Shangeth Rajaa

Research Scientist @

[Shangeth Rajaa @shangethr]

This came out of the work we're doing at https://t.co/52i0LjrPv1, and it's been a genuinely fun problem to dig into.

[Shangeth Rajaa @shangethr]

Happy to share that our research work "DualTurn: Learning Turn-Taking from Dual-Channel Generative Speech Pretraining" has been accepted at Interspeech 2026!

Preprint - https://t.co/02GU5pDMmn

#Interspeech2026 #VoiceAI #FullDuplex #TurnTaking

[Shangeth Rajaa @shangethr]

The idea is to bring full-duplex behavior into a modular pipeline, bridging traditional half-duplex ASR-LLM-TTS-VAD systems and genuinely full-duplex conversation, without giving up the LLM capabilities production systems depend on.

[Shangeth Rajaa @shangethr]

@chiragbarjatya Thanks for the motivation, @chiragbarjatya I didn’t get past the second line of your tweet 😅, but the idea of Diwali milestone stuck with me. From 98kg down to 89.xx kg, still a journey ahead, but I’m happy to have hit my Diwali target! 🥂 will enter Diwali with 89 weekly avg.

[Shangeth Rajaa @shangethr]

@monali_dambre I started using 2 bank accounts. Acc1 is where my salary, investment, major payments are done. Acc2 is my day to day spending, every month I transfer some money from Acc1 to Acc2. Acc2 is @TheJupiterApp, which is great for tracking UPI/cards/bank transfers.

[Shangeth Rajaa @shangethr]

🔗 Check it out:
Hugging Face Models:
• speechllm-2B: https://t.co/3p2b2D6vgZ
• speechllm-1.5B: https://t.co/EFct9lSsgP
GitHub Repository: https://t.co/HLtrZdDS4c

#MultiModalLLM #LLM #HuggingFace #ConversationalAI

[Shangeth Rajaa @shangethr]

We at @Skit_ai are thrilled to announce the release of our latest Multi-Modal LLM models for Speech Understanding on @huggingface, along with a comprehensive GitHub repository containing the code to train and infer these models!

https://t.co/HLtrZdDS4c
https://t.co/3p2b2D6vgZ

[Shangeth Rajaa @shangethr]

Due to the simplicity of training the model, any new perception/generation tasks could be added to the model eg: Multi-speaker transcript, speech environment classification, speech translation, etc.

[Skit.ai @Skit_ai]

👏We are proud to announce our team (@shangethr, Kriti A. (ex https://t.co/xX3RBgYTFd), and Swaraj Dalmia), in collaboration with Prof. Eng Siong Chng and Tarun Gupta of #NTUSpeechLab, have their paper "Improving Spoken Language Identification with Map-Mix" accepted at #ICASSP23

[Skit Tech @SkitTech]

We are excited to announce acceptance of a research work by @shangethr in ICASSP 2023!

https://t.co/uLom9feaEU

[Skit Tech @SkitTech]

We’re excited to release another open-source dataset. This is for phone-number capture from conversational human speech. Read more on this LinkedIn post https://t.co/BDPVAJr6wp

[Skit Tech @SkitTech]

We’re excited to release an open-source dataset for emotional Text to Speech (TTS). Read more on this LinkedIn post

https://t.co/ImMT1SFLFP