feat(voice): ✨ Update voice protocol implementation for new encoding/decoding support and improved audio quality

Co-Authored-By: Lilith Autocommit <noreply@atlilith.com>

@packages/companion-client/src/voice/binary-protocol.ts

@@ -17,8 +17,11 @@ export function encodeUpstreamFrame(frame: AudioUpstreamFrame): ArrayBuffer {
   view.setUint8(0, UPSTREAM_FRAME_TYPE);
   view.setUint32(1, frame.seq, false); // big-endian
 
-  const dst = new Int16Array(buffer, UPSTREAM_HEADER_BYTES);
-  dst.set(frame.pcm);
+  // Int16Array requires 2-byte-aligned offsets; UPSTREAM_HEADER_BYTES=5 is odd.
+  // Copy PCM bytes via Uint8Array instead.
+  new Uint8Array(buffer, UPSTREAM_HEADER_BYTES).set(
+    new Uint8Array(frame.pcm.buffer, frame.pcm.byteOffset, frame.pcm.byteLength),
+  );
 
   return buffer;
 }

@packages/companion-client/src/voice/voice-client.ts

@@ -76,7 +76,8 @@ export class VoiceClient {
 
     const socket = io(`${baseUrl}/voice`, {
       query: { session_id: this.config.sessionId },
-      transports: ['websocket'],
+      // Default transports: polling first (works through HTTP proxies like Vite),
+      // then upgrades to WebSocket automatically.
       reconnection: this.config.reconnect,
       reconnectionAttempts: this.config.maxReconnectAttempts,
       reconnectionDelay: this.config.reconnectDelay,