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 { createRendererServer, installBrowserGlobals } = require("../lib/rendererTestHarness");
const RAW = "hey are you around this evening thinking coffee around six";
// Tone & Style classifies the dictation's app at recording start and the
// cleanup request carries the tone picked for that kind of app.
test("dictation cleanup carries the tone for the app it was captured in", async (t) => {
const { window } = installBrowserGlobals(t, { window: { dispatchEvent() {} } });
Object.defineProperty(navigator, "onLine", { configurable: true, value: true });
t.after(() => delete navigator.onLine);
const vite = await createRendererServer(t, {
cachePrefix: "openwhispr-audio-tone-style-",
mockModules: {
"/utils/logger":
"export default { debug() {}, info() {}, warn() {}, error() {}, logReasoning() {} };",
},
});
const AudioManager = (await vite.ssrLoadModule("/helpers/audioManager.js")).default;
const service = (await vite.ssrLoadModule("/services/ReasoningService.ts")).default;
t.after(() => service.destroy());
const { useSettingsStore } = await vite.ssrLoadModule("/stores/settingsStore.ts");
const { usePolicyStore } = await vite.ssrLoadModule("/stores/policyStore.ts");
usePolicyStore.setState({ status: "unmanaged", policy: null });
useSettingsStore.setState({
isSignedIn: true,
useCleanupModel: true,
cleanupMode: "openwhispr",
cleanupCloudMode: "openwhispr",
useDictationAgent: false,
customPrompts: { cleanup: "" },
preferredLanguage: "en",
dataRetentionEnabled: true,
toneStyleEnabled: true,
toneStyles: { personal: "very_casual", work: "casual", email: "excited", other: "formal" },
});
window.electronAPI.cloudTranscribe = async () => ({ success: true, text: RAW });
let capturedCategory;
window.electronAPI.captureDictationAppCategory = async () => capturedCategory;
const reasonRequests = [];
window.electronAPI.cloudReason = async (_text, options) => {
reasonRequests.push(options);
return { success: true, text: "hey u around this evening? thinking coffee around 6 lmk" };
};
const createManager = () =>
Object.assign(Object.create(AudioManager.prototype), {
voiceAgentRequested: false,
translationRequested: false,
pendingCleanupFailure: null,
getEffectiveSttLanguage: () => "en",
getWhisperPrompt: () => null,
isDictionaryEcho: () => false,
isReasoningAvailable: async () => true,
finalizeChineseScript: async (text) => text,
});
const lastRequest = () => reasonRequests[reasonRequests.length - 1];
await t.test("Cloud batch cleanup sends the captured app's tone", async () => {
capturedCategory = "personal";
const manager = createManager();
manager.beginToneStyleCapture();
await manager.processWithOpenWhisprCloud(new Blob(["synthetic audio"]));
assert.equal(lastRequest().tone, "very_casual");
assert.equal(lastRequest().appCategory, "personal");
});
await t.test("streaming Cloud cleanup sends the captured app's tone", async () => {
capturedCategory = "email";
window.electronAPI.cloudStreamingUsage = async () => ({ success: true });
const manager = Object.assign(createManager(), {
streamingFinalText: RAW,
streamingCleanupFns: [],
micRecovery: { stop() {} },
finishStreamingFallbackSegment: async () => {},
mergeRecordedSegments: async () => null,
awaitStreamingTextSettled: async () => {},
getStreamingProvider: () => ({
awaitsFinalTranscript: true,
stop: async () => ({ success: true }),
}),
getStreamingProviderName: () => "openwhispr",
shouldUseStreaming: () => false,
onTranscriptionComplete: () => {},
});
manager.beginToneStyleCapture();
await manager._finalizeStreamingRecording(null);
assert.equal(lastRequest().tone, "excited");
assert.equal(lastRequest().appCategory, "email");
});
await t.test("an app the classifier can't place gets the other-apps tone", async () => {
capturedCategory = null;
const manager = createManager();
manager.beginToneStyleCapture();
await manager.processWithOpenWhisprCloud(new Blob(["synthetic audio"]));
assert.equal(lastRequest().tone, "formal");
assert.equal(lastRequest().appCategory, "other");
});
await t.test("no tone without a capture, or once Tone & Style is off", async () => {
capturedCategory = "email";
await createManager().processWithOpenWhisprCloud(new Blob(["synthetic audio"]));
assert.equal("tone" in lastRequest(), false);
const manager = createManager();
manager.beginToneStyleCapture();
useSettingsStore.setState({ toneStyleEnabled: false });
await manager.processWithOpenWhisprCloud(new Blob(["synthetic audio"]));
useSettingsStore.setState({ toneStyleEnabled: true });
assert.equal("tone" in lastRequest(), false);
});
await t.test("an app classified after 1.5 s of speech still gets its tone", async (st) => {
st.mock.timers.enable({ apis: ["setTimeout"] });
let classify;
capturedCategory = new Promise((resolve) => (classify = resolve));
const manager = createManager();
manager.beginToneStyleCapture();
st.mock.timers.tick(1600);
classify("email");
assert.deepEqual(await manager.consumeToneStyle(), { appCategory: "email", tone: "excited" });
});
await t.test("a cancelled upload that still resolves leaves the next dictation's tone alone", async () => {
capturedCategory = "work";
const manager = createManager();
// The next dictation started while the cancelled one's upload was in flight.
manager.beginToneStyleCapture();
await manager.processWithOpenWhisprCloud(new Blob(["synthetic audio"]), {}, () => true);
assert.equal("tone" in lastRequest(), false);
assert.deepEqual(await manager.consumeToneStyle(), { appCategory: "work", tone: "casual" });
});
await t.test("a new recording drops the previous recording's capture", async () => {
capturedCategory = "work";
const manager = createManager();
manager.beginToneStyleCapture();
manager.setVoiceAgentRequested(false);
assert.equal(await manager.consumeToneStyle(), null);
});
await t.test("local and BYOK cleanup get the tone through the reasoning config", async () => {
capturedCategory = "email";
const configs = [];
const manager = createManager();
manager.processWithReasoningModel = async (_text, _model, _agentName, config) => {
configs.push(config);
return "Thanks so much for today!";
};
manager.beginToneStyleCapture();
await manager.processTranscription(RAW, "local");
assert.deepEqual(configs[0].toneStyle, { appCategory: "email", tone: "excited" });
assert.equal(configs[0].inferenceScope, "dictationCleanup");
});
});