Enhance audio input/output configuration by adding deviceId support. …

packages/client/src/BaseConversation.ts

@@ -14,6 +14,7 @@ import {
   UserTranscriptionEvent,
 } from "./utils/events";
 import type { InputConfig } from "./utils/input";
+import { OutputConfig } from "./utils/output";
 
 export type Role = "user" | "ai";
 
@@ -28,12 +29,14 @@ export type Status =
 export type Options = SessionConfig &
   Callbacks &
   ClientToolsConfig &
-  InputConfig;
+  InputConfig &
+  OutputConfig;
 
 export type PartialOptions = SessionConfig &
   Partial<Callbacks> &
   Partial<ClientToolsConfig> &
-  Partial<InputConfig>;
+  Partial<InputConfig> &
+  Partial<OutputConfig>;
 
 export type ClientToolsConfig = {
   clientTools: Record<

packages/client/src/VoiceConversation.ts

@@ -33,17 +33,22 @@ export class VoiceConversation extends BaseConversation {
       // some browsers won't allow calling getSupportedConstraints or enumerateDevices
       // before getting approval for microphone access
       preliminaryInputStream = await navigator.mediaDevices.getUserMedia({
-        audio: true,
+        audio: fullOptions.deviceId ? { deviceId: fullOptions.deviceId } : true,
       });
 
       await applyDelay(fullOptions.connectionDelay);
       connection = await Connection.create(options);
       [input, output] = await Promise.all([
         Input.create({
           ...connection.inputFormat,
-          preferHeadphonesForIosDevices: options.preferHeadphonesForIosDevices,
+          preferHeadphonesForIosDevices:
+            fullOptions.preferHeadphonesForIosDevices,
+          deviceId: fullOptions.deviceId,
+        }),
+        Output.create({
+          ...connection.outputFormat,
+          outputDeviceId: fullOptions.outputDeviceId,
         }),
-        Output.create(connection.outputFormat),
       ]);
 
       preliminaryInputStream?.getTracks().forEach(track => track.stop());

packages/client/src/utils/input.ts

@@ -4,6 +4,7 @@ import { isIosDevice } from "./compatibility";
 
 export type InputConfig = {
   preferHeadphonesForIosDevices?: boolean;
+  deviceId?: string;
 };
 
 const LIBSAMPLERATE_JS =
@@ -14,6 +15,7 @@ export class Input {
     sampleRate,
     format,
     preferHeadphonesForIosDevices,
+    deviceId,
   }: FormatConfig & InputConfig): Promise<Input> {
     let context: AudioContext | null = null;
     let inputStream: MediaStream | null = null;
@@ -25,7 +27,9 @@ export class Input {
         noiseSuppression: { ideal: true },
       };
 
-      if (isIosDevice() && preferHeadphonesForIosDevices) {
+      if (deviceId) {
+        options.deviceId = deviceId;
+      } else if (isIosDevice() && preferHeadphonesForIosDevices) {
         const availableDevices =
           await window.navigator.mediaDevices.enumerateDevices();
         const idealDevice = availableDevices.find(

packages/client/src/utils/output.ts

@@ -1,17 +1,28 @@
 import { loadAudioConcatProcessor } from "./audioConcatProcessor";
 import { FormatConfig } from "./connection";
 
+export type OutputConfig = {
+  outputDeviceId?: string;
+};
+
 export class Output {
   public static async create({
     sampleRate,
     format,
-  }: FormatConfig): Promise<Output> {
+    outputDeviceId,
+  }: FormatConfig & OutputConfig): Promise<Output> {
     let context: AudioContext | null = null;
     try {
       context = new AudioContext({ sampleRate });
       const analyser = context.createAnalyser();
       const gain = context.createGain();
       gain.connect(analyser);
+
+      // Try to set the output device if specified and supported
+      if (outputDeviceId && "setSinkId" in AudioContext.prototype) {
+        await (context as any).setSinkId(outputDeviceId);
+      }
+
       analyser.connect(context.destination);
       await loadAudioConcatProcessor(context.audioWorklet);
       const worklet = new AudioWorkletNode(context, "audio-concat-processor");