Ankit Kumar is a Sesame co-founder and a co-author of its conversational speech research. His earlier work includes co-founding Ubiquity6 and leading Discord’s Clyde AI engineering effort.[1][2]
Start with Conversational Speech Model →
Career
Kumar’s former Ubiquity6 co-founder Anjney Midha describes their earlier company and Kumar’s later work productionizing language and speech models at Discord. Kumar’s public professional profile lists mathematics studies at Stanford from 2011 to 2015. His Sesame work brings that research and product experience into conversational computing.[1][5]
Contributions
Generate speech using conversational history
Kumar is among the authors of Sesame’s Conversational Speech Model technical account. CSM conditions generation on interleaved text and audio history. A backbone predicts the first audio codebook, and a smaller decoder generates the remaining codebooks. The design keeps speech generation within one multimodal system while reducing latency.[2]
Keep a reproducible implementation available
Sesame released a CSM implementation and a 1B-parameter checkpoint for experimentation. The repository shows how text, speaker labels and preceding audio segments form an input. The released checkpoint is a base model, not the complete production assistant behind Sesame’s public app.[3]
Tools, writing & talks
Tools
Start here
Conversational Speech Model
github.comOpen code and a speech-generation checkpoint that conditions on preceding conversation.
Start with the example showing preceding audio context and speaker labels. Python and model-inference experience help; the checkpoint is a research component, not the complete Sesame assistant.
Research co-author[3]