[Online]A Mel-Scale STFT-Domain TinyUNet for Speech Enhancement on an ESP32-S3 Microcontroller

A Mel-Scale STFT-Domain TinyUNet for Speech Enhancement on an ESP32-S3 Microcontroller
ID:83 Submission ID:69 View Protection:ATTENDEE Updated Time:2026-10-04 23:34:43 Hits:208 Online

Start Time:2026-10-13 11:15 (Asia/Ho_Chi_Minh)

Duration:15min

Session:[S2] Track 2: IoT and applications » [S2-3] Track 2: IoT and applications

Video No Permission Presentation File

Tips: The file permissions under this presentation are only for participants. You have not logged in yet and cannot view it temporarily.

Abstract
Deploying speech enhancement directly on microcontroller units provides an energy-efficient, privacy-preserving front-end for voice-enabled edge computing. However, conventional deep neural architectures impose computational loads and memory footprints that exceed the strict resource boundaries of commodity microcontrollers. This paper introduces a Mel-scale STFT-domain TinyUNet comprising 13,651 parameters, optimized for on-device speech enhancement on an ESP32-S3 microcontroller operating at 240 MHz. Rather than processing full-resolution linear spectrograms, the architecture utilizes a 64-channel Mel filterbank representation, effectively halving the acoustic feature dimensionality while retaining the primary spectral envelope for causal mask estimation. Under full 8-bit integer (INT8) post-training quantization, the model occupies 45.57 KB of Flash memory and 18.06 KB of internal SRAM. Experimental evaluation on the DEMAND dataset across signal-to-noise ratios from -5 dB to +10 dB confirms that the quantized model achieves robust noise suppression, providing consistent improvements in objective quality and signal clarity over traditional statistical baselines with negligible performance loss relative to floating-point precision. Measured on-device execution of the complete pipeline, encompassing hardware-accelerated Fourier transforms, neural mask inference, and overlap-add synthesis, processes a 72 ms audio context window in 22.97 ms entirely within on-chip SRAM, establishing a practical baseline for deep neural speech enhancement on low-cost edge devices.
Keywords
Speech enhancement,short-time Fourier transform (STFT),Microcontroller,Unet
Speaker
Kien Trong Nguyen
Lecturer Posts and Telecommunications Institute of Technology;Ho Chi Minh City

Submission Author
Thuan Quang Tran Posts and Telecommunications Institute of Technology;Ho Chi Minh City
Xuan Viet Le Posts and Telecommunications Institute of Technology (PTIT);Ho Chi Minh City
Kien Trong Nguyen Posts and Telecommunications Institute of Technology;Ho Chi Minh City
Comment submit
Verification code Change another
All comments

CONTACT US

Email: asiancomnet2026@usssociety.org

Website & IT Support: hi@aconf.org

Registration Submit Paper