Varpy example: line-by-line stack breakdown

Host uses Varpy to explain STT → LLM → TTS pipeline, transports, and cost components per minute.

Play episode from 01:36

chevron_right

Transcript

chevron_right

Transcript

Episode notes

TLDR: Voice AIs aren't that much cheaper in the year 2025

My friend runs a voice agent startup in Canada for walk-in clinics. The AI takes calls and uses tools to book appointments in the EMR (electronic medical record) system. In theory, this helps the clinic hire less front desk staff and the startup makes infinite money. In reality, the margins are brutal and they barely charge above cost. This is surprising to me: surely a living, breathing, squishy human costs more per hour than a GPU in a datacenter somewhere?

An industry overview of voice AIs

Broadly speaking there are 3 types of companies in the voice AI industry

Foundation model companies:
1. These companies actually train the text to speech and realtime audio models
2. Openai, Elevenlabs, Cartesia
Pipeline companies
1. Infrastructure companies that aggregate multiple foundation model providers and help you experiment with multiple providers, build agents, and connect with SIP and WebRTC transports (think OpenRouter but with extra steps).
2. Developer focused: N8n, Bland, Vapi
3. Enterprise focused: Ada, Sierra, Fin
Vertical startups
1. Startups that do "voice agents for {healthcare | logistics | real estate | etc }"
2. Here's 142 of them

Of course, these [...]

---

Outline:

(00:47) An industry overview of voice AIs

(01:57) The line by line breakdown

(02:22) Speech to Text (STT) = LLM = Text to Speech (TTS)

(03:09) Realtime API

(03:42) Comparison to Humans and Business Process Outsourcing (BPO)

(04:46) Assumptions

(05:20) Limitations

(06:40) The Future

(07:43) Conclusion