evOo-local/packages/adapters/src/voice-factory.test.ts

import { afterEach, describe, expect, it, vi } from "vitest";
import { KOKORO_LOCAL_TOKEN, KokoroVoiceProvider } from "./kokoro-voice.js";
import {
  SCRIPTED_MPEG,
  SCRIPTED_TRANSCRIPT,
  SCRIPTED_VOICE_ID,
  ScriptedVoiceProvider,
} from "./scripted-voice.js";
import {
  createVoiceProvider,
  isVoiceProviderId,
  listVoiceCatalog,
  VOICE_CATALOG,
} from "./voice-factory.js";
import { MAX_SYNTHESIZED_AUDIO_BYTES } from "./voice-http.js";

const ctx = {
  operationId: "voice",
  traceId: "voice",
  spaceId: "w",
  userId: "u",
  signal: new AbortController().signal,
};

const previousRuntime = process.env.AGENT_RUNTIME;

afterEach(() => {
  vi.unstubAllGlobals();
  vi.unstubAllEnvs();
  if (previousRuntime === undefined) delete process.env.AGENT_RUNTIME;
  else process.env.AGENT_RUNTIME = previousRuntime;
});

describe("createVoiceProvider", () => {
  it("offers only the local voice server", () => {
    process.env.AGENT_RUNTIME = "pi";
    expect(VOICE_CATALOG.map((entry) => entry.id)).toEqual(["kokoro"]);
    expect(listVoiceCatalog().map((entry) => entry.id)).toEqual(["kokoro"]);
    expect(createVoiceProvider("kokoro").describe().id).toBe("kokoro");
    expect(createVoiceProvider("kokoro").describe().capabilities.transcribe).toBe(false);
    expect(isVoiceProviderId("kokoro")).toBe(true);
    expect(isVoiceProviderId("elevenlabs")).toBe(false);
    expect(() => createVoiceProvider("elevenlabs")).toThrow(/unknown voice provider/i);
    expect(isVoiceProviderId("scripted")).toBe(false);
    expect(isVoiceProviderId("piper")).toBe(false);
    expect(() => createVoiceProvider("piper")).toThrow(/unknown voice provider/i);
    expect(() => createVoiceProvider("scripted")).toThrow(/unknown voice provider/i);
  });

  it("adds the scripted fixture only when the agent runtime is scripted", () => {
    process.env.AGENT_RUNTIME = "scripted";
    expect(listVoiceCatalog().some((entry) => entry.id === "scripted")).toBe(true);
    expect(isVoiceProviderId("scripted")).toBe(true);
    expect(createVoiceProvider("scripted").describe().id).toBe("scripted");
  });
});

describe("KokoroVoiceProvider", () => {
  it("verifies against the local /v1/audio/voices endpoint", async () => {
    const fetchMock = vi.fn().mockResolvedValue({
      ok: true,
      json: async () => ({ voices: ["af_heart"] }),
    });
    vi.stubGlobal("fetch", fetchMock);
    const provider = new KokoroVoiceProvider();
    await expect(provider.verify(KOKORO_LOCAL_TOKEN, ctx)).resolves.toEqual({ ok: true });
    expect(String(fetchMock.mock.calls[0]?.[0])).toBe("http://127.0.0.1:8880/v1/audio/voices");
  });

  it("returns a helpful message when the local server is down", async () => {
    vi.stubGlobal("fetch", vi.fn().mockRejectedValue(new Error("fetch failed")));
    const provider = new KokoroVoiceProvider();
    const result = await provider.verify(KOKORO_LOCAL_TOKEN, ctx);
    expect(result.ok).toBe(false);
    expect(result.message).toMatch(/Lokaler Sprachdienst nicht erreichbar/i);
  });

  it("maps plain-string and object voice payloads", async () => {
    vi.stubGlobal(
      "fetch",
      vi.fn().mockResolvedValue(
        new Response(
          JSON.stringify({
            voices: ["af_heart", { id: "af_bella", name: "Bella" }],
          }),
        ),
      ),
    );
    const voices = await new KokoroVoiceProvider().listVoices(KOKORO_LOCAL_TOKEN, ctx);
    expect(voices.map((voice) => voice.id)).toEqual(["af_heart", "af_bella"]);
    expect(voices[0]?.label).toBe("Heart");
  });

  it("posts non-streaming speech to /v1/audio/speech", async () => {
    const fetchMock = vi.fn().mockResolvedValue({
      ok: true,
      arrayBuffer: async () => new Uint8Array([6, 7]).buffer,
    });
    vi.stubGlobal("fetch", fetchMock);
    const clip = await new KokoroVoiceProvider().synthesize(
      { text: "Hallo", voiceId: "af_heart", apiKey: KOKORO_LOCAL_TOKEN },
      ctx,
    );
    expect(clip.mimeType).toBe("audio/mpeg");
    expect([...clip.bytes]).toEqual([6, 7]);
    expect(String(fetchMock.mock.calls[0]?.[0])).toBe("http://127.0.0.1:8880/v1/audio/speech");
    const init = fetchMock.mock.calls[0]?.[1] as RequestInit;
    expect(JSON.parse(String(init.body))).toMatchObject({
      model: "kokoro",
      input: "Hallo",
      voice: "af_heart",
      response_format: "mp3",
      stream: false,
    });
  });

  it("honors KOKORO_BASE_URL", async () => {
    vi.stubEnv("KOKORO_BASE_URL", "http://kokoro.test:9000/");
    const fetchMock = vi.fn().mockResolvedValue({ ok: true, json: async () => ({ voices: [] }) });
    vi.stubGlobal("fetch", fetchMock);
    await new KokoroVoiceProvider().verify(KOKORO_LOCAL_TOKEN, ctx);
    expect(String(fetchMock.mock.calls[0]?.[0])).toBe("http://kokoro.test:9000/v1/audio/voices");
  });
});

