Everything on voice and chat AI agents — real-time speech processing, telephony and phone-system automation, conversational AI, voice UX design, and production deployment.
From the blog
Vendor TTS demos always sound great. Production with your prompts on your audio path is a different story. Here is how we monitor MOS, CMOS, and prosody drift across ElevenLabs, OpenAI, and Cartesia in production.
Human turn gaps run 200-300ms. Most agents hit 800-1500ms. We unpack VAD vs semantic turn detection, duplex audio pipelines, and CallSphere's per-vertical barge-in thresholds.
92% of nonprofits use AI but only 7% report major impact. The gap is voice — AI voice campaigns reach mid-tier donors humans never call. Here is the compliant nonprofit fundraising build.
Krisp's VIVA SDK isolates the primary speaker before STT. Wire it as a pre-processor in front of LiveKit/Pipecat for 30%+ WER drop in noisy calls.
ElevenLabs shipped v3 expressive TTS and consolidated its agents stack into ElevenAgents in March 2026. The MCP-native, action-taking voice agent has arrived.
Mistral's Voxtral 4B-TTS (March 26 2026) is open-weights, 3 GB quantized, and runs on a single 16 GB GPU. Here's the full local voice agent build using Voxtral Mini for STT and Voxtral TTS.
Build a triage RealtimeAgent that hands off to specialist agents (billing, scheduling) in a single voice session — TypeScript code, real handoff events, and CallSphere patterns.
Ofcom's January 2025 CLI guidance, the new international scam-call blocking, the PECR live-vs-automated distinction, and what UK-bound AI voice agents must do to stay legal.
Expo Go cannot run WebRTC. EAS Dev Client can. Here is the 2026 config-plugin recipe for shipping AI voice agents on Expo without abandoning the managed workflow.
