# Shangeth Rajaa — Voice AI Researcher

> Senior ML Scientist specializing in Turn-Taking, Full-Duplex Spoken Dialogue Systems, Multi-Modal Speech LLMs, and Conversational AI. 6+ years of Voice AI research and engineering experience across industry and academia. Built and led ML and research teams in early-stage startups. 7 peer-reviewed publications at Interspeech, ICASSP, NeurIPS, and PMLR.

## Contact & Links

- Website: https://shangeth.com
- Email: shangethrajaa@gmail.com
- LinkedIn: https://www.linkedin.com/in/shangeth
- GitHub: https://github.com/shangeth
- Google Scholar: https://scholar.google.com/citations?user=apmFPkAAAAAJ
- Book a call: https://calendly.com/shangeth-anyreach/30min

## Current Role

Senior Machine Learning Scientist at Anyreach AI (Feb 2025 — Present)
- Turn-Taking in Full-Duplex Spoken Dialogue Systems
- Multi-Modal LLMs for Speech Understanding and Synthesis
- Automatic Speech Translation with Multi-Modal LLMs

## Research Interests

- Turn-Taking & Full-Duplex Spoken Dialogue
- Multi-Modal Speech LLMs
- Spoken Language Understanding (SLU)
- Speaker Profiling & Representation Learning
- Conversational AI Systems (ASR, TTS, NLU, Dialogue Management)
- RLHF/DPO for goal-driven spoken dialogue

## Experience

- Anyreach AI — Senior ML Scientist (Feb 2025 — Present)
- ScoreTravel AI — Senior ML Researcher (Aug 2024 — Feb 2025)
- Skit.ai (Vernacular.ai) — ML Researcher (Jun 2021 — Jun 2024)
- Speech and Language Lab, NTU Singapore — Research Assistant (Aug 2020 — Jun 2021)
- IBM Research Labs — Research Intern (May 2020 — Aug 2020)
- INRIA Paris — Research Collaborator, AutoDL (Apr 2019 — Jul 2020)

## Publications

1. DualTurn: Learning Turn-Taking from Dual-Channel Generative Speech Pretraining — Submitted to Interspeech 2026. https://arxiv.org/abs/2603.08216
2. Improving End-to-End SLU performance with Prosodic Attention and Distillation — Interspeech 2023. https://arxiv.org/abs/2305.08067
3. Improving Spoken Language Identification with MAP-Mix — ICASSP 2023. https://arxiv.org/abs/2302.08229
4. Skit-S2I: An Indian Accented Speech to Intent Dataset — arXiv 2022. https://arxiv.org/abs/2212.13015
5. Learning Speaker Representation with Semi-supervised Learning for Speaker Profiling — arXiv 2021. https://arxiv.org/abs/2110.13653
6. Towards Automated Deep Learning: Analysis of the AutoDL Challenge Series 2019 — NeurIPS / PMLR 2020. https://proceedings.mlr.press/v123/liu20a.html
7. Overview and Unifying Conceptualization of Automated Machine Learning — ECML PKDD 2019.
8. RL based framework for optimal data quality remediation sequence for ML — SIGMOD 2021.
9. Convolutional Feature Extraction and Neural Arithmetic Logic Units for Stock Prediction — ICACDS 2019. https://arxiv.org/abs/1905.07581
10. SpeechLLM: Multi-Modal LLM for Speech Understanding — GitHub/HuggingFace. https://github.com/skit-ai/SpeechLLM

## Education

Dual Degree — B.E. Electrical & Electronics Engineering and M.Sc. Mathematics, BITS Pilani, India

## Availability

Open to Voice AI consulting, research advisory, and technical collaborations.
Rate reflects senior research expertise. Book a 30-min call: https://calendly.com/shangeth-anyreach/30min