feat(streaming): add speakerLabelsRevisionIntervalMs param

CHANGELOG.md

@@ -1,5 +1,9 @@
 # Changelog
 
+## [4.41.5]
+
+- Add `speakerLabelsRevisionIntervalMs` streaming param (sent as `speaker_labels_revision_interval_ms`) — the cadence, in ms of audio time, at which the server emits mid-stream `speakerRevision` events when `speakerLabels` is enabled. Unset or `0` sends only the end-of-stream revision; non-zero values are clamped server-side to 120 000–300 000 ms. The first mid-stream revision needs ~120 s of streamed speech, and one is sent only when some earlier turn's label changed
+
 ## [4.41.4]
 
 - Fix `RealtimeTranscriber.connect()` promise hanging indefinitely when a socket error or unexpected close occurs before `SessionBegins`. The promise now rejects, and the socket reference is discarded so that connections can be safely retried.

CLAUDE.md

@@ -103,7 +103,11 @@ when `speakerLabels` is enabled). Sent once per offline-recluster resolve; each
 message batches every earlier Turn whose speaker label changed (unchanged turns
 are omitted). For each `revisions[i]`, match `turn_order` against the original
 Turn and replace its per-word `speaker` (and the turn-level `speaker_label`)
-with the revision's values — text and word timestamps are unchanged:
+with the revision's values — text and word timestamps are unchanged.
+By default only the end-of-stream revision is sent; set
+`speakerLabelsRevisionIntervalMs` (ms of audio time, clamped server-side to
+120 000–300 000) to also get them mid-stream — the first can only arrive after
+~120 s of speech, and one is sent only when a label actually changed:
 
 ```typescript
 transcriber.on("speakerRevision", (event) => {

package.json

@@ -1,6 +1,6 @@
 {
   "name": "assemblyai",
-  "version": "4.41.4",
+  "version": "4.41.5",
   "description": "The AssemblyAI JavaScript SDK provides an easy-to-use interface for interacting with the AssemblyAI API, which supports async and real-time transcription.",
   "engines": {
     "node": ">=18"

src/services/streaming/service.ts

@@ -406,6 +406,13 @@ export class StreamingTranscriber {
       searchParams.set("max_speakers", this.params.maxSpeakers.toString());
     }
 
+    if (this.params.speakerLabelsRevisionIntervalMs !== undefined) {
+      searchParams.set(
+        "speaker_labels_revision_interval_ms",
+        this.params.speakerLabelsRevisionIntervalMs.toString(),
+      );
+    }
+
     if (this.params.voiceFocus) {
       searchParams.set("voice_focus", this.params.voiceFocus);
     }

src/types/streaming/index.ts

@@ -139,6 +139,19 @@ export type StreamingTranscriberParams = {
   inactivityTimeout?: number;
   speakerLabels?: boolean;
   maxSpeakers?: number;
+  /**
+   * Cadence, in milliseconds of audio time, at which the server emits
+   * mid-stream `speakerRevision` events with corrected speaker labels for
+   * earlier turns (plus a final one at end of stream). Unset or `0` sends no
+   * mid-stream revisions — only the end-of-stream one. Non-zero values are
+   * clamped server-side to 120 000–300 000 ms; the server rejects values
+   * outside 0–86 400 000 (24 h). Only used when `speakerLabels` is enabled.
+   *
+   * The first mid-stream revision can only arrive after ~120 s of speech has
+   * been streamed, and a revision is sent only when some earlier turn's label
+   * actually changed — so expect "at most one per interval", not exactly one.
+   */
+  speakerLabelsRevisionIntervalMs?: number;
   voiceFocus?: VoiceFocusModel;
   voiceFocusThreshold?: number;
   continuousPartials?: boolean;

tests/unit/streaming.test.ts

@@ -127,6 +127,44 @@ describe("streaming", () => {
     await connect(rt, server);
   });
 
+  it("should include speaker_labels_revision_interval_ms in connection URL", async () => {
+    await cleanup();
+    WS.clean();
+
+    const wsUrl = `${websocketBaseUrl}?token=123&sample_rate=16000&speech_model=universal-streaming-english&speaker_labels=true&speaker_labels_revision_interval_ms=120000`;
+    server = new WS(wsUrl);
+    rt = new StreamingTranscriber({
+      websocketBaseUrl,
+      token: "123",
+      sampleRate: 16_000,
+      speechModel: "universal-streaming-english",
+      speakerLabels: true,
+      speakerLabelsRevisionIntervalMs: 120_000,
+    });
+    onOpen = jest.fn();
+    rt.on("open", onOpen);
+    await connect(rt, server);
+  });
+
+  it("should send speaker_labels_revision_interval_ms=0 when explicitly set to 0", async () => {
+    await cleanup();
+    WS.clean();
+
+    const wsUrl = `${websocketBaseUrl}?token=123&sample_rate=16000&speech_model=universal-streaming-english&speaker_labels=true&speaker_labels_revision_interval_ms=0`;
+    server = new WS(wsUrl);
+    rt = new StreamingTranscriber({
+      websocketBaseUrl,
+      token: "123",
+      sampleRate: 16_000,
+      speechModel: "universal-streaming-english",
+      speakerLabels: true,
+      speakerLabelsRevisionIntervalMs: 0,
+    });
+    onOpen = jest.fn();
+    rt.on("open", onOpen);
+    await connect(rt, server);
+  });
+
   it("should normalize deprecated minEndOfTurnSilenceWhenConfident to min_turn_silence in connection URL", async () => {
     await cleanup();
     WS.clean();