assemblyai 4.35.0 → 4.35.4

package/dist/types/streaming/index.d.ts

@@ -87,7 +87,12 @@ 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;
     /**
@@ -104,7 +109,19 @@ export type StreamingTranscriberParams = {
     prompt?: string;
     agentContext?: string;
     speechModel?: StreamingSpeechModel;
+    /**
+     * @deprecated Use `languageCodes` instead (pass a single-element array, e.g. `["es"]`,
+     * for the same behavior). Still supported for backward compatibility.
+     */
     languageCode?: string;
+    /**
+     * Recommended way to select languages. Steers transcription toward a set of
+     * languages by biasing output toward them on a per-token basis while still
+     * allowing native code-switching among them. Pass the languages you expect
+     * (e.g. `["en", "es"]`), or a single-element array (e.g. `["es"]`) for a
+     * monolingual session. Universal-3.5 Pro Streaming only.
+     */
+    languageCodes?: string[];
     languageDetection?: boolean;
     domain?: StreamingDomain;
     inactivityTimeout?: number;
@@ -255,6 +272,12 @@ export type StreamingUpdateConfiguration = {
     filter_profanity?: boolean;
     interruption_delay?: number;
     turn_left_pad_ms?: number;
+    /**
+     * Steer transcription toward a set of languages mid-stream. Pass an empty
+     * array (`[]`) to clear steering and restore the model's default
+     * multilingual code-switching. Universal-3.5 Pro Streaming only.
+     */
+    language_codes?: string[];
 };
 export type StreamingForceEndpoint = {
     type: "ForceEndpoint";

package/dist/workerd.mjs

@@ -28,7 +28,7 @@ if (typeof navigator !== "undefined" && navigator.userAgent) {
     defaultUserAgentString += navigator.userAgent;
 }
 const defaultUserAgent = {
-    sdk: { name: "JavaScript", version: "4.35.0" },
+    sdk: { name: "JavaScript", version: "4.35.4" },
 };
 if (typeof process !== "undefined") {
     if (process.versions.node && defaultUserAgentString.indexOf("Node") === -1) {
@@ -1036,6 +1036,10 @@ class StreamingTranscriber {
         if (!(this.token || this.apiKey)) {
             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("StreamingTranscriber.channels must have exactly 2 entries.");
@@ -1061,10 +1065,12 @@ class StreamingTranscriber {
                 "speaker-history") {
                 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));
-            this.minChunkSamples = Math.max(1, Math.round(params.sampleRate * (MIN_CHUNK_MS / 1000)));
-            this.maxChunkSamples = Math.max(this.minChunkSamples, Math.round(params.sampleRate * (MAX_CHUNK_MS / 1000)));
+            // 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(sampleRate * (MIN_CHUNK_MS / 1000)));
+            this.maxChunkSamples = Math.max(this.minChunkSamples, Math.round(sampleRate * (MAX_CHUNK_MS / 1000)));
             this.channelBuffers = new Map(names.map((n) => [n, []]));
             this.channelSamplesReceived = new Map(names.map((n) => [n, 0]));
             this.channelVadFloatBuffers = new Map(names.map((n) => [n, new Float32Array(this.vadFrameSamples)]));
@@ -1082,7 +1088,9 @@ class StreamingTranscriber {
         if (this.token) {
             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("end_of_turn_confidence_threshold", this.params.endOfTurnConfidenceThreshold.toString());
         }
@@ -1134,8 +1142,12 @@ class StreamingTranscriber {
             searchParams.set("speech_model", this.params.speechModel.toString());
         }
         if (this.params.languageCode !== undefined) {
+            console.warn("[Deprecation Warning] `languageCode` is deprecated and will be removed in a future release. Please use `languageCodes` instead.");
             searchParams.set("language_code", this.params.languageCode);
         }
+        if (this.params.languageCodes !== undefined) {
+            searchParams.set("language_codes", JSON.stringify(this.params.languageCodes));
+        }
         if (this.params.languageDetection !== undefined) {
             searchParams.set("language_detection", this.params.languageDetection.toString());
         }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "assemblyai",
-  "version": "4.35.0",
+  "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"

package/src/exports/streaming.ts

@@ -1,7 +1,7 @@
 export * from "../types/asyncapi.generated";
 export * from "../types/realtime";
 export * from "../types/helpers";
-export * from "../types/streaming/dual-channel";
+export * from "../types/streaming";
 export * from "../services/realtime/service";
 export * from "../services/streaming/service";
 export * from "../services/streaming/factory";

package/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(
@@ -313,9 +324,19 @@ export class StreamingTranscriber {
     }
 
     if (this.params.languageCode !== undefined) {
+      console.warn(
+        "[Deprecation Warning] `languageCode` is deprecated and will be removed in a future release. Please use `languageCodes` instead.",
+      );
       searchParams.set("language_code", this.params.languageCode);
     }
 
+    if (this.params.languageCodes !== undefined) {
+      searchParams.set(
+        "language_codes",
+        JSON.stringify(this.params.languageCodes),
+      );
+    }
+
     if (this.params.languageDetection !== undefined) {
       searchParams.set(
         "language_detection",
@@ -743,7 +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++) {

package/src/types/asyncapi.generated.ts

@@ -25,7 +25,7 @@ export type AudioData = ArrayBufferLike;
  * The encoding of the audio data
  * @defaultValue "pcm_s16"le
  */
-export type AudioEncoding = "pcm_s16le" | "pcm_mulaw";
+export type AudioEncoding = "pcm_s16le" | "pcm_mulaw" | "opus" | "ogg_opus";
 
 /**
  * Configure the threshold for how long to wait before ending an utterance. Default is 700ms.

package/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;
   /**
@@ -109,7 +114,19 @@ export type StreamingTranscriberParams = {
   prompt?: string;
   agentContext?: string;
   speechModel?: StreamingSpeechModel;
+  /**
+   * @deprecated Use `languageCodes` instead (pass a single-element array, e.g. `["es"]`,
+   * for the same behavior). Still supported for backward compatibility.
+   */
   languageCode?: string;
+  /**
+   * Recommended way to select languages. Steers transcription toward a set of
+   * languages by biasing output toward them on a per-token basis while still
+   * allowing native code-switching among them. Pass the languages you expect
+   * (e.g. `["en", "es"]`), or a single-element array (e.g. `["es"]`) for a
+   * monolingual session. Universal-3.5 Pro Streaming only.
+   */
+  languageCodes?: string[];
   languageDetection?: boolean;
   domain?: StreamingDomain;
   inactivityTimeout?: number;
@@ -359,6 +376,12 @@ export type StreamingUpdateConfiguration = {
   filter_profanity?: boolean;
   interruption_delay?: number;
   turn_left_pad_ms?: number;
+  /**
+   * Steer transcription toward a set of languages mid-stream. Pass an empty
+   * array (`[]`) to clear steering and restore the model's default
+   * multilingual code-switching. Universal-3.5 Pro Streaming only.
+   */
+  language_codes?: string[];
 };
 
 export type StreamingForceEndpoint = {