Skip to content

Voice Infrastructure

The components a voice agent is made of — recognition, synthesis, transport, telephony. Bought directly by teams building their own stack, and rebilled to everyone else inside a platform fee.

The thing buyers get wrong here

Latency numbers are quoted as time-to-first-byte, which is not what a caller experiences. Mouth-to-ear includes endpointing, the model and the network, and it is typically three to five times the quoted figure.

All 35

Published prices only

30 of 35 publish a price you can read without a sales call. The rest are shown as Custom, which is a fact about them rather than a gap in our research.

ToolEntry priceModelFree tierSelf-serveScore
Fish Audio$0.00/minflatYesYes4.1
Kokoro$0.00/minflatYesYes4.1
Whisper$0.00/minflatYesYes4.1
AssemblyAIFree tierusageYesYes3.9
Azure AI SpeechFree tierusageYesYes3.9
BasetenFree tierusageYesYes3.9
CartesiaFree tierusageYesYes3.9
DeepgramFree tierusageYesYes3.9
ElevenLabs$6/mousageYesYes3.9
Gemini Live APIFree tierusageYesYes3.9
GladiaFree tierusageYesYes3.9
GroqFree tierusageYesYes3.9
Hume AIFree tierusageYesYes3.9
ModalFree tierusageYesYes3.9
NeuphonicFree tierusageYesYes3.9
Plivo$0.0115/minusageYesYes3.9
Rev AIFree tierusageYesYes3.9
RimeFree tierusageYesYes3.9
Rime ArcanaFree tierusageYesYes3.9
SieveFree tierusageYesYes3.9
SignalWire$0.008/minusageYesYes3.9
Smallest.aiFree tierusageYesYes3.9
SonioxFree tierusageYesYes3.9
SpeechmaticsFree tierusageYesYes3.9
Telnyx$0.007/minusageYesYes3.9
Twilio$0.014/minusageYesYes3.9
OpenAI Realtime APICustomusageYes3.8
DailyFree tierusageYesYes3.7
LiveKitFree tierusageYesYes3.7
VonageFree tierusageYesYes3.5

Other categories