Beyond Prompting: Efficient and Robust Contextual Biasing for Speech LLMs via Logit-Space Integration (LOGIC)

Name: Beyond Prompting: Efficient and Robust Contextual Biasing for Speech LLMs via Logit-Space Integration (LOGIC)
Uploaded: 2026-09-13T20:50:38.507073+00:00
Duration: 743 s
Description: The episode discusses 'Beyond Prompting: Efficient and Robust Contextual Biasing for Speech LLMs via Logit-Space Integration (LOGIC)'. Hosts analyze how LOGIC improves speech LLMs by integrating context directly into the prediction math, making the AI highly reliable, robust to noise, and adaptable for specialized or niche domains.

The episode discusses 'Beyond Prompting: Efficient and Robust Contextual Biasing for Speech LLMs via Logit-Space Integration (LOGIC)'. Hosts analyze how LOGIC improves speech LLMs by integrating context directly into the prediction math, making the AI highly reliable, robust to noise, and adaptable for specialized or niche domains.

Open the full episode