Is there a per-minute limit like other sites?

No. Sites limit minutes because server transcription costs them money per minute. This tool runs on your device, so a 3-hour MP3 costs us the same as a 3-minute one: nothing.

Can I get timestamps with the text?

Yes — after transcribing, use the .SRT or .VTT buttons to download the text with timestamps, or .TXT for clean prose.

What about music or lyrics?

Whisper is built for speech. It will attempt lyrics but accuracy drops sharply with singing and heavy instrumentation — spoken-word content is where it shines.

Does it tell me who is speaking?

No. The transcript is a single flow of text without speaker labels. For an interview or a panel, add names as you edit; the timestamps in an SRT export help you find where each person starts.

Which model should I pick for a long episode?

Fast gives a quick first draft and is the sensible start on a laptop. Best is more accurate but needs more memory and much more time, so try it on a short clip before committing an hour of audio.

MP3 to Text — Transcribe MP3 Free (No Upload)

Convert an MP3 into text right in your browser — podcasts, lectures, recorded calls. Free and private.

Model

Spoken

Output

Timestamps [0:05]

🌍 The spoken language is detected automatically and the transcript comes out in that same language. Set it manually above if your audio is hard to hear, and use Output to translate into English.

⬇️ One-time download: 104 MB (Accurate model 73 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 4-8 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 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.

An MP3 is usually long: a podcast episode, a two-hour lecture, an interview recorded on a dictaphone, a call exported from a phone app. Reading is faster than listening, and text can be searched, quoted and pasted into notes. Transcription services know this, which is why they charge by the minute, and why they need your recording on their servers to do it.

On this page the Whisper speech model from OpenAI does the work inside your browser. The MP3 stays on your machine, there is no per-minute fee, and what you get is a plain text transcript to copy or download, with SRT and VTT subtitle files if you want timestamps. The first run downloads the audio engine and the speech model, which takes a moment; later runs reuse them when the browser has kept the cache.

A clear single voice comes out close to publishable; a kitchen-table conversation with crosstalk and clatter needs more editing. Check names, figures and anything you will quote. Three model sizes are offered, from Fast, which is enough for a first pass, to Best, which wants more memory and time but misses less. It does not label who is speaking, so for an interview add the speaker names yourself. Speed depends on your computer; an hour of audio is a job for a laptop rather than a phone.

More sizes & options

Frequently asked questions

Is there a per-minute limit like other sites?: No. Sites limit minutes because server transcription costs them money per minute. This tool runs on your device, so a 3-hour MP3 costs us the same as a 3-minute one: nothing.
Can I get timestamps with the text?: Yes — after transcribing, use the .SRT or .VTT buttons to download the text with timestamps, or .TXT for clean prose.
What about music or lyrics?: Whisper is built for speech. It will attempt lyrics but accuracy drops sharply with singing and heavy instrumentation — spoken-word content is where it shines.
Does it tell me who is speaking?: No. The transcript is a single flow of text without speaker labels. For an interview or a panel, add names as you edit; the timestamps in an SRT export help you find where each person starts.
Which model should I pick for a long episode?: Fast gives a quick first draft and is the sensible start on a laptop. Best is more accurate but needs more memory and much more time, so try it on a short clip before committing an hour of audio.

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.