feat: support nova-3 and keyterms in stt

Author: naomi-lgbt

examples/node-live/index.js

@@ -10,7 +10,7 @@ const live = async () => {
   let is_finals = [];
 
   const connection = deepgram.listen.live({
-    model: "nova-2",
+    model: "nova-3",
     language: "en-US",
     // Apply smart formatting to the output
     smart_format: true,
@@ -20,6 +20,8 @@ const live = async () => {
     vad_events: true,
     // Time in milliseconds of silence to wait for before finalizing speech
     endpointing: 300,
+    // Keyterm Prompting allows you improve Keyword Recall Rate (KRR) for important keyterms or phrases up to 90%.
+    keyterms: ["BBC"],
   });
 
   connection.on(LiveTranscriptionEvents.Open, () => {

examples/node-prerecorded/index.js

@@ -11,7 +11,8 @@ const transcribeUrl = async () => {
       url: "https://dpgr.am/spacewalk.wav",
     },
     {
-      model: "nova-2",
+      model: "nova-3",
+      keyterms: ["spacewalk"],
     }
   );
 
@@ -28,7 +29,8 @@ const transcribeFile = async () => {
 
   console.log("Transcribing file", file);
   const { result, error } = await deepgram.listen.prerecorded.transcribeFile(file, {
-    model: "nova-2",
+    model: "nova-3",
+    keyterms: ["spacewalk"],
   });
 
   if (error) console.error(error);

src/lib/types/TranscriptionSchema.ts

@@ -94,6 +94,11 @@ interface TranscriptionSchema extends Record<string, unknown> {
    */
   keywords?: string[] | string;
 
+  /**
+   * @see https://developers.deepgram.com/docs/keyterm
+   */
+  keyterms?: string[] | string;
+
   /**
    * @see https://developers.deepgram.com/docs/tagging
    */

src/packages/ListenLiveClient.ts

@@ -1,6 +1,7 @@
 import { AbstractLiveClient } from "./AbstractLiveClient";
 import { LiveTranscriptionEvents } from "../lib/enums";
 import type { LiveSchema, LiveConfigOptions, DeepgramClientOptions } from "../lib/types";
+import { DeepgramError } from "../lib/errors";
 
 /**
  * The `ListenLiveClient` class extends the `AbstractLiveClient` class and provides functionality for setting up and managing a WebSocket connection for live transcription.
@@ -34,6 +35,13 @@ export class ListenLiveClient extends AbstractLiveClient {
   ) {
     super(options);
 
+    if (
+      transcriptionOptions.keyterms?.length &&
+      !transcriptionOptions.model?.startsWith("nova-3")
+    ) {
+      throw new DeepgramError("Keyterms are only supported with the Nova 3 models.");
+    }
+
     this.connect(transcriptionOptions, endpoint);
   }
 

src/packages/ListenRestClient.ts

@@ -52,6 +52,10 @@ export class ListenRestClient extends AbstractRestClient {
         );
       }
 
+      if (options?.keyterms?.length && !options.model?.startsWith("nova-3")) {
+        throw new DeepgramError("Keyterms are only supported with the Nova 3 models.");
+      }
+
       const requestUrl = this.getRequestUrl(endpoint, {}, { ...{}, ...options });
       const result: SyncPrerecordedResponse = await this.post(requestUrl, body).then((result) =>
         result.json()