describe("German voices through Piper", () => {
  function routeFetch(routes: Record<string, () => Response | Promise<Response>>) {
    const fetchMock = vi.fn(async (url: string | URL, _init?: RequestInit) => {
      const route = routes[String(url)];
      if (!route) throw new Error("fetch failed");
      return route();
    });
    vi.stubGlobal("fetch", fetchMock);
    return fetchMock;
  }

  it("lists the German voices whose Piper model is installed next to Kokoro's", async () => {
    vi.stubEnv("PIPER_BASE_URL", "http://piper.test:5150/");
    routeFetch({
      "http://127.0.0.1:8880/v1/audio/voices": () =>
        new Response(JSON.stringify({ voices: ["af_heart"] })),
      "http://piper.test:5150/voices": () =>
        new Response(JSON.stringify({ "de_DE-thorsten-high": {}, "en_US-lessac-medium": {} })),
    });
    const voices = await new KokoroVoiceProvider().listVoices(KOKORO_LOCAL_TOKEN, ctx);
    expect(voices.map((voice) => voice.id)).toEqual(["af_heart", "dm_thorsten"]);
    expect(voices[1]?.label).toBe("Thorsten");
  });

  it("keeps German voices when Kokoro is down and skips Piper when it is not configured", async () => {
    vi.stubEnv("PIPER_BASE_URL", "http://piper.test:5150");
    routeFetch({
      "http://piper.test:5150/voices": () =>
        new Response(JSON.stringify({ "de_DE-eva_k-x_low": {} })),
    });
    const voices = await new KokoroVoiceProvider().listVoices(KOKORO_LOCAL_TOKEN, ctx);
    expect(voices.map((voice) => voice.id)).toEqual(["df_eva"]);

    vi.stubEnv("PIPER_BASE_URL", "");
    const fetchMock = routeFetch({
      "http://127.0.0.1:8880/v1/audio/voices": () =>
        new Response(JSON.stringify({ voices: ["af_heart"] })),
    });
    await new KokoroVoiceProvider().listVoices(KOKORO_LOCAL_TOKEN, ctx);
    expect(fetchMock).toHaveBeenCalledTimes(1);
  });

  it("speaks a German voice through Piper's /synthesize as WAV", async () => {
    vi.stubEnv("PIPER_BASE_URL", "http://piper.test:5150");
    const fetchMock = routeFetch({
      "http://piper.test:5150/synthesize": () => new Response(new Uint8Array([0x52, 0x49])),
    });
    const clip = await new KokoroVoiceProvider().synthesize(
      { text: "Hallo", voiceId: "dm_thorsten", apiKey: KOKORO_LOCAL_TOKEN },
      ctx,
    );
    expect(clip.mimeType).toBe("audio/wav");
    expect([...clip.bytes]).toEqual([0x52, 0x49]);
    const init = fetchMock.mock.calls[0]?.[1] as RequestInit;
    expect(JSON.parse(String(init.body))).toEqual({ text: "Hallo", voice: "de_DE-thorsten-high" });
  });

  it("names Piper when a German voice is picked but Piper is missing", async () => {
    vi.stubGlobal("fetch", vi.fn());
    await expect(
      new KokoroVoiceProvider().synthesize(
        { text: "Hallo", voiceId: "df_eva", apiKey: KOKORO_LOCAL_TOKEN },
        ctx,
      ),
    ).rejects.toThrow(/Piper/);
  });
});

describe("voice response limits", () => {
  it.each([["Kokoro", () => new KokoroVoiceProvider(), "af_heart"] as const])(
    "rejects an oversized %s speech response before buffering it",
    async (_name, create, voiceId) => {
      const cancel = vi.fn().mockResolvedValue(undefined);
      vi.stubGlobal(
        "fetch",
        vi.fn().mockResolvedValue({
          ok: true,
          headers: new Headers({
            "content-length": String(MAX_SYNTHESIZED_AUDIO_BYTES + 1),
          }),
          body: { cancel },
        }),
      );

      await expect(
        create().synthesize({ text: "Hi", voiceId, apiKey: "sk-test" }, ctx),
      ).rejects.toThrow("Voice response is too large.");
      expect(cancel).toHaveBeenCalledOnce();
    },
  );
});

describe("ScriptedVoiceProvider", () => {
  it("verifies, lists, speaks, and transcribes without a network", async () => {
    const provider = new ScriptedVoiceProvider();
    await expect(provider.verify("short", ctx)).resolves.toEqual({
      ok: false,
      message: "That key is too short.",
    });
    await expect(provider.verify("fake-scripted-voice-key", ctx)).resolves.toEqual({ ok: true });
    expect(await provider.listVoices("fake-scripted-voice-key", ctx)).toEqual([
      { id: SCRIPTED_VOICE_ID, label: "Scripted", description: "Test voice" },
    ]);
    const clip = await provider.synthesize(
      { text: "Hello", voiceId: SCRIPTED_VOICE_ID, apiKey: "fake-scripted-voice-key" },
      ctx,
    );
    expect(clip.mimeType).toBe("audio/mpeg");
    expect([...clip.bytes]).toEqual([...SCRIPTED_MPEG]);
    await expect(
      provider.transcribe(
        {
          audio: new Uint8Array([1]),
          mimeType: "audio/webm",
          apiKey: "fake-scripted-voice-key",
        },
        ctx,
      ),
    ).resolves.toEqual({ text: SCRIPTED_TRANSCRIPT });
  });
});