FREE · OPEN · ON YOUR DEVICE

Turkish subtitles, for everyone.

Drop a video or audio file and QiTV's open models write its Turkish subtitles on your own device. Nothing is uploaded.

Drop your video or audio file here

NOTHING IS UPLOADED · UP TO 30 MINUTES · .SRT / .VTT

DEMO Subtitles written in advance by QiTV ASR (Best mode)

How it works

THREE STEPS
  1. 01

    Choose a file

    Video or audio: MP4, WebM, MP3, M4A, WAV… Up to 30 minutes. The file never leaves your device.

  2. 02

    Our model writes on your device

    The first time, the model downloads once (144 MB or 278 MB, shown before it starts) and your browser keeps it for your next visit. It cuts the speech at pauses and writes numbers as digits, as QiTV does: “saat on beş otuzda” → “saat 15.30'da”.

  3. 03

    Watch and download

    The subtitles appear under the video as they're written. Watch the video with them, then download .srt or .vtt.

Why we made it

FROM THE MODEL CARDS

We care about accessibility. Subtitles let deaf and hard-of-hearing people follow Turkish TV, films and talk, and they help everyone learning Turkish. Good Turkish captions shouldn't depend on a paid service or an internet connection.

01

Free to use

CC BY 4.0: personal or commercial, no fees, no sign-up. Just give credit.

02

Free to deploy

It runs offline on an ordinary CPU, on the viewer's own device. No audio leaves it.

03

Free to develop further

The weights, the ONNX files, and the training and evaluation scripts are all public. Fine-tune it on your own data, shrink it, ship it in your app.

04

Better open models, together

We test against every open Turkish model we could find and publish where others do better. If you build a better one, on ours or from scratch, tell us: we'd be proud to have helped.

The models

THREE OPEN MODELS · CC BY 4.0

MOONSHINE · TR + EN

Medium

The default Turkish model for QiTV's on-device AI subtitles. Turkish and English in one model; cased and punctuated.

  • 238M PARAMETERS
  • 278 MB · INT8
  • TR + EN
DRAMA, TV CHANNEL A · WORDS WRONG, LOWER IS BETTER
  • This model11.8 %
  • mihu-v219.2 %
  • zipformer-asr-turkish-v122.4 %
  • ufakzeka-kulak24.8 %
  • balbal-v226.2 %

Where others are better: on read speech and news, several open Turkish models are more accurate; mihu-v2 is the most accurate there.

On Hugging Face

MOONSHINE · TR + EN

Small

For computers where the medium model is too slow, at half the size. It writes Turkish and English too; cased and punctuated.

  • 117M PARAMETERS
  • 144 MB · INT8
  • TR + EN
DRAMA, TV CHANNEL A · WORDS WRONG, LOWER IS BETTER
  • This model14.3 %
  • mihu-v219.2 %
  • zipformer-asr-turkish-v122.4 %
  • ufakzeka-kulak24.8 %
  • balbal-v226.2 %

Where others are better: on read speech and news, several open Turkish models are more accurate; mihu-v2 is the most accurate there. Our medium model is more accurate on every test.

On Hugging Face

FASTCONFORMER · TR ONLY

Conformer

Built on NVIDIA FastConformer, Turkish only. It runs in your own app with sherpa-onnx or NeMo, and the NeMo file lets you keep training it. This page runs the Moonshine models.

  • 115M PARAMETERS
  • 142 MB · INT8
  • TR
DRAMA, TV CHANNEL A · WORDS WRONG, LOWER IS BETTER
  • This model16.5 %
  • mihu-v219.2 %
  • zipformer-asr-turkish-v122.4 %
  • ufakzeka-kulak24.8 %
  • balbal-v226.2 %

Where others are better: on read speech and news, several open Turkish models are more accurate; mihu-v2 is the most accurate there, and on drama our Moonshine models are.

On Hugging Face

All the scores

Share of words wrong (WER, %), lower is better; the lowest in each row is marked. The drama and song tests are private, the others public. Numbers and method from the model cards.

Words wrong (%) for seven open Turkish models on eight tests
TestMediumSmallConformermihu-v2ufakzeka-kulakbalbal-v2zipformer-asr-turkish-v1
TV channel A11.8 (lowest)14.316.519.224.826.222.4
TV channel B15.0 (lowest)17.319.122.227.329.025.3
TV channel C19.0 (lowest)22.023.125.131.533.828.4
Sung lyrics38.9 (lowest)48.550.668.377.779.077.2
FLEURS (read)11.713.611.47.9 (lowest)9.48.09.4
Common Voice (read)10.013.111.06.0 (lowest)9.37.09.6
MediaSpeech (news)12.213.412.710.8 (lowest)11.811.311.9
Talk videos25.128.724.422.5 (lowest)24.824.423.1

Use it yourself

OPEN WEIGHTS · OPEN CODE

On Hugging Face

qitv-asr-tr-moonshine-mid 278 MB qitv-asr-tr-moonshine-small 144 MB qitv-asr-tr-conformer 142 MB

Not only for QiTV: use them in your own app, your site, any project. CC BY 4.0; credit them as the model card asks and say what you changed.

The scripts

github.com/ozankaraali/qitv-asr MIT

The training, export and test scripts, the converter that writes numbers as digits among them. MIT.

On live TV

qitv.app QiTV

QiTV is a free IPTV player for Windows, macOS and Linux. With these models it writes subtitles for what you watch, on your device.