Conversational AI Insight Discovery Tool — 60+ curated research papers spanning 75 years, from Turing Test (1950) to Full-Duplex Real-time Interaction (2026).
| Metric | Value | Description |
|---|---|---|
| Papers | 60+ | Curated research collection |
| Human Turn Gap | ~200ms | Natural conversation pause |
| GPT-4o Latency | ~320ms | Approaching human-level |
| Target | Full-Duplex | Listening while Speaking |
Visual tier-based taxonomy for Voice Agent architecture:
| Tier | Category | Description |
|---|---|---|
| 0 | Hall of Fame | Foundation classics (Turing, ELIZA, Transformer) |
| 2 | Ear (ASR) | Audio input processing |
| 3 | Brain (LLM) | Core intelligence layer |
| 4 | Voice (TTS) | Audio output synthesis |
| 5 | Support | Benchmarks & Tools (VoiceAgentEval) |
| 6 | Global | System-level design |
Why typical voice bots feel robotic:
- Turn-Taking Gap — 1-3s response delay breaks flow
- Barge-In — Can't hear interruptions while speaking
- Prosody & Affect — Text conversion loses emotion
- Context Loss — Single models can't handle orchestration
Solution: Omni-Duplex Architecture with streaming audio tokens
- 🗺️ Interactive Pipeline Map — Visual tier-based paper taxonomy
- 📊 Core Metrics Dashboard — WER, latency, research trends
- 🎯 Experience Gaps Analysis — Why bots feel robotic
- 📅 Historical Timeline — 1950-2026 milestones
- 🔍 Smart Search — Filter by era, tier, tags
Built with ❤️ for the Voice Agent community.
© 2026 PaperHunt