fix(frontend): start voice sessions without update acknowledgment

s2s-frontend/README.md

@@ -147,6 +147,13 @@ credentials, and ApiPilot execution. Browser `SpeechSynthesis` is used only
 when the configured external reasoning service returns an answer instead of
 `log_only`.
 
+Session startup waits for audio initialization and `session.created`, then sends
+`session.update` before microphone audio or the initial greeting. The speech
+server processes those frames in order and does not emit `session.updated` on
+success. That acknowledgment is therefore optional; receiving it later does not
+restart the greeting or reset an active response. The startup timeout still
+closes the microphone and socket if the server never creates a session.
+
 ## References
 
 - [Vite getting started](https://vite.dev/guide/)

s2s-frontend/src/realtime/client.test.ts

@@ -1,9 +1,11 @@
 import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
 
+import type { AudioEngine } from "../audio/engine";
+
 import { RealtimeClient, SessionError, type ClientOptions } from "./client";
 
 const audio = vi.hoisted(() => ({
-  start: vi.fn<(...callbacks: unknown[]) => Promise<void>>(),
+  start: vi.fn<AudioEngine["start"]>(),
   close: vi.fn<() => Promise<void>>(),
   play: vi.fn(),
   clear: vi.fn(),
@@ -68,7 +70,6 @@ async function connect(client: RealtimeClient, pending = client.connect()) {
   await vi.waitFor(() => {
     expect(socket.sent[0]?.type).toBe("session.update");
   });
-  socket.receive({ type: "session.updated" });
   await pending;
   return socket;
 }
@@ -88,6 +89,59 @@ afterEach(() => {
 });
 
 describe("realtime session", () => {
+  it("starts microphone streaming without a session.updated acknowledgment", async () => {
+    const client = new RealtimeClient(options());
+    const pending = client.connect();
+    const settled = pending.catch(() => undefined);
+    try {
+      await vi.waitFor(() => {
+        expect(audio.start).toHaveBeenCalledOnce();
+      });
+      const socket = Socket.instances[0];
+      const capture = audio.start.mock.calls[0]?.[0];
+      if (!socket || !capture) {
+        throw new Error("Missing socket or microphone callback");
+      }
+      const pcm = new ArrayBuffer(2);
+      capture(pcm);
+      expect(socket.sent).toHaveLength(0);
+
+      socket.receive({ type: "session.created" });
+      await vi.waitFor(() => {
+        expect(client.status).toBe("listening");
+      });
+      await expect(pending).resolves.toBeUndefined();
+      capture(pcm);
+      expect(socket.sent).toEqual([
+        expect.objectContaining({ type: "session.update" }),
+        { type: "input_audio_buffer.append", audio: "AAA=" },
+      ]);
+      await vi.advanceTimersByTimeAsync(21000);
+      expect(client.status).toBe("listening");
+      expect(socket.close).not.toHaveBeenCalled();
+    } finally {
+      await client.close();
+      await settled;
+    }
+    expect(audio.close).toHaveBeenCalledOnce();
+    expect(vi.getTimerCount()).toBe(0);
+  });
+
+  it("times out and releases the microphone when the server never creates a session", async () => {
+    const client = new RealtimeClient(options());
+    const rejected = expect(client.connect()).rejects.toThrow("El servidor de voz no respondió a tiempo.");
+    await vi.waitFor(() => {
+      expect(audio.start).toHaveBeenCalledOnce();
+    });
+    await vi.advanceTimersByTimeAsync(20000);
+    await rejected;
+    expect(Socket.instances[0]?.sent).toHaveLength(0);
+    expect(Socket.instances[0]?.close).toHaveBeenCalledExactlyOnceWith(1000, "client closed");
+    expect(audio.close).toHaveBeenCalledOnce();
+    expect(client.status).toBe("closed");
+    expect(vi.getTimerCount()).toBe(0);
+  });
+
   it("connects with a plain load-balancer grant and closes without further HTTP requests", async () => {
     const fetchMock = vi.fn().mockResolvedValueOnce(
       new Response(
@@ -263,7 +317,7 @@ describe("realtime session", () => {
     expect(socket.sent.at(-1)).toMatchObject({ item: { role: "assistant", content: [{ text: "Respuesta externa" }] } });
     await client.close();
   });
-  it("waits for configuration before greeting and serializes response requests", async () => {
+  it("sends configuration before greeting without an acknowledgment and serializes response requests", async () => {
     const client = new RealtimeClient({ ...options(), greeting: "hola vocero" });
     const socket = await connect(client);
     expect(socket.sent.map((event) => event.type)).toEqual([
@@ -275,6 +329,12 @@ describe("realtime session", () => {
     await vi.waitFor(() => {
       expect(client.status).toBe("processing");
     });
+    const sentCount = socket.sent.length;
+    socket.receive({ type: "session.updated" });
+    socket.receive({ type: "session.updated" });
+    await vi.advanceTimersByTimeAsync(0);
+    expect(client.status).toBe("processing");
+    expect(socket.sent).toHaveLength(sentCount);
     client.sendUserText("Otra pregunta");
     client.requestResponse();
     expect(socket.sent.filter((event) => event.type === "response.create")).toHaveLength(1);

s2s-frontend/src/realtime/client.ts

@@ -320,8 +320,8 @@ export class RealtimeClient {
             tool_choice: this.options.tools.length ? "auto" : "none",
           },
         });
-        break;
-      case "session.updated":
+        // The speech server applies frames in order without acknowledging session.update.
+        // Send configuration before microphone audio or the greeting, then consider it ready.
         if (!this.configured) {
           this.configured = true;
           clearTimeout(this.readyTimer);