Open to opportunities

Aditya Srivastava

@adityasrivastava4

Message

I build speech and acoustic machine-learning systems for enhancement, localization, and real-time detection.

India

Message

What I'm looking for

I'm looking for a Machine Learning Scientist role in speech processing and automatic speech recognition, where I can build acoustic and speech ML systems spanning audio signal processing, speech enhancement, sound-source localization, and efficient real-time inference.

I'm building acoustic UAV detection at IIT Delhi's CARE Lab, developing a physics-grounded dataset pipeline and benchmarking STFT, MFCC, and log-mel frontends across CNN, CRNN, and RNN detectors.

At CSIR–Central Scientific Instruments Organisation, I built multichannel CRNN and graph-inspired direction-of-arrival models, reaching 97.8% azimuth accuracy and 2.3°/0.9° MAE. This work contributed to my IEEE Sensors Letters publication on direction-of-arrival estimation under sensor failure conditions.

I've also reimplemented GTCRN speech enhancement in PyTorch, achieving PESQ 2.82 and STOI 0.941, then built an INT8 ONNX quantization pipeline for lower-latency deployment. I enjoy turning audio research into practical speech, localization, and embedded-inference systems.

Experience

Work history, roles, and key accomplishments

Current

Research Intern

Current

Indian Institute of Technology Delhi

May 2026 - Present (3 months)

Building acoustic UAV detection at IIT Delhi's CARE Lab, including a physics-grounded dataset pipeline and benchmarking of frontends and detectors.

Log Mel CNN RNN SincNet SNR Sweep Onset Latency

Research Intern

CSIR – Central Scientific Instruments Organisation

May 2025 - Jun 2025 (1 month)

Built a CRNN on multichannel STFT features for 3D direction classification, achieving 97.8% azimuth accuracy, and developed a graph-inspired pairwise-MLP regressor for continuous DoA estimation.

Batch Normalization BiGRU Multi Channel Audio Azimuth Doa Pairwise MLP IEEE Sensors Letters