feat(realtime): add usage field to input audio transcription completed event

Author: Max

packages/agents-realtime/src/openaiRealtimeEvents.ts

@@ -138,6 +138,18 @@ export const conversationItemInputAudioTranscriptionCompletedEventSchema =
     content_index: z.number(),
     transcript: z.string(),
     logprobs: z.array(z.any()).nullable().optional(),
+    usage: z
+      .object({
+        type: z.literal('tokens'),
+        total_tokens: z.number(),
+        input_tokens: z.number(),
+        input_token_details: z.object({
+          text_tokens: z.number(),
+          audio_tokens: z.number(),
+        }),
+        output_tokens: z.number(),
+      })
+      .optional(),
   });
 
 export const conversationItemInputAudioTranscriptionDeltaEventSchema = z.object(

packages/agents-realtime/src/transportLayerEvents.ts

@@ -44,6 +44,16 @@ export type InputAudioTranscriptionCompletedEvent = {
   type: 'conversation.item.input_audio_transcription.completed';
   item_id: string;
   transcript: string;
+  usage?: {
+    type: 'tokens';
+    total_tokens: number;
+    input_tokens: number;
+    input_token_details: {
+      text_tokens: number;
+      audio_tokens: number;
+    };
+    output_tokens: number;
+  };
 };
 
 export type TransportLayerTranscriptDelta = {