Transcribe Audio to Text Privately, on Your Own Device

To transcribe a recording to text without uploading it, open Speech to Text, drop in the audio or video file, choose the Spoken language, and click Transcribe. OpenAI's open-source Whisper model then runs inside your browser and gives you an editable transcript plus SRT and VTT subtitle files. To get a PDF instead, use Audio Transcript to PDF, which works the same way and adds a Download PDF button.

How does on-device transcription work?

Most online transcription services upload your audio to their servers. These tools do it differently: the speech-recognition model is downloaded into your browser and runs on your own processor. Your recording is never sent anywhere, which matters for interviews, medical or legal notes, HR conversations and anything else you'd rather not hand to a third party.

The model is Whisper, released by OpenAI as open source. You choose between two sizes under Accuracy:

• Fast (≈64 MB): the Whisper "tiny" model. Quicker to download and run, less accurate.
• Better (≈105 MB): the Whisper "base" model. Noticeably more accurate, a little slower.

The model downloads the first time you use it and is then cached by your browser, so later transcriptions start straight away, even offline. Speech to Text starts on Better; Audio Transcript to PDF starts on Fast.

Behind the scenes the audio is decoded, mixed down to mono and resampled to 16 kHz, which is what Whisper expects. It's then transcribed in 30-second windows that overlap by 5 seconds, so words at the edges of each window aren't lost. Each window comes back with timestamps, which is how the subtitle files are built.

Which languages can it transcribe?

The Spoken language list includes English, Hindi, Bengali, Tamil, Telugu, Marathi, Gujarati, Kannada, Malayalam, Punjabi and Urdu, plus Spanish, French, German, Portuguese, Italian, Russian, Arabic, Chinese, Japanese, Korean, Indonesian, Turkish and Dutch. Auto-detect is available too, but choosing the language yourself is faster and more accurate.

Accuracy is best for clear English. Other languages are usable but expect more corrections, especially with the Fast model.

How do I transcribe a recording, step by step?

1. Open Speech to Text and drop in a file (MP3, WAV, M4A, OGG, FLAC, WEBM, or a video). Or click Start recording under the upload area, speak, then click Stop & use recording.
2. Under Accuracy, choose Better unless you're on a slow device.
3. Under Spoken language, pick the language. Leave Auto-detect only if you don't know it.
4. Click Transcribe. The first time, you see "Downloading Whisper model… (first time only)" with a megabyte count; after that the progress line reads "Transcribing … of audio on your device".
5. Read through the transcript in the text box and correct names, numbers and jargon. Tick Show timestamps to see each line with its start time.
6. Export with Copy text, .txt, .srt subtitles or .vtt subtitles.

Export the subtitle files before you edit. They're built from the original timed transcript, so the .srt and .vtt buttons switch off once you change the text.

How do I save a transcript as a PDF?

Audio Transcript to PDF adds a PDF step:

1. Upload a file, or click Record with microphone and then Stop recording.
2. Set Accuracy and Spoken language.
3. Leave Include timestamps in the PDF ticked if you want each line to start with its time, such as [01:05].
4. Click Transcribe to PDF.
5. Edit the text under Transcript — edit before exporting, then click Download PDF. You can also save .txt or .srt subtitles.

The PDF has a title, a line with the file name, length and date, then the transcript. Timestamps are only included while the transcript is unedited; once you change the text, the PDF uses your edited version. To split it into paragraphs, leave a blank line between them while editing.

Note the script limit: the PDF supports Latin, Cyrillic, Greek and Vietnamese text. For a Hindi, Tamil or other Indic transcript, or for Arabic, Chinese, Japanese or Korean, download the .txt file and open it in a word processor instead.

How can I get a more accurate transcript?

• Record close to the speaker. A phone held near the speaker beats a laptop across the room.
• Keep background noise down. Music, fans and crosstalk cause most errors. When recording with the Voice Recorder, leave Noise suppression & echo cancellation ticked for speech.
• Pick the language rather than relying on Auto-detect.
• Use Better for anything you'll publish or quote.
• Expect to label speakers yourself. Whisper doesn't say who is talking, so add names while editing.
• Cut long silences. Whisper can occasionally repeat or invent a phrase over long stretches of silence or noise. Trimming them out with the Audio Cutter helps.
• Check every name and number. These are the most common errors on any speech recogniser.

Can I transcribe long recordings?

There's no fixed length limit. Because the audio is processed in 30-second windows, a one-hour lecture works the same way as a one-minute voice note; it just takes longer. Speed depends entirely on your device: a recent laptop is much faster than an older phone, and the Fast model is quicker than Better.

The whole recording is decoded into memory first, so very long files (several hours) can be slow or fail on phones. For those, split the recording into parts of 30–60 minutes with the Audio Cutter, transcribe each part, and paste the texts together. Keep the tab open while it works; closing it stops the transcription.

Troubleshooting

• "No speech was detected": the file may be silent, very quiet or mostly music. Play it first to check.
• Microphone errors: if access was blocked, allow the microphone in your browser's site settings and try again. Recording needs a secure (https) page.
• "Could not decode": the browser can't read that audio format. Convert it to MP3 or WAV with the Audio Converter first.
• Gibberish in another language: Auto-detect guessed wrong. Choose the language and run it again.
• Live captions: these tools don't do word-by-word live captions. Record first, then transcribe.

Quick answers

Can I transcribe audio to text without uploading it?

Yes. EasyConvert's Speech to Text runs OpenAI's Whisper model inside your browser, so the recording stays on your device. Only the model itself is downloaded, once, and then cached.

Does Speech to Text work for Hindi and other Indian languages?

Yes. You can choose Hindi, Bengali, Tamil, Telugu, Marathi, Gujarati, Kannada, Malayalam, Punjabi or Urdu as the spoken language. Accuracy is best for clear English, so proofread Indian-language transcripts carefully and use the Better model.

How do I make subtitles from a video?

Drop the video into Speech to Text, choose the language and click Transcribe. Then click .srt subtitles or .vtt subtitles before editing the text, because the subtitle buttons switch off once the text is changed.

Does it work offline?

After the Whisper model has downloaded once, transcription runs locally on your device. The model stays in your browser's cache until you clear it.