@drax/ai-back 3.42.0 → 3.43.0

package/test/DraxAgent.test.ts

@@ -144,6 +144,70 @@ describe("DraxAgent", () => {
         ]);
     });
 
+    test("passes audioResponse to provider and returns prompt audio", async () => {
+        class AudioMockProvider extends MockProvider {
+            async prompt(input: IPromptParams): Promise<IPromptResponse> {
+                this.requests.push(input);
+
+                return {
+                    output: "respuesta hablada",
+                    tokens: 10,
+                    inputTokens: 6,
+                    outputTokens: 4,
+                    time: 12,
+                    audio: {
+                        audio: Buffer.from("audio").toString("base64"),
+                        contentType: "audio/mpeg",
+                        encoding: "base64",
+                        meta: {
+                            provider: "elevenlabs",
+                            model: "eleven_multilingual_v2",
+                            voiceId: "voice-1",
+                            size: 5,
+                            time: 20,
+                        },
+                    },
+                };
+            }
+        }
+
+        const provider = new AudioMockProvider();
+        const agent = new DraxAgent().configure({
+            provider,
+            systemPrompt: "Sos un asistente.",
+            sessionService: false,
+            toolBuilders: undefined,
+            tools: undefined,
+        });
+
+        const response = await agent.sendMessage({
+            userId: "user-1",
+            message: "Hola",
+            audioResponse: {
+                provider: "ElevenLabs",
+                languageCode: "es",
+            },
+        });
+
+        expect(provider.requests[0].audioResponse).toEqual({
+            provider: "ElevenLabs",
+            languageCode: "es",
+        });
+        expect(response.message).toBe("respuesta hablada");
+        expect(response.audio).toEqual({
+            audio: Buffer.from("audio").toString("base64"),
+            contentType: "audio/mpeg",
+            encoding: "base64",
+            meta: {
+                provider: "elevenlabs",
+                model: "eleven_multilingual_v2",
+                voiceId: "voice-1",
+                size: 5,
+                time: 20,
+            },
+        });
+    });
+
     test("returns a navigation path from tool execution metadata", async () => {
         const provider = new MockProvider();
         const agent = new DraxAgent().configure({

package/test/PromptAudioService.test.ts

@@ -0,0 +1,115 @@
+import {afterEach, describe, expect, test, vi} from "vitest";
+import {PromptAudioService, TTSProviderFactory} from "../src";
+
+describe("PromptAudioService Test", () => {
+
+    afterEach(() => {
+        vi.restoreAllMocks()
+    })
+
+    test("does not generate audio when audioResponse is not requested", async () => {
+        const instanceSpy = vi.spyOn(TTSProviderFactory, "instance")
+
+        const audio = await PromptAudioService.build({
+            systemPrompt: "You are an assistant.",
+        }, "Hola")
+
+        expect(audio).toBeUndefined()
+        expect(instanceSpy).not.toHaveBeenCalled()
+    })
+
+    test("generates base64 prompt audio with default ElevenLabs provider", async () => {
+        const textToSpeech = vi.fn(async () => ({
+            audio: Buffer.from("audio-bytes"),
+            contentType: "audio/mpeg",
+            size: 11,
+            time: 25,
+            provider: "elevenlabs",
+            model: "eleven_multilingual_v2",
+            voiceId: "voice-1",
+            outputFormat: "mp3_44100_128",
+        }))
+        const instanceSpy = vi.spyOn(TTSProviderFactory, "instance").mockReturnValue({
+            textToSpeech,
+        })
+
+        const audio = await PromptAudioService.build({
+            systemPrompt: "You are an assistant.",
+            audioResponse: true,
+            operationTitle: "prompt-title",
+            operationGroup: "prompt-group",
+            ip: "127.0.0.1",
+            userAgent: "vitest",
+            tenant: "tenant-1",
+            user: "user-1",
+        }, "Hola mundo")
+
+        expect(instanceSpy).toHaveBeenCalledWith("ElevenLabs")
+        expect(textToSpeech).toHaveBeenCalledWith({
+            text: "Hola mundo",
+            voiceId: undefined,
+            model: undefined,
+            outputFormat: undefined,
+            voiceSettings: undefined,
+            previousText: undefined,
+            nextText: undefined,
+            languageCode: undefined,
+            seed: undefined,
+            operationTitle: "prompt-title",
+            operationGroup: "prompt-group",
+            ip: "127.0.0.1",
+            userAgent: "vitest",
+            tenant: "tenant-1",
+            user: "user-1",
+        })
+        expect(audio).toEqual({
+            audio: Buffer.from("audio-bytes").toString("base64"),
+            contentType: "audio/mpeg",
+            encoding: "base64",
+            meta: {
+                provider: "elevenlabs",
+                model: "eleven_multilingual_v2",
+                voiceId: "voice-1",
+                outputFormat: "mp3_44100_128",
+                size: 11,
+                time: 25,
+            },
+        })
+    })
+
+    test("supports custom TTS provider params", async () => {
+        const textToSpeech = vi.fn(async () => ({
+            audio: Buffer.from("custom-audio"),
+            contentType: "audio/ogg",
+            size: 12,
+            time: 31,
+            provider: "custom",
+            model: "custom-model",
+            voiceId: "voice-2",
+        }))
+        const instanceSpy = vi.spyOn(TTSProviderFactory, "instance").mockReturnValue({
+            textToSpeech,
+        })
+
+        await PromptAudioService.build({
+            systemPrompt: "You are an assistant.",
+            audioResponse: {
+                provider: "CustomTTS",
+                voiceId: "voice-2",
+                model: "custom-model",
+                languageCode: "es",
+                operationTitle: "tts-title",
+            },
+        }, {message: "Hola"})
+
+        expect(instanceSpy).toHaveBeenCalledWith("CustomTTS")
+        expect(textToSpeech).toHaveBeenCalledWith(expect.objectContaining({
+            text: "{\"message\":\"Hola\"}",
+            voiceId: "voice-2",
+            model: "custom-model",
+            languageCode: "es",
+            operationTitle: "tts-title",
+        }))
+    })
+
+})