on_speech_started event is unusable

[Vaibhav-Lodha]

What is the current behavior?

on_speech_started gets triggered in cases where there is just background noise, making it unusable with cases where we want to use it as a VAD where it can tell in exactly when user has started speaking.

Steps to reproduce

1. Use the given config -

sample_rate = 16000
model = "nova-2"
default_options = LiveOptions(
            language=Language.HI,
            model=model,
            sample_rate=sample_rate,
            vad_events=True,
            endpointing=300,
            utterance_end_ms="1000",
            interim_results=True,
            punctuate=False,
        )

2. Register on_speech_started event, and log when it gets triggered without saying anything, it will keep on logging event.

Expected behavior

1. Only trigger when there is actual word / speech detected.

Please tell us about your environment

• Operating System/Version: MacOS 15.3 / Ubuntu 22.04
• Python Version: 3.12