Skip to content

Engineering

Speech-to-Speech vs Chained STT-LLM-TTS: Choosing a Voice Architecture

Speech-to-speech wins on latency and naturalness. Chained pipelines win on control, cost and auditability. How to pick, and why most production systems I build end up chained.

There are two ways to build a voice agent and the choice shapes everything downstream: what it costs, how fast it feels, what you can log, and whether you can answer a compliance question six months from now.

Chained: speech to text, model, text to speech

Audio becomes text, text goes to a language model, the model's text becomes audio. Three services, three hops, and a clean transcript at every step.

Speech-to-speech

Audio goes straight into a multimodal model and audio comes back out. The OpenAI Realtime API is the best-known implementation. Two hops disappear, and so does the information a transcript throws away: tone, emphasis, hesitation, emotion.

The trade, plainly

Speech-to-speechChained
First audio latency500 - 900 ms800 ms - 1.4 s
NaturalnessHigher: keeps tone and emphasisFlatter: tone is lost at transcription
Model choiceWhoever ships a realtime modelAny model you like
Transcript qualitySecondary output, sometimes lossyFirst-class at every turn
Cost on long callsHigher: audio tokens both waysLower: text tokens
Per-layer controlLittleTotal
AuditabilityHarderStraightforward

How to choose

Speech-to-speech when the conversation is short, latency is the product, and emotional nuance matters: companions, tutors, consumer conversational apps.

Chained when calls are long, when you need a specific model, when cost per call matters at volume, or when someone will eventually ask for a transcript of what the agent said. That covers most business use cases: support, scheduling, qualification, intake.

What I usually ship

Chained, inside LiveKit, for most production systems. The 200 to 400 milliseconds it costs is worth a reliable transcript, free model choice and predictable economics. I reach for speech-to-speech when the product is a conversation rather than a transaction, and the voice AI companion I spent a year on is exactly that case.

You are not locked in either way if you build the pipeline behind an abstraction. That is worth doing on day one; both sides of this comparison are moving fast. See OpenAI Realtime API development for the speech-to-speech side, and AI voice agent development for the chained build.

Direct Project Enquiry

Tell me what you are building

Share what you are building and where it is stuck. If a managed platform already solves it, I will say so. I review every enquiry personally and reply within about four hours.

  1. 1Send your project goal and the services you think you need.
  2. 2I review the context and whether I am the right fit.
  3. 3You get a direct reply with a practical next step.

International projects welcome

I work in English by email, video call or through Upwork, from GMT+7 with overlap hours for US and EU teams.

Read the Privacy Policy, Terms of Service and Cookie Policy.

Client project enquiries only

This form is for businesses and teams looking to hire. Sales pitches, recruitment messages, guest posts and link-building outreach are not reviewed.

Select Service *

In a hurry? Message me on WhatsApp at nguyentienm@gmail.com

Message me