Sync from internal repo (2026/07/07)

CHANGELOG.md

@@ -1,5 +1,9 @@
 # Changelog
 
+## [4.35.4]
+
+- `sampleRate` is now optional when `encoding` is `opus` or `ogg_opus` (the Opus stream is self-describing and the server ignores the value). It remains required for PCM encodings and for dual-channel mode; omitting it there now throws at construction time
+
 ## [4.35.3]
 
 - Allow `language_codes` in `updateConfiguration()` — re-steer the transcription language mid-stream without reconnecting; pass `[]` to clear steering and restore the model's default multilingual code-switching (Universal-3.5 Pro Streaming only)

package.json

@@ -1,6 +1,6 @@
 {
   "name": "assemblyai",
-  "version": "4.35.3",
+  "version": "4.35.4",
   "description": "The AssemblyAI JavaScript SDK provides an easy-to-use interface for interacting with the AssemblyAI API, which supports async and real-time transcription, as well as the latest LeMUR models.",
   "engines": {
     "node": ">=18"

src/services/streaming/service.ts

@@ -164,6 +164,13 @@ export class StreamingTranscriber {
       throw new Error("API key or temporary token is required.");
     }
 
+    const isOpus = params.encoding === "opus" || params.encoding === "ogg_opus";
+    if (params.sampleRate === undefined && (!isOpus || params.channels)) {
+      throw new Error(
+        '`sampleRate` is required; it may only be omitted when `encoding` is "opus" or "ogg_opus" (the Opus stream is self-describing) and dual-channel mode is not used.',
+      );
+    }
+
     if (params.channels) {
       if (params.channels.length !== 2) {
         throw new Error(
@@ -194,15 +201,17 @@ export class StreamingTranscriber {
       ) {
         this.speakerHistory = new Map();
       }
-      // 20 ms VAD frames at the transcriber's target sample rate.
-      this.vadFrameSamples = Math.max(1, Math.round(params.sampleRate * 0.02));
+      // 20 ms VAD frames at the transcriber's target sample rate. The
+      // constructor check above guarantees sampleRate in dual-channel mode.
+      const sampleRate = params.sampleRate!;
+      this.vadFrameSamples = Math.max(1, Math.round(sampleRate * 0.02));
       this.minChunkSamples = Math.max(
         1,
-        Math.round(params.sampleRate * (MIN_CHUNK_MS / 1000)),
+        Math.round(sampleRate * (MIN_CHUNK_MS / 1000)),
       );
       this.maxChunkSamples = Math.max(
         this.minChunkSamples,
-        Math.round(params.sampleRate * (MAX_CHUNK_MS / 1000)),
+        Math.round(sampleRate * (MAX_CHUNK_MS / 1000)),
       );
       this.channelBuffers = new Map(names.map((n) => [n, [] as number[]]));
       this.channelSamplesReceived = new Map(names.map((n) => [n, 0]));
@@ -230,7 +239,9 @@ export class StreamingTranscriber {
       searchParams.set("token", this.token);
     }
 
-    searchParams.set("sample_rate", this.params.sampleRate.toString());
+    if (this.params.sampleRate !== undefined) {
+      searchParams.set("sample_rate", this.params.sampleRate.toString());
+    }
 
     if (this.params.endOfTurnConfidenceThreshold) {
       searchParams.set(
@@ -753,7 +764,7 @@ Learn more at https://github.com/AssemblyAI/assemblyai-node-sdk/blob/main/docs/c
     let vadIdx = this.channelVadBufferIdx!.get(name)!;
     let received = this.channelSamplesReceived!.get(name)!;
     const vad = this.channelVads!.get(name)!;
-    const sampleRate = this.params.sampleRate;
+    const sampleRate = this.params.sampleRate!;
     const frameSize = this.vadFrameSamples;
 
     for (let i = 0; i < samples.length; i++) {

src/types/streaming/index.ts

@@ -92,7 +92,12 @@ export type StreamingTranscriberParams = {
    * Milliseconds to wait between connection attempts. Defaults to 500.
    */
   connectionRetryDelay?: number;
-  sampleRate: number;
+  /**
+   * Required for PCM encodings (and for dual-channel mode). May be omitted
+   * for Opus encodings (`opus`, `ogg_opus`) — the stream is self-describing
+   * and the server ignores the value.
+   */
+  sampleRate?: number;
   encoding?: AudioEncoding;
   endOfTurnConfidenceThreshold?: number;
   /**

tests/unit/streaming.test.ts

@@ -408,6 +408,55 @@ describe("streaming", () => {
     },
   );
 
+  it.each(["opus", "ogg_opus"] as const)(
+    "should allow omitting sampleRate for %s encoding",
+    async (encoding) => {
+      await cleanup();
+      WS.clean();
+
+      const wsUrl = `${websocketBaseUrl}?token=123&encoding=${encoding}`;
+      server = new WS(wsUrl);
+      rt = new StreamingTranscriber({
+        websocketBaseUrl,
+        token: "123",
+        encoding,
+      });
+      onOpen = jest.fn();
+      rt.on("open", onOpen);
+      await connect(rt, server);
+    },
+  );
+
+  it("should throw when sampleRate is omitted for non-Opus encodings", () => {
+    expect(
+      () =>
+        new StreamingTranscriber({
+          websocketBaseUrl,
+          token: "123",
+        }),
+    ).toThrow("`sampleRate` is required");
+    expect(
+      () =>
+        new StreamingTranscriber({
+          websocketBaseUrl,
+          token: "123",
+          encoding: "pcm_mulaw",
+        }),
+    ).toThrow("`sampleRate` is required");
+  });
+
+  it("should throw when sampleRate is omitted in dual-channel mode even with Opus", () => {
+    expect(
+      () =>
+        new StreamingTranscriber({
+          websocketBaseUrl,
+          token: "123",
+          encoding: "ogg_opus",
+          channels: [{ name: "mic" }, { name: "system" }],
+        }),
+    ).toThrow("`sampleRate` is required");
+  });
+
   it("should include whisper-rt speech model in connection URL", async () => {
     await cleanup();
     WS.clean();