CrossOracle: An Agentic Framework for Real-Time Expert Personas with Parallelized Acoustic Synthesis
Year: 2026
Type: conference-paper
Abstract: We present CrossOracle, a productiondeployed spoken dialogue system (SDS) optimized for real-time Hinglish (Hindi-English code-switched) voice interaction across eleven distinct Expert Persona identities.The system addresses the Latency Wall inherent to cascaded Automatic Speech Recognition (ASR) → Large Language Model (LLM) → Text-to-Speech (TTS) pipelines by introducing a parallelized asynchronous TTS queue and a language-aware semantic chunking heuristic.Key empirical results across 8,000 conversational turns include a mean TTS Real-Time Factor (RTF) of 0.107, a mean Time-to-First-Byte (TTFB) of 4.9s, and a barge-in resolution rate of 100% with a mean abort latency of < 1 ms.The live system is accessible at https://crossoracle.io.
Author: Yash Raj Singh
Topic: Speech and dialogue systems
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W7214005291
FWCI:
Cites:
Cited by:
Related to: