Skip to content

Speech to Text

Speech to Text is the hearing of a voice agent with any speech recognition provider in NappAI: you choose Provider and the agent hears exactly as with that provider’s own component, such as Deepgram STT or OpenAI STT.

The difference is that switching provider does not mean switching node or rewiring the agent: you change it in the field, and the node remembers the credential of every provider you already set up.

It plugs into the STT connector of the Voice Agent node.

When to use it

  • You want to try several providers with the same agent and compare how well they understand your callers.
  • You want to be able to move from a paid provider to a local one, or the other way round, without touching the rest of the flow.

If the provider will not change, use its own component: it shows all its fields directly.

Configuration

FieldWhat it does
ProviderThe speech recognition provider. Default: OpenAI. The options are Deepgram, OpenAI, Gemini, ElevenLabs, Azure, GroqWhisper, AssemblyAI, Cartesia, Gladia, Inworld, WitAI and the local Parakeet and Vosk. Choose it first: when you change it, the node asks for that provider’s credential.
Credential (required if the provider uses one)The credential of the chosen provider. The list only shows that provider’s credentials. Parakeet and Vosk need none. See Credentials.
ModelThe provider’s transcription model. When you switch provider it is filled with that provider’s default. You can type any model name.
LanguageCode of the language the caller is expected to speak, for example es or en. Default: es.
Detect languageWhen on, the agent switches language when the caller does, with the providers that allow it. Off by default.

Advanced options

FieldWhat it does
Provider SettingsOther fields of the provider’s own component, each with its own name. For example, region = eu with AssemblyAI. Fields the provider does not have are ignored.
Credentials by ProviderThe credential the node remembers for each provider. It fills itself in when you choose a credential.

Each provider’s own options are explained on its page, for example in Deepgram STT.

Output

The STT output holds the speech recognition settings. It plugs into the STT connector of the Voice Agent node, or into Fallback STT if you turn on Fallback Engines and want to use it as a backup.

Example

A dental clinic wants to know which provider best understands the names of its treatments.

  1. Connect Speech to Text to the STT connector of the Voice Agent.
  2. Choose Deepgram and its credential, and test the agent in the playground.
  3. Switch to OpenAI, choose its credential and repeat the same test.
  4. Keep the one that works best and deploy the flow.

Good practices

  • The voice agent answers calls with the deployed version of the flow: after switching provider, deploy so that real calls use it.
  • The spend is recorded on the chosen provider’s credential. The differences between local and cloud engines are explained in Quality, noise and latency.