About Speech to Text Online
This speech to text tool is free, needs no sign-up and runs in your browser, so your audio is never uploaded. Record with your microphone or choose an audio or video file, and the tool decodes the sound on your device and turns it into text with timestamps. You can copy the transcript or save it as plain text, SubRip subtitles (SRT) or WebVTT.
It runs OpenAI Whisper, in the tiny (41 MB) or base (77 MB) size, both multilingual and under the Apache-2.0 licence. The model downloads from Hugging Face the first time, after you confirm, and your browser may cache it. Only the model file is downloaded. The audio itself never leaves the page.
Long recordings are handled in windows of 30 seconds with a 5 second overlap, so words on a cut are not lost. You can choose the spoken language, including Bangla, or let the model detect it, and you can ask for a translation into English instead of a transcript in the original language.
Be realistic about accuracy. Tiny and base are the small Whisper models, picked so they download quickly and run on a phone. They do well on clear English speech and are weaker on noisy audio, accents and music. Bangla accuracy is limited with these sizes, so read the output and fix the mistakes before you rely on it.
How to use Speech to Text Online
- Choose Audio or video file, or Microphone.
- For a file, drop it on the box, tap to choose it, or paste it. For the microphone, press Start recording and Stop recording when you are done.
- Pick the model (Tiny or Base), the spoken language, and turn on Translate to English if you want English text.
- Press Download model, confirm, and wait for it to load. This is needed once per visit; your browser may keep the file.
- Press Transcribe. Long audio takes longer than short audio, and a computer with WebGPU is faster than a phone.
- Read the transcript, then switch between Text, SRT and VTT and press Copy or Download.
Features
- Whisper tiny and base, multilingual, in your browser.
- Microphone recording, or any audio or video file your browser can decode.
- Language choice including Bangla, English, Hindi, Arabic, Urdu and more, or automatic detection.
- Translate to English with the Whisper translate task.
- Timestamps for every phrase, shown in the list.
- Export as TXT, SRT subtitles or WebVTT.
- Long audio in 30 second windows with a 5 second overlap, up to 60 minutes.
- Model size and licence shown before the download, with a confirm step.
Worked examples
Two phrases: "Welcome to the show." from 0 to 2.4 s, "Today we talk about sleep." from 2.4 to 5.75 s gives 1 00:00:00,000 --> 00:00:02,400 Welcome to the show. 2 00:00:02,400 --> 00:00:05,750 Today we talk about sleep.
SRT numbers each cue and uses a comma before the milliseconds.
The same two phrases as WebVTT gives WEBVTT 00:00:00.000 --> 00:00:02.400 Welcome to the show. 00:00:02.400 --> 00:00:05.750 Today we talk about sleep.
VTT starts with WEBVTT and uses a dot before the milliseconds.
A 2 minute recording gives 6 windows of 30 seconds, each overlapping the next by 5 seconds on both sides.
Tiny or base
| Model | Download | Speed | Accuracy |
|---|---|---|---|
| Whisper tiny | 41 MB | Faster | Fine for clear speech |
| Whisper base | 77 MB | Slower | Better, including Bangla |
Good to know
About accuracy and Bangla
Whisper tiny and base are small models. They can miss words, merge them or invent text in silent parts, and they make more mistakes in Bangla than in English. Choose Base for Bangla, set the language to Bangla instead of automatic detection, and use clear audio with little background noise.
The transcript is a draft. Do not use it for legal, medical or other work where an error matters without reading it against the recording.
Which files work
The browser decodes the audio, so a file works if your browser can play it: usually MP3, WAV, M4A, OGG and WebM audio, and MP4 or WebM video. If a file is refused, convert it first with the file converter on this site.
Privacy: how it works
- Your audio and video are decoded and transcribed in your browser. They are never uploaded to Ulkaa or to anyone else, and a microphone recording stays in the tab.
- One download is needed: the Whisper model files come from Hugging Face the first time you confirm. That request does not include your audio.
Tips
- Record close to the microphone in a quiet room. Clean audio changes the result more than the model choice.
- Set the spoken language when you know it. Automatic detection can guess wrong on short clips.
- Use Base if you can wait a little longer and the audio is not English.
- For subtitles, check the timing against the video, because the timestamps are per phrase, not per word.
Frequently asked questions
Is it really free?
Yes. There is no price, no trial and no credit card. Ulkaa Tool does the work with your own device, so there is no server bill to pass on to you.
Do I need an account?
No. There is no sign-up and no login. Open the page and use it. An Ulkaa account is not needed for anything on this tool.
Does it run in my browser?
Yes. Your audio are processed by code running in your browser tab, on your own device.
Is my data uploaded?
No. Your audio are never sent to Ulkaa or to anyone else. You can check this yourself by watching the Network tab in your browser developer tools while you use the tool.
Is anything downloaded?
Yes, once: the Whisper model (41 MB for tiny, 77 MB for base) from Hugging Face, after you confirm. Your audio is never uploaded.
How accurate is it?
Good on clear English speech, weaker with noise, accents and music. Tiny and base are small models, so always read the result.
Does it understand Bangla?
It can transcribe Bangla, but accuracy is limited with the tiny and base models. Use Base and set the language to Bangla.
Can it translate to English?
Yes. Turn on Translate to English and Whisper writes English text from speech in another language.
How long can the audio be?
Up to 60 minutes. It is processed in 30 second windows, so long files take a while, especially on a phone.
Can I get subtitles?
Yes. Choose SRT or VTT and press Download. The timestamps are per phrase.
Which models and licences are used?
OpenAI Whisper tiny and base, in ONNX form from onnx-community on Hugging Face, under the Apache-2.0 licence.
Popular Speech to Text Online pages
- Audio to Text ConverterDrop an audio file and get a transcript with timestamps.
- Video to Text ConverterChoose a video and transcribe what is said in it.
- Voice to Text OnlineRecord your voice and turn it into text.
- Transcribe Bangla AudioTurn Bangla speech into text, and read the result carefully.
- Speech to Text With Subtitles (SRT)Transcribe a recording and download it as SRT or VTT subtitles.