Latest articles
Everything Abhishek has published, most recent first.
VoiceHow to detect an AI-generated voice on live calls
Real-time AI voice detection on live phone calls: how it works, where it fits in a call pipeline, and what the PSTN transport does to detection accuracy.
Conversational AIYou can now use Deepgram Flux on Telnyx
Host Deepgram’s Flux CSR model inside Telnyx’s global network to reduce latency and build smoother voice-AI conversations.
- Agent Primitive
Benchmarking Tavily, Telnyx, DuckDuckGo, and Brave for agent workflows
We ran 20 queries through Tavily, Telnyx Web Search, DuckDuckGo, and Brave. Brave was fastest at 561ms avg with zero errors. Tavily returned the richest snippets. DuckDuckGo failed 8 out of 20 times.
MessagingHow to give your Muse Agent an inbox
Muse has no inbox of its own. Give any AI agent a real email address with the Telnyx Email API, on the same platform as voice and SMS.
- Conversational AI
An open source medical pronunciation dictionary for voice AI
960 medical terms with phonetic alias and IPA pronunciations, pre-built for Telnyx, ElevenLabs, Vapi, Retell, and Amazon Polly. Open source, provider-agnostic, 30 seconds to import.
TTS APITTS architecture decides what breaks first
Text-to-speech models do not only differ by voice quality. The architecture underneath the voice determines latency, cloning quality, streaming behavior, controllability, and cost.
STT APIHow I built a voice dictation app for macOS using Telnyx
A full breakdown of building a Wispr Flow alternative on macOS using Telnyx STT and LLM inference APIs — pipeline architecture, permissions, hotkeys, and real-world latency.
IntegrationsBuild Voice AI Agents With Vision Agents and Telnyx
The Telnyx plugin for GetStream Vision Agents now covers the full voice AI stack: PSTN phone calling, LLM, STT, and TTS, all from a single install.
Conversational AIHow we built multi-participant voice AI calls
Adding a second human to an active voice AI call breaks assumptions your conversation layer has held since day one. Here is what changes in the audio layer, the transcript state, and the tool design when you build for multiple speakers.
Conversational AIImproving Voice AI Latency with Co-Located Infrastructure
Most voice AI feels robotic because of latency, not bad voices. Every network hop adds delay. We fixed it by colocating our GPUs with our telephony core, cutting response times sub second and making conversations feel human again.
