Conversational AI
<- Benchmarks

Text-to-Speech APIs

Streaming speech synthesis services and open-source engines for conversational AI output.

ElevenLabs

character or usage-based

Website Docs
Deployment
managed API
Modalities
streaming TTS, voice cloning
Strengths
voice quality; voice customization; agent platform adjacency
Tradeoffs
voice cloning needs clear consent processes

Cartesia

usage-based

Website Docs
Deployment
managed API
Modalities
streaming TTS
Strengths
low-latency streaming; voice-agent focus; realtime synthesis
Tradeoffs
managed provider dependency

Deepgram Aura

usage-based

Website Docs
Deployment
managed API
Modalities
streaming TTS
Strengths
pairs with Deepgram STT; streaming output; voice-agent workflows
Tradeoffs
best fit when Deepgram is already in the stack

Piper

open source

Website Docs
Deployment
local, self-hosted, edge
Modalities
local TTS
Strengths
fast local synthesis; edge-friendly; open-source
Tradeoffs
voice quality and language coverage depend on available models

What to measure next

  • time to first byte
  • streaming smoothness
  • voice naturalness
  • pronunciation controls
  • voice cloning consent model

V1 methodology note

This page is a curated directory, not a measured leaderboard. Scores will be added only when the methodology, scenarios, and raw assumptions are public.