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 {
parseFfmpegDuration,
parseWavFormat,
isPcm16Mono16kWav,
wavToFloat32Samples,
} = require("../../src/helpers/ffmpegUtils");
const { wavHeader, pcm16Mono16kWav, insertWavChunk } = require("./harness/wavFixtures");
test("parseFfmpegDuration reads the input duration from ffmpeg output", () => {
const stderr = "Input #0, mp3, from 'recording.mp3':\n Duration: 01:13:00.25, start: 0.000000";
assert.equal(parseFfmpegDuration(stderr), 4380.25);
});
test("parseFfmpegDuration returns null when ffmpeg reports no duration", () => {
assert.equal(parseFfmpegDuration("Duration: N/A"), null);
assert.equal(parseFfmpegDuration(""), null);
});
test("isPcm16Mono16kWav accepts only what the local engines decode as-is", () => {
assert.equal(isPcm16Mono16kWav(wavHeader()), true);
assert.equal(isPcm16Mono16kWav(wavHeader({ sampleRate: 48000 })), false);
assert.equal(isPcm16Mono16kWav(wavHeader({ channels: 2 })), false);
assert.equal(isPcm16Mono16kWav(wavHeader({ audioFormat: 3, bitsPerSample: 32 })), false);
assert.equal(isPcm16Mono16kWav(Buffer.from("not a wav")), false);
});
const samples = [-32768, -16384, 0, 16384, 32767];
const wav = pcm16Mono16kWav(samples);
for (const [name, input] of [
["no metadata", wav],
["zero-length metadata", insertWavChunk(wav, "JUNK", Buffer.alloc(0), 12)],
["even-length metadata", insertWavChunk(wav, "JUNK", Buffer.from([1, 2]), 12)],
["odd-length metadata before fmt", insertWavChunk(wav, "JUNK", Buffer.from([42]), 12)],
["odd-length metadata before data", insertWavChunk(wav, "JUNK", Buffer.from([42]), 36)],
["three-byte metadata", insertWavChunk(wav, "JUNK", Buffer.from([1, 2, 3]), 36)],
[
"consecutive odd-length chunks",
insertWavChunk(
insertWavChunk(wav, "JUNK", Buffer.from([42]), 12),
"JUNK",
Buffer.from([7]),
12
),
],
]) {
test(`WAV chunk traversal preserves format and samples with ${name}`, () => {
assert.deepEqual(parseWavFormat(input), {
audioFormat: 1,
channels: 1,
sampleRate: 16000,
bitsPerSample: 16,
});
assert.equal(isPcm16Mono16kWav(input), true);
const decoded = wavToFloat32Samples(input);
assert.equal(decoded.length, samples.length * 4);
assert.deepEqual(
samples.map((_, index) => decoded.readFloatLE(index * 4)),
samples.map((sample) => sample / 32768)
);
});
}