feat: implement tier-a4 tts voice output replies

2026-02-18 10:22:28 -08:00
parent 3eb07875f1
commit a71aa5992d
11 changed files with 482 additions and 4 deletions
@@ -1122,6 +1122,156 @@ describe('daemon audio routing integration', () => {
  });
 });

+describe('daemon tts routing integration', () => {
+  afterEach(() => {
+    vi.restoreAllMocks();
+  });
+
+  it('attaches synthesized audio reply when tts is enabled for the channel', async () => {
+    const processSpy = vi.spyOn(AgentOrchestrator.prototype, 'process').mockResolvedValue('voice-enabled response');
+    const fetchSpy = vi.spyOn(globalThis, 'fetch').mockResolvedValue({
+      ok: true,
+      status: 200,
+      statusText: 'OK',
+      arrayBuffer: async () => Uint8Array.from([7, 8, 9]).buffer,
+    } as Response);
+
+    const session = {
+      id: 'telegram:tts-user-1',
+      addMessage: vi.fn(),
+      getHistory: vi.fn(() => []),
+      clear: vi.fn(),
+      replaceHistory: vi.fn(),
+      getConfig: vi.fn(() => undefined),
+      setConfig: vi.fn(),
+      deleteConfig: vi.fn(),
+    };
+
+    const router = createMessageRouter({
+      sessionManager: { getSession: vi.fn(() => session) } as unknown as MessageRouterDeps['sessionManager'],
+      modelRouter: {
+        getAvailableTiers: () => ['default'],
+        getAllLabels: () => ({ default: 'default' }),
+        getLabel: (tier: string) => tier,
+      } as unknown as MessageRouterDeps['modelRouter'],
+      systemPrompt: 'test prompt',
+      toolRegistry: { clone() { return this; }, register: vi.fn() } as unknown as MessageRouterDeps['toolRegistry'],
+      toolExecutor: {} as unknown as MessageRouterDeps['toolExecutor'],
+      config: {
+        agents: {
+          primary_tier: 'default',
+          delegation: {
+            compaction: 'default',
+            memory_extraction: 'default',
+            classification: 'default',
+            tool_summarisation: 'default',
+            complex_reasoning: 'default',
+          },
+          max_delegation_depth: 1,
+          max_iterations: 3,
+        },
+        compaction: { enabled: false },
+        models: { default: { provider: 'anthropic', model: 'claude' } },
+        tts: {
+          enabled: true,
+          enabled_channels: ['telegram'],
+          provider: {
+            type: 'custom',
+            endpoint: 'https://example.com/v1/audio/speech',
+            api_key: 'sk-test',
+            model: 'gpt-4o-mini-tts',
+            voice: 'alloy',
+            format: 'mp3',
+          },
+        },
+      } as unknown as MessageRouterDeps['config'],
+    });
+
+    const reply = vi.fn(async (_message: OutboundMessage) => {});
+    await router.handler({
+      id: 'tts-1',
+      channel: 'telegram',
+      senderId: 'tts-user-1',
+      text: 'say hello',
+      timestamp: Date.now(),
+    } as MessageRouterInput, reply);
+
+    expect(processSpy).toHaveBeenCalledTimes(1);
+    expect(fetchSpy).toHaveBeenCalledTimes(1);
+    const outbound = reply.mock.calls[0]?.[0] as OutboundMessage | undefined;
+    expect(outbound?.attachments).toBeDefined();
+    expect(outbound?.attachments?.[0]).toMatchObject({
+      mimeType: 'audio/mpeg',
+      data: 'BwgJ',
+    });
+  });
+
+  it('does not synthesize tts when channel is not enabled', async () => {
+    vi.spyOn(AgentOrchestrator.prototype, 'process').mockResolvedValue('text-only response');
+    const fetchSpy = vi.spyOn(globalThis, 'fetch');
+
+    const session = {
+      id: 'discord:tts-user-2',
+      addMessage: vi.fn(),
+      getHistory: vi.fn(() => []),
+      clear: vi.fn(),
+      replaceHistory: vi.fn(),
+      getConfig: vi.fn(() => undefined),
+      setConfig: vi.fn(),
+      deleteConfig: vi.fn(),
+    };
+
+    const router = createMessageRouter({
+      sessionManager: { getSession: vi.fn(() => session) } as unknown as MessageRouterDeps['sessionManager'],
+      modelRouter: {
+        getAvailableTiers: () => ['default'],
+        getAllLabels: () => ({ default: 'default' }),
+        getLabel: (tier: string) => tier,
+      } as unknown as MessageRouterDeps['modelRouter'],
+      systemPrompt: 'test prompt',
+      toolRegistry: { clone() { return this; }, register: vi.fn() } as unknown as MessageRouterDeps['toolRegistry'],
+      toolExecutor: {} as unknown as MessageRouterDeps['toolExecutor'],
+      config: {
+        agents: {
+          primary_tier: 'default',
+          delegation: {
+            compaction: 'default',
+            memory_extraction: 'default',
+            classification: 'default',
+            tool_summarisation: 'default',
+            complex_reasoning: 'default',
+          },
+          max_delegation_depth: 1,
+          max_iterations: 3,
+        },
+        compaction: { enabled: false },
+        models: { default: { provider: 'anthropic', model: 'claude' } },
+        tts: {
+          enabled: true,
+          enabled_channels: ['telegram'],
+          provider: {
+            type: 'custom',
+            endpoint: 'https://example.com/v1/audio/speech',
+          },
+        },
+      } as unknown as MessageRouterDeps['config'],
+    });
+
+    const reply = vi.fn(async (_message: OutboundMessage) => {});
+    await router.handler({
+      id: 'tts-2',
+      channel: 'discord',
+      senderId: 'tts-user-2',
+      text: 'respond as text',
+      timestamp: Date.now(),
+    } as MessageRouterInput, reply);
+
+    expect(fetchSpy).not.toHaveBeenCalled();
+    const outbound = reply.mock.calls[0]?.[0] as OutboundMessage | undefined;
+    expect(outbound?.attachments).toBeUndefined();
+  });
+});
+
 describe('daemon auto-escalate integration', () => {
  afterEach(() => {
    vi.restoreAllMocks();