Agenda

Speech AI for games must work fast, integrate seamlessly, and perform under demanding conditions: noisy in-game comms, emotionally charged callouts, and rapid meta shifts. This session will cover an end-to-end approach to training and shipping speech-driven systems for games—real-time comms moderation and coaching, a multilingual AI teammate that understands and speaks naturally, and simultaneous translation for e-sports commentary. The talk will explore how Tencent balanced latency, cost, safety, and live-ops iteration using AWS as the infrastructure layer.

Takeaway

Attendees will learn a practical blueprint for building and shipping speech AI in games: training data strategy, low-latency inference, evaluation for noisy multiplayer audio, safety/moderation, multilingual voice interactions, and live-ops iteration. Includes pitfalls, system patterns, and integration guidance for comms, AI teammates, and live translation.

Intended Audience

For game AI/ML engineers, audio/voice engineers, backend/platform teams, and producers planning speech features for multiplayer, companions, or broadcast. Best for teams shipping live services. Pre-reqs: basic ML concepts and familiarity with game networking/online latency constraints; no deep speech research required.