Skip to content

Speech To Text ​

Overview ​

The Speech To Text node accepts an audio input and returns a plain-text transcript using a configured speech recognition provider. Use it to transcribe battle audio submissions, voice prompts, or recorded model responses before passing the text downstream to other nodes. The node requires a valid provider credential and emits an error port for unrecognisable audio or API failures. Language and model quality can be tuned via configuration to balance cost against accuracy.

Configuration ​

FieldTypeRequiredDescription
providerenumYesSpeech recognition provider to use. Supported values: openai_whisper, google_speech, assemblyai, deepgram.
credentialIdstringYesID of the stored credential used to authenticate with the selected provider.
languagestringNoBCP-47 language tag (e.g. en-US, tr-TR) hinting the expected spoken language. Defaults to auto-detect when omitted.
modelenumNoTranscription model/quality tier. Values vary by provider (e.g. whisper-1, nova-2). Falls back to provider default when omitted.
promptstringNoOptional context string passed to the provider to improve accuracy for domain-specific vocabulary or proper nouns.
timestampsEnabledbooleanNoWhen true, the transcript output includes word-level timestamps in addition to the plain text.

Inputs ​

PortTypeDescription
audioaudio buffer or file referenceRaw audio data or a file reference (URL or storage key) pointing to the audio to transcribe. Accepted formats depend on the provider; WAV, MP3, and FLAC are broadly supported.
promptstringOptional runtime context prompt that overrides the static prompt set in config, useful for dynamic vocabulary hints.

Outputs ​

PortTypeDescription
transcriptstringThe plain-text transcript of the audio input.
transcriptDataTranscriptResultStructured transcription result containing the text, detected language, confidence score, and optional word-level timestamps when timestampsEnabled is true.
errorNodeErrorEmitted when transcription fails — e.g. unsupported audio format, inaudible input, or provider API error. Contains a code and human-readable message.

Example ​

json
{
  "nodeType": "speech_to_text",
  "config": {
    "provider": "openai_whisper",
    "credentialId": "cred_openai_prod",
    "language": "en-US",
    "timestampsEnabled": false
  }
}