Module livekit.agents.inference.eot.base

Audio EOT detector base, the per-window inference stream (with built-in cloud→local fallback), and the transport Protocol concrete backends satisfy.

Lives next to its transports rather than in voice/turn so the fallback logic is a peer of the transports it switches between rather than a template-method-across-packages.

Global variables

var MIN_SILENCE_DURATION_MS

Minimum VAD silence the audio EOT detector needs before it sends an inference request. Enforced against the caller-supplied VAD's min_silence_duration in AudioRecognition.

Classes

class TurnDetectorOptions (sample_rate: int, thresholds: ThresholdOptions)

Expand source code

@dataclass
class TurnDetectorOptions:
    sample_rate: int
    thresholds: ThresholdOptions

TurnDetectorOptions(sample_rate: 'int', thresholds: 'ThresholdOptions')

Instance variables

var sample_rate : int

var thresholds : ThresholdOptions