Does it add nikud (vowel points)?

No. The output is standard unpointed Hebrew, as in newspapers and messages. That is what nearly everyone wants for a transcript.

Why is Best the default here?

Because the smaller models make noticeably more mistakes in Hebrew. Best is slower but accurate enough to edit.

Does mixed Hebrew and English work?

The model follows the selected language, so English passages inside Hebrew speech may be inconsistent. Choose the language that dominates the recording.

Will the Hebrew display correctly when I paste it into Word or Google Docs?

Yes. Hebrew characters carry their own right-to-left direction, so editors render the text correctly without settings. Lines that mix Hebrew with numbers or English can look jumbled at the boundaries; set the paragraph direction to RTL if that happens.

How long does the first download take?

Best is the largest of the three models, so the one-time download is longer than on other pages, from a minute to several on a slow connection. It is cached when possible, so later sessions usually start straight away.

Transcribe Hebrew Audio to Text — Free, No Upload

Turn Hebrew audio or video into right-to-left Hebrew text in your browser. 100% free, no per-minute limit, no signup — nothing is uploaded. Opens with the most accurate model.

Model

Spoken

Output

Timestamps [0:05]

⬇️ One-time download: 269 MB (Best model 238 MB + 31 MB audio engine, shared with the video tools) — only the first time: it stays saved in your browser, so next time it starts right away. (Translating or switching model needs its own file; a browser can clear this cache if storage runs very low.) Processing takes about 25-50 min per 10 minutes of audio on a typical laptop.

What exactly is downloaded — and is it safe?

• The speech model is OpenAI's Whisper, open source, converted to run in browsers. The audio engine is FFmpeg, the open-source tool behind most video software.
• They come from Hugging Face and unpkg — public repositories used by researchers and developers worldwide, not from us.
• They are not programs. Nothing is installed, no permission is asked, and your browser keeps them in its own cache exactly like it keeps a website's images.
• They can't read your files or touch anything outside this page. Remove them any time by clearing this site's data in your browser.
• Your audio never goes the other way: this site has no server receiving files.

Drop your video or audio here or click to choose

transcribed on your device by OpenAI's Whisper — your recording is not uploaded

ℹ️ Use a recording on your device that you own or have permission to process. This tool cannot fetch YouTube or social video links. Start with a short clip. Larger models can improve accuracy but require more memory and download data. Processing time varies by device, model and recording length. Always review the result.

Hebrew gives a transcription model more to resolve than most languages. Everyday writing leaves out the vowel points, so the same letters can be several words, and the right one depends on context; and the script runs right to left. The output here is standard unpointed Hebrew with punctuation, and in our test with real Hebrew speech the language was identified correctly.

Because Hebrew is harder for the model than English or Spanish, this page opens with the Best model already chosen. That means a larger one-time download than the other transcription pages, cached when possible, and slower processing, but the accuracy gap is large enough that a lecture, an interview or a meeting comes out as something to edit rather than retype. Everything runs in your browser: the recording is not uploaded, there is no minute limit and nothing is charged.

What to check afterwards: names, numbers, dates and any English words embedded in Hebrew speech, which the model may render inconsistently because it follows the selected language. Nikud is not added, which is what nearly everyone wants for a transcript. On a phone, the Best model is a heavy load, so a laptop is the better place for a long recording. Copy the result, download it as TXT, or export SRT or VTT subtitles.

More sizes & options

Frequently asked questions

Does it add nikud (vowel points)?: No. The output is standard unpointed Hebrew, as in newspapers and messages. That is what nearly everyone wants for a transcript.
Why is Best the default here?: Because the smaller models make noticeably more mistakes in Hebrew. Best is slower but accurate enough to edit.
Does mixed Hebrew and English work?: The model follows the selected language, so English passages inside Hebrew speech may be inconsistent. Choose the language that dominates the recording.
Will the Hebrew display correctly when I paste it into Word or Google Docs?: Yes. Hebrew characters carry their own right-to-left direction, so editors render the text correctly without settings. Lines that mix Hebrew with numbers or English can look jumbled at the boundaries; set the paragraph direction to RTL if that happens.
How long does the first download take?: Best is the largest of the three models, so the one-time download is longer than on other pages, from a minute to several on a slow connection. It is cached when possible, so later sessions usually start straight away.

Guides

Related tools

🔒 Private mode (the default): your recording never leaves your device. Instant mode is optional, you switch it on yourself, and it sends the audio to the browser's speech service.