SpeakerIdTool interface

Speaker identification tool configuration (CAM++ voice embeddings against an enrolled gallery). Requires the vad tool in localAnalysis — only VAD-active audio is accumulated, and utterances are scored at speech end (or at maxUtteranceMs during long speech). Broadcast audio carries music beds and channel colour, hence the conservative default matchThreshold; lower it for clean speech.

Signature:

export interface SpeakerIdTool

Properties

Property	Type	Description
gallery	{ name: string; wav: Uint8Array; }[]	Enrolled voices: name + a 16kHz mono WAV (PCM16 or float32) of that speaker. Embedded at node creation; a bad WAV fails configuration.
gate?	Gate	(Optional) Only run speaker identification while the gate holds. Video-derived signals (face.identity, object_detection.label, scene state, …) reach the audio chain via the shared signal bus, so referenced video tools may appear anywhere in the localAnalysis list — e.g. only attribute speech while an enrolled face is on screen. While the gate is closed, part-accumulated utterances are dropped.
id?	string	(Optional) Instance id (G3): names this instance so gates can reference it and the same tool may appear more than once. Default: the tool name. Must be unique across localAnalysis.
matchThreshold?	number	(Optional) Cosine similarity an enrolled voice must reach. Default 0.7.
maxUtteranceMs?	number	(Optional) Score-and-restart cap during continuous speech, ms. Default 6000.
minUtteranceMs?	number	(Optional) Shortest utterance worth scoring, ms. Default 1000.
model?	"campplus"	(Optional) Today’s model, named for forward compatibility.
tool	"speaker_id"
windowSecs	number

Property

Type

Description

gallery

{ name: string; wav: Uint8Array; }[]

Enrolled voices: name + a 16kHz mono WAV (PCM16 or float32) of that speaker. Embedded at node creation; a bad WAV fails configuration.

gate?

Gate

(Optional) Only run speaker identification while the gate holds. Video-derived signals (face.identity, object_detection.label, scene state, …) reach the audio chain via the shared signal bus, so referenced video tools may appear anywhere in the localAnalysis list — e.g. only attribute speech while an enrolled face is on screen. While the gate is closed, part-accumulated utterances are dropped.

id?

string

(Optional) Instance id (G3): names this instance so gates can reference it and the same tool may appear more than once. Default: the tool name. Must be unique across localAnalysis.

matchThreshold?

number

(Optional) Cosine similarity an enrolled voice must reach. Default 0.7.

maxUtteranceMs?

number

(Optional) Score-and-restart cap during continuous speech, ms. Default 6000.

minUtteranceMs?

number

(Optional) Shortest utterance worth scoring, ms. Default 1000.

model?

"campplus"

(Optional) Today’s model, named for forward compatibility.