Refactor the speaker detection logic into observeSpeaker and add tests (#2814)

* Refactor the speaker detection logic into observeSpeaker and add tests

@robintown the tests pass, but some of the values were off by 1ms from what I was expecting. Please can you sanity check them?

* Extra test cases and clean up

* Make distinctUntilChanged part of the observable itself

* More suggestions from code review

src/state/observeSpeaker.ts

@@ -0,0 +1,36 @@
+/*
+Copyright 2024 New Vector Ltd.
+
+SPDX-License-Identifier: AGPL-3.0-only
+Please see LICENSE in the repository root for full details.
+*/
+import {
+  Observable,
+  audit,
+  merge,
+  timer,
+  filter,
+  startWith,
+  distinctUntilChanged,
+} from "rxjs";
+
+/**
+ * Require 1 second of continuous speaking to become a speaker, and 60 second of
+ * continuous silence to stop being considered a speaker
+ */
+export function observeSpeaker(
+  isSpeakingObservable: Observable<boolean>,
+): Observable<boolean> {
+  const distinct = isSpeakingObservable.pipe(distinctUntilChanged());
+
+  return distinct.pipe(
+    // Either change to the new value after the timer or re-emit the same value if it toggles back
+    // (audit will return the latest (toggled back) value) before the timeout.
+    audit((s) =>
+      merge(timer(s ? 1000 : 60000), distinct.pipe(filter((s1) => s1 !== s))),
+    ),
+    // Filter the re-emissions (marked as: | ) that happen if we toggle quickly (<1s) from false->true->false|->..
+    startWith(false),
+    distinctUntilChanged(),
+  );
+}