Shatrughan Singh

About Shatrughan Singh

With a deep-rooted passion for advancing conversational AI, my role as Architect and…

Experience

1. Architect

   Samsung Electronics

   Jun 2007 — Present · Bengaluru, IN

   Automatic Speech Recognition(ASR) for Bixby and Galaxy AI- ASR Lead for Galaxy AI, developed ASR models for 5 languages. Based on pretrained(BEST-RQ) Causal Non-causal Cascaded Conformer. Enables applications like Live Translate, Voice Recorder transcription, Interpreter etc- Product owner for development and productisation of On-device ASR models for On-device Bixby(Mobile, TV and appliances) for Indian and UK English- Product owner for Bixby Text Call / Call Screening for Indian and UK EnglishResearch Lead for End to End ASR model development in collaboration with Samsung Research, Seoul- Streaming ASR based on Monotonic Chunkwise Attention(MoChA)- 2-pass ASR based on RNN Transducers(RNN-T LAS 2-pass) and Conformer Transducers(Cascaded Causal Non-Causal)- Model compression for on-device deployments based on SVD and Knowledge Distillation- Multilingual ASR based on self-supervised learning(wav2vec2)- LoRA based language adapters for multilingual ASR- SpeechLLMs Autoregressive Language Models- Language models based on Transformer and Tranformer-XL- LM fusion to improve ASR accuracyASR Components- Named Entity Correction to address errors in ASR hypothesis for named entities- Neural Confidence Score to indicate the confidence of ASR system in its transcriptionsProducts and services for mobile platforms- Cloud based Gallery and Files sync service(Samsung Cloud)- Knox security platform for Android- Device based search service- Text input prediction for mobile keyboard

Skills

• Debugging
• Machine Learning
• Rtos
• Android
• Software Design
• Cloud Storage
• Software Engineering
• Clearcase
• Embedded Software
• C
• Data Structures
• Python
• Embedded C
• Embedded Systems
• Arm
• Embedded Linux
• Software Development
• Device Drivers
• Algorithms