Batched inference uses Voice Activity Detection (VAD) filter and ignores the following paramters: compression_ratio_threshold, logprob_threshold, no_speech_threshold, condition_on_previous_text, ...