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 { loadAudioManager } = require("./harness/audioManager");
// transcriptions.timestamp is what the history list sorts and groups on. A
// successful dictation stores when speech started, so every other row a
// recording can produce has to be dated the same way -- otherwise a failed
// take sorts above the dictation that was still being transcribed when it
// happened, and the column means two different things depending on the row.
async function loadManagerClass(t) {
return loadAudioManager(t, {
cachePrefix: "openwhispr-occurrence-time-test-",
settingsKey: "__occurrenceTimeSettings",
settings: { dataRetentionEnabled: true, audioRetentionDays: 0 },
});
}
test("a failed take is dated when it was recorded, not when the failure was written", async (t) => {
const { AudioManager, window } = await loadManagerClass(t);
const saved = [];
window.electronAPI.saveTranscription = async (text, rawText, options) => {
saved.push(options);
return { id: 1 };
};
const manager = Object.assign(Object.create(AudioManager.prototype), {
lastAudioBlob: null,
lastAudioMetadata: null,
translationRequested: false,
});
await manager.saveFailedTranscription("boom", null, {
analyticsOccurredAt: "2026-04-01T09:00:00.000Z",
});
assert.equal(saved.length, 1);
assert.equal(saved[0].analyticsOccurredAt, "2026-04-01T09:00:00.000Z");
});
test("a discarded take carries its recording time past the state reset", async (t) => {
const { AudioManager } = await loadManagerClass(t);
const manager = Object.assign(Object.create(AudioManager.prototype), {
recordingStartTime: Date.parse("2026-04-01T09:00:00.000Z"),
audioChunks: [],
_batchSegments: [],
recordingMimeType: "audio/webm",
});
// The snapshot exists because the save is async and runs after the manager
// has already been reset for the next recording, so recordingStartTime is
// gone by then -- the occurrence time has to be captured with the audio.
const snapshot = manager.takeDiscardedBatchSnapshot();
assert.equal(snapshot.analyticsOccurredAt, "2026-04-01T09:00:00.000Z");
});