Sarir
Small Data Generative AI for Culturally Specific Artistic Creation: A Case Study on the Sonic Space of sarīr
ISEA 2026 Anonymous Submission
1. Results from Open-Source Text-to-Audio Models
| AudioLDM2 | Stable Audio Open | AudioGen |
|---|---|---|
| prompt: "Arabic calligraphy sarir." | ||
| prompt: "Persian calligraphy sarir." | ||
| prompt: "The creaking sound of a reed pen during Arabic calligraphy writing." | ||
| prompt: "The creaking sound of a reed pen during Persian calligraphy writing." | ||
| prompt: "The sound of a reed pen writing in Arabic calligraphy." | ||
| prompt: "The sound of a reed pen writing in Persian calligraphy." | ||
| prompt: "صدای قلم نی هنگام خوشنویسی" | ||
2. Audio and Spectrogram Samples from the sarīr Dataset
Example 1
Example 2
Example 3
Example 4
Example 5
Example 6
3. Results from RAVE Model Training
3.1 Generated Samples from RAVE v2 Configuration+Prior (our preferred results)
Example 1
Example 2
Example 3
Example 4
Example 5
Example 6
3.2 Generated Samples from RAVE v2 Configuration
Example 1
Example 2
Example 3
Example 4
Example 5
Example 6
3.3 Generated Samples from RAVE v2-small Configuration
Example 1
Example 2
Example 3
Example 4
Example 5
Example 6
3.4 Generated Samples from RAVE v1 Configuration
Example 1
Example 2
Example 3
Example 4
Example 5
Example 6
4. Results from StyleGAN2 Model Training
4.1 Generated Spectrograms → Audio Reconstruction → RAVE Correction
Example 1
Generated Spectrogram
Reconstructed Audio
RAVE-Corrected Audio
Reconstructed Audio
RAVE-Corrected Audio
Example 2
Generated Spectrogram
Reconstructed Audio
RAVE-Corrected Audio
Reconstructed Audio
RAVE-Corrected Audio
Example 3
Generated Spectrogram
Reconstructed Audio
RAVE-Corrected Audio
Reconstructed Audio
RAVE-Corrected Audio
Example 4
Generated Spectrogram
Reconstructed Audio
RAVE-Corrected Audio
Reconstructed Audio
RAVE-Corrected Audio
Example 5
Generated Spectrogram
Reconstructed Audio
RAVE-Corrected Audio
Reconstructed Audio
RAVE-Corrected Audio
Example 6
Generated Spectrogram
Reconstructed Audio
RAVE-Corrected Audio
Reconstructed Audio
RAVE-Corrected Audio
4.2 StyleGAN2 Latent Space Interpolation
4.3 Generated Samples Based on Latent Space Interpolation Using Concatenative algorithm
Example 1
Example 2
Example 3
Example 4
Example 5
