pdum_aiui

Appearance

Interface: OpenaiTranscriptionTransportOptions ​

Defined in: openai.ts:151

@habemus-papadum/aiui-stt — realtime speech-to-text as a pencil-shaped component. Contract of record: the aiui-stt proposal (pdum_aiui git history); the wire facts underneath are the channel's, hard-won and ported verbatim (see each engine module's header).

The contract is the pencil's, with text in place of ink: start a session, feed it audio, watch a stream of CUMULATIVE text deltas, and know when a segment is final. The default engine is ElevenLabs Scribe v2 realtime (scribeTransport); the OpenAI transcription-type realtime flavor (openaiTranscriptionTransport) is the alternate. Both sit behind one seam (SttTransport → SttConnection), so the vendor difference is confined to the open.

Credentials are injected, never owned (D3): Scribe takes a connectUrl() minting one single-use token per connect; OpenAI takes the oracle's KeySource. Audio is injected too (D4): an AudioSource of 24 kHz PCM16 frames — aiui-stt/mic ships the real microphone; a test or a file player is the same seam. Display is LiveDiffText (aiui-viz/modal) — this package ships no renderer (D6).

Properties ​

delay? ​

optional delay?: string

Defined in: openai.ts:158

The latency/accuracy knob — a gpt-realtime-whisper-ONLY param (other models over this wire reject it); omitted when unset.

keySource ​

keySource: KeySource

Defined in: openai.ts:153

The oracle's credential seam, reused verbatim (chain/caching compose).

now? ​

optional now?: () => number

Defined in: openai.ts:164

Injected clock (tests).

Returns ​

number

socketFactory? ​

optional socketFactory?: SttSocketFactory

Defined in: openai.ts:162

Injected socket (tests); defaults to the platform WebSocket.

transcriptionModel? ​

optional transcriptionModel?: string

Defined in: openai.ts:155

Default DEFAULT_OPENAI_TRANSCRIPTION_MODEL.

url? ​

optional url?: string

Defined in: openai.ts:160

Override the endpoint (tests).