TranscriptionResult

Result of a speech-to-text transcription.

Defined in: src/core/types/providers.ts:225

Result of a speech-to-text transcription.

Remarks

Represents a transcription result emitted by an STT provider. The result can be interim (partial, still being refined) or final. For providers like Deepgram that support multi-segment utterances, additional flags (speechFinal, isPreflight) indicate when the complete utterance is finished and when speculative processing can begin.

Example

sttProvider.onTranscription((result: TranscriptionResult) => {
  if (result.isPreflight) {
    console.log('Preflight:', result.text);
  } else if (result.speechFinal) {
    console.log('Final utterance:', result.text);
  } else if (result.isFinal) {
    console.log('Segment final:', result.text);
  } else {
    console.log('Interim:', result.text);
  }
});

See

RestSTTProvider.onTranscription for registering transcription callbacks
LiveSTTProvider.onTranscription for registering transcription callbacks
TranscriptionFinalEvent for the corresponding SDK event

Properties

Property	Type	Description	Defined in
`confidence?`	`number`	Confidence score for the transcription, from 0 (lowest) to 1 (highest). Remarks Not all providers supply confidence scores. When unavailable, this is `undefined`.	src/core/types/providers.ts:292
`isFinal`	`boolean`	Whether this is a final transcription segment or an interim (partial) result. Remarks Interim results update as the user speaks and are replaced by subsequent results. Final results represent a committed segment of the transcription.	src/core/types/providers.ts:236
`isPreflight?`	`boolean`	Whether this is a preflight/eager-end-of-turn signal. Remarks DeepgramFlux (e.g., `flux-general-en`) can emit this before `speech_final` to allow the next pipeline stage (LLM) to start generating speculatively. The text may still change slightly when the confirmed `speech_final` arrives. See - EagerLLMConfig for enabling speculative generation - TranscriptionPreflightEvent for the corresponding SDK event	src/core/types/providers.ts:284
`metadata?`	`Record`<`string`, `unknown`>	Additional provider-specific metadata. Remarks May contain information such as word-level timestamps, speaker diarization data, or other provider-specific fields.	src/core/types/providers.ts:301
~~`speechFinal?`~~	`boolean`	Whether this result marks the end of a complete utterance. Remarks For Deepgram: `true` when `speech_final=true` (the speaker has stopped talking). For NativeSTT and other providers that emit one result per utterance: equals isFinal. Multi-segment providers (Deepgram) may emit several `isFinal: true` chunks for a single utterance — only the last one has `speechFinal: true`. Deprecated Use utteranceComplete instead. This field is retained for informational/display purposes but is no longer used by CompositeVoice to trigger LLM processing.	src/core/types/providers.ts:252
`text`	`string`	The transcribed text.	src/core/types/providers.ts:227
`utteranceComplete?`	`boolean`	Whether this result should trigger the next pipeline stage (LLM). Remarks This is the single flag CompositeVoice checks to decide when to send transcribed text to the LLM. The STT provider is responsible for setting this — it knows its own end-of-utterance semantics. - DeepgramSTT: `true` when `speech_final=true` (after buffering all `is_final` segments into a complete utterance). - NativeSTT: `true` when `isFinal=true` (one result per utterance). - AssemblyAI: `true` on `FinalTranscript` messages. - ElevenLabs: `true` on `committed_transcript` / `final_transcript`. If not set, CompositeVoice will NOT trigger LLM processing for this result.	src/core/types/providers.ts:270