Voice-to-text dictation app with local (Nvidia Parakeet/Whisper) and cloud models (BYOK). Privacy-first and available cross-platform.
JavaScript aianthropiccross-platformgeminigroqlinuxmacosnvidiaopen-sourceopenaiparakeetspeech-to-texttranscribewhisperwindows
const test = require("node:test");
const assert = require("node:assert/strict");
const React = require("react");
const { createRoot } = require("react-dom/client");
const {
createRendererServer,
installBrowserGlobals,
installHookDom,
} = require("../lib/rendererTestHarness");
// A minimal AudioManager stand-in whose startRecording() fails without ever
// invoking the onStateChange callback registered via setCallbacks — this
// mirrors the real failure path (mic-open rejection, permission denial,
// etc.), where only onError fires. reportLifecycle's own report inside
// performStartRecording, and its finally-block reset, are the only things
// this test exercises.
const FAKE_AUDIO_MANAGER_SOURCE = `
export default class FakeAudioManager {
constructor() {
this.voiceAgentRequested = false;
this.translationRequested = false;
this.sttConfig = { success: true };
}
getState() {
return {};
}
setCallbacks() {}
setVoiceAgentRequested(value) {
this.voiceAgentRequested = value;
}
setAssistantSelectionContext() {}
setTranslationRequested(value) {
this.translationRequested = value;
}
shouldUseStreaming() {
return false;
}
isSttConfigStale() {
return false;
}
prepareMicCapture() {}
cancelPreparedMicCapture() {}
cleanup() {}
async startRecording() {
return false;
}
}
`;
test("a failed dictation start reports the lifecycle back to idle instead of sticking on preparing", async (t) => {
// t.after hooks run in registration order, so this unmount must be
// registered before installBrowserGlobals/installHookDom's own cleanup —
// otherwise window/document are already torn down when unmount runs.
let root = null;
t.after(async () => {
if (root) await React.act(async () => root.unmount());
});
const reported = [];
const noopDispose = () => () => {};
installBrowserGlobals(t, {
window: {
electronAPI: {
// The mount effect registers all seven of these unconditionally (or
// via optional chaining); onToggleDictation in particular is called
// without "?.", so it must exist even though this test drives
// startRecording() directly rather than through any of these events.
onToggleDictation: noopDispose,
onToggleVoiceAgent: noopDispose,
onToggleTranslation: noopDispose,
onStartDictation: noopDispose,
onPrepareDictation: noopDispose,
onCancelDictationPreparation: noopDispose,
onStopDictation: noopDispose,
dictationLifecycleStateChanged: (state, inputKind) =>
reported.push(`${state}:${inputKind}`),
},
},
});
const container = installHookDom(t);
// The start path awaits two animation frames before opening the mic;
// resolve them synchronously instead of riding out the 250ms/frame
// fallback timeout twice in this test.
globalThis.requestAnimationFrame = (callback) => {
callback();
return 1;
};
const vite = await createRendererServer(t, {
cachePrefix: "openwhispr-audio-recording-preparing-reset-",
mockModules: {
"/helpers/audioManager": FAKE_AUDIO_MANAGER_SOURCE,
},
});
const { useAudioRecording } = await vite.ssrLoadModule("/hooks/useAudioRecording.js");
let api;
function Harness() {
api = useAudioRecording(() => {}, { onDemoEvent: () => {} });
return null;
}
root = createRoot(container);
await React.act(async () => {
root.render(React.createElement(Harness));
});
// The mount effect reports its own "idle" reset; isolate the assertion to
// what performStartRecording itself reports.
reported.length = 0;
let started;
await React.act(async () => {
started = await api.startRecording();
});
assert.equal(started, false);
assert.deepEqual(reported, ["preparing:dictation", "idle:dictation"]);
});