//! # Speaker diarization: the parts that do not depend on the model
//!
//! A diarizer answers "who spoke when". Everything about that which does not
//! depend on the model lives here; a model plugs in as a
//! [`SpeakerActivityModel`].
//!
//! - [`SpeakerActivity`]: what a diarizer emits, the probability that each
//! speaker is talking in each frame, `[frames, speakers]`. Speakers are
//! numbered by the diarizer, in the order it first heard them.
//! - [`SpeakerActivityModel`] and [`SpeakerActivityStream`]: the diarizer seam.
//! The model is shared and immutable; each stream opens its own stream state,
//! takes samples in pieces of any size, and hands back the frames it has
//! finalized.
//! - [`SpeakerTurn`] and [`TurnFilter`]: activity to turns, one speaker's
//! stretch of speech each, by `NeMo`'s diarization post-processing
//! ([`TurnFilterConfig`]: hysteresis thresholds, padding, and the short-turn
//! and short-gap filters), streamed or over a whole recording.
//! - [`RttmSegment`]: turns as NIST RTTM lines, written and read.
//!
//! A diarizer's activity and its turns are on the stream's sample clock: a
//! turn's times are seconds from the first frame the filter saw.
pub use *;
pub use *;
pub use *;