Audiofilter
    Transcribe audio

    AI SPEECH TO TEXT

    Free AI speech to text in your browser

    Audiofilter turns your recordings into editable text and subtitles with speech recognition that runs in your browser. Add audio or record your microphone, choose a language, and export TXT, SRT, VTT or JSON. Your audio stays on your device.

    LOCAL AUDIO LAB

    Transcribe a recording

    Choose your audio, review the text and export a transcript or subtitles.
    Up to 5 minutes per fileAudio stays on your device. Models download when needed; a desktop browser is recommended.

    1. Add audio

    For more accurate speech recognition, use clear audio with as little background noise as possible. Noise and interference can reduce recognition accuracy.

      2. Choose recognition settings

      Automatic language detection uses Whisper Tiny and may be inaccurate, especially with noisy audio. Select the spoken language manually when possible.

      When choosing a model manually, use SenseVoice Small for Chinese, English, Cantonese, Japanese and Korean; use Whisper Base for other languages.

      Selecting a language manually automatically chooses a compatible model.

      Advanced recognition options

      Hotword and beam behavior depends on the selected model. Whisper uses hotwords as context.

      Only settings are saved, never your audio or transcript.Your settings are saved on this device.This browser could not save your settings.

      Add audio to begin. Models download only when needed.

      How automatic routing works

      SenseVoice handles Chinese, English, Cantonese, Japanese and Korean. Whisper handles other languages and uncertain detection. You can choose the language and model yourself.

      Local processing and downloads

      Your audio stays on your device. Models download once and are cached in this browser. Future runs reuse the cache, including after reloading the page.

      Whisper Tiny detects the language. SenseVoice and Whisper Base transcribe speech. A desktop browser is recommended.

      Recordings are transcribed after you stop recording. Long files use separate windows; review words near window boundaries. Timestamps refer to segments, not individual words.

      Short, quiet, noisy or mixed-language recordings may be misidentified. Select the spoken language manually when needed.

      Remove background noise