Claim Missing Document
Check
Articles

Found 11 Documents
Search

PATSE: PyTorch-Native Polar Encoding for Runtime Stability in Audio Processing Nuk Ghurroh Setyoningrum; Ema Utami; Kusrini Kusrini; Ferry Wahyu Wibowo
Jurnal RESTI (Rekayasa Sistem dan Teknologi Informasi) Vol 10 No 5 (2026): October 2026
Publisher : Ikatan Ahli Informatika Indonesia (IAII)

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.29207/resti.v10i5.7776

Abstract

Efficient and stable audio feature extraction is essential for real-time AI systems, yet conventional spectral representations often incur preprocessing overhead and framework-conversion dependencies. This study proposes PATSE (Polarized Amplitude Time Spiral Encoding), a PyTorch-native spatial–temporal encoding framework that converts one-dimensional audio into CNN-compatible two-dimensional polar representations while preserving temporal–amplitude information. Evaluated on 100 infant cry samples, PATSE achieved competitive CPU runtime (5.37 ms), improved latency stability over Mel-Spectrogram, and 91% classification accuracy with a Macro-F1 score of 0.85 using a lightweight CNN. These results suggest that PATSE provides an effective balance between computational efficiency, framework integration, and discriminative representation in the evaluated CPU-based infant cry setting, while broader validation is reserved for future work.