Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.
Python agentaichatgptdeepseekfine-tuninggemmaimage-generationllamallmllmsopenaipythonqwenreinforcement-learningself-hostedstable-diffusiontext-to-speechttsuiunsloth
// SPDX-License-Identifier: AGPL-3.0-only
// Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
import assert from "node:assert/strict";
import test from "node:test";
import {
type TranscriptExport,
exportFileName,
exportTranscript,
toJson,
toSrt,
toTxt,
toVtt,
} from "../src/features/audio/transcript-export.ts";
import { EMPTY_TRANSCRIPT_DETAILS } from "../src/features/audio/transcript-model.ts";
const seg = (start: number, end: number, text: string, speaker?: string) => ({
start,
end,
text,
speaker,
});
const plain: TranscriptExport = {
title: "note.m4a",
text: "Just text.",
model: "whisper-small",
names: {},
details: { ...EMPTY_TRANSCRIPT_DETAILS, duration: 4.5 },
};
const timed: TranscriptExport = {
...plain,
names: { S02: "Bea" },
details: {
...EMPTY_TRANSCRIPT_DETAILS,
segments: [
seg(1.234, 3, "Hello there.", "S01"),
seg(3, 3.5, "Hi!", "S02"),
seg(3661.001, 3662.25, "<b>Tom & Jerry</b>", "S02"),
],
words: [{ start: 1.234, end: 1.6, word: "Hello" }],
speakers: [
{ id: "S01", label: "Speaker 1" },
{ id: "S02", label: "Speaker 2" },
],
},
};
test("SRT numbers cues with comma milliseconds; VTT has a header, voice tags and escaped text", () => {
assert.equal(
toSrt(timed),
"1\n00:00:01,234 --> 00:00:03,000\nSpeaker 1: Hello there.\n\n" +
"2\n00:00:03,000 --> 00:00:03,500\nBea: Hi!\n\n" +
"3\n01:01:01,001 --> 01:01:02,250\nBea: <b>Tom & Jerry</b>\n",
);
assert.equal(
toVtt(timed),
"WEBVTT\n\n00:00:01.234 --> 00:00:03.000\n<v Speaker 1>Hello there.\n\n" +
"00:00:03.000 --> 00:00:03.500\n<v Bea>Hi!\n\n" +
"01:01:01.001 --> 01:01:02.250\n<v Bea><b>Tom & Jerry</b>\n",
);
});
test("cues: no names without speakers, no early end on a blank line, one cue over the clip without timing", () => {
const unnamed = { ...timed, details: { ...timed.details, speakers: [] } };
assert.doesNotMatch(toSrt(unnamed) + toVtt(unnamed), /Speaker 1|<v /);
const segments = [seg(0, 1, "one\n\n two"), seg(1, 2, " "), seg(2, 3, "x")];
const gaps = { ...plain, details: { ...plain.details, segments } };
assert.equal(
toSrt(gaps),
"1\n00:00:00,000 --> 00:00:01,000\none\ntwo\n\n2\n00:00:02,000 --> 00:00:03,000\nx\n",
);
assert.equal(toSrt(plain), "1\n00:00:00,000 --> 00:00:04,500\nJust text.\n");
assert.equal(
toVtt(plain),
"WEBVTT\n\n00:00:00.000 --> 00:00:04.500\nJust text.\n",
);
assert.equal(toSrt({ ...plain, text: "" }), "");
assert.equal(toVtt({ ...plain, text: "" }), "WEBVTT\n");
});
test("TXT is the plain text or paragraphs led by the current names; JSON keeps keys, names and words", () => {
assert.equal(toTxt(plain), "Just text.");
assert.equal(
toTxt(timed),
"Speaker 1: Hello there.\n\nBea: Hi! <b>Tom & Jerry</b>\n",
);
const json = JSON.parse(toJson(timed));
assert.deepEqual(
Object.keys(json).join(),
"title,model,language,duration,text,segments,words,speakers",
);
assert.deepEqual(json.speakers, [
{ id: "S01", label: "Speaker 1", name: null },
{ id: "S02", label: "Speaker 2", name: "Bea" },
]);
assert.equal("words" in JSON.parse(toJson(plain)), false);
});
test("each format has its media type, and file names drop the extension and unsafe characters", () => {
for (const [format, mime, write] of [
["txt", "text/plain;charset=utf-8", toTxt],
["srt", "application/x-subrip;charset=utf-8", toSrt],
["vtt", "text/vtt;charset=utf-8", toVtt],
["json", "application/json;charset=utf-8", toJson],
] as const)
assert.deepEqual(exportTranscript(format, timed), {
content: write(timed),
mime,
});
for (const [title, name] of [
["meeting.wav", "meeting.srt"],
['a/b\\c:d*e?"f<g>h|i.mp3', "a_b_c_d_e__f_g_h_i.srt"],
["tab\there", "tab_here.srt"],
["", "transcript.srt"],
[".wav", "transcript.srt"],
])
assert.equal(exportFileName(title, "srt"), name);
});