Q n A with Dr Ananth Sankar
summary
2005 single-author Nuance paper demonstrates system combination that measurably improves recognition accuracy; recent deep learning work drives word error rates down and Microsoft Research claims near-human WER by using a dialogue-level language model that conditions on entire conversations. Remaining challenges are making models real-time and streaming and handling noisy, multi‑speaker meeting transcription for true human-to-human speech recognition. Advice: master the underlying ML math and transfer it across applications; domain knowledge is learned on the job.
AI-written summary · about
related talks
25mBeyond Transcription: Building Voice AI That Understands Conversations
13mAI: Big Expectations
19mLLM-Enhanced Multimodal AI: Revolutionizing Audio/Video Interaction
04mMike Tamir Interview
27mThe Cutting Edge of AI: From LLMs to Multimodal Intelligence
42mdata.bythebay.io: Shubho Sengupta,
27mApplying ML at scale for New User Experiences
46mBeyond Deep Learning: The Cognitive AI Frontier