Question 1

Which languages can it recognize?

Accepted Answer

The tool uses OpenAI's Whisper multilingual model to recognize dozens of languages, including Chinese, English, Japanese, and Korean. Choosing 'Auto Detect' allows the model to determine the language automatically. If you know the audio language, manual specification is usually more accurate. Transcription results can be exported as plain text or SRT subtitles with timestamps.

Question 2

Why is it slow the first time and requires an internet connection?

Accepted Answer

The first use requires downloading the Whisper model file (tens of MB) from the internet. After downloading, the browser will cache it, allowing reuse and even offline operation later. Recognition is performed locally on your device, so longer audio may take more time. It's recommended to start with shorter clips to experience the effect.

Question 3

Will the audio be uploaded to a server? How is it different from online transcription?

Accepted Answer

No. Except for the initial model download, audio decoding and speech recognition are completed entirely in the local browser without uploading to any server. This is fundamentally different from transcription services that send recordings to the cloud. Note: This tool does not use the browser's cloud speech interface precisely to ensure audio does not leave your local device.

Speech to Text

Frequently Asked Questions