)}{\sqrt{\mathbb{E}\left\{|X(f + \frac{\alpha}{2})|^2\right\}\mathbb{E}\left\{|X(f-\frac{\alpha}{2})|^2\right\}}}
\]

- **Discrete Fourier transform (DFT)**:
\[
X(f)=\frac{1}{N}\sum_{k = 0}^{N-1}\left[\sum_{n = 0}^{N-1}x(n)e^{-j2\pi nk/N}\delta(2\pi f - 2\pi k/N)\right]
\]

- **Feature definition**:
- \(\Phi_1(x)=\text{Mean}(S_{XX})\)
- \(\Phi_2(x)=\text{Var}(S_{XX})\)
- \(\Phi_3(x)=G(S_{XX})\)
- \(\Phi_4(x)=\text{Mean}(S_{XX}^{\text{max}})\)
- \(\Phi_5(x)=\text{Var}(S_{XX}^{\text{max}})\)