Audio to Text, Free nothing stored

SRT, VTT or plain text: picking the file to save

Set Save as to SRT subtitles, VTT subtitles or Plain text; the starting choice is SRT subtitles. SRT numbers every subtitle and writes its times with a comma before the milliseconds, like 00:01:02,345. Most video editors and upload forms accept it.

VTT opens with a WEBVTT line, leaves out the numbers and puts a period in its times. Pick it for a web page’s own video player. Plain text drops the time codes and joins the lines into paragraphs, which suits show notes, quotes and searchable lecture notes.

How subtitle lines are split and timed

The model returns whole sentences, and one sentence can run for half a minute. The tool breaks anything longer than 84 characters or 7 seconds: first at sentence ends, then between words, preferring a comma. Each subtitle shows on at most two lines, broken near the middle on a space.

When a long sentence is split, its time is shared out by the length of each piece, because the model times sentences rather than single words. Inside a long, fast sentence a subtitle can therefore land a moment early or late. Play the file against the video before you publish it.

Lines invented over music or silence are dropped

Speech models sometimes write a polite line, like "Thank you.", over music, engine noise or silence. To catch these, a second small model scores every 10 milliseconds of the recording for voice; it is the same noise model behind the free Noise Remover. A subtitle survives only when a voice is heard for at least 30% of its time. Markers such as [Music] or ♪ are removed too.

The check also works the other way. If the subtitles cover less than 85% of the moments where a voice is heard, that block is transcribed again with a tenth of a second of silence in front, and the better attempt is kept. If the voice model cannot load, no line is dropped.

Choosing the spoken language or English output

Spoken language starts at Detect automatically and lists every language the tool is set up for. Naming the language means the model does not have to guess it. The result’s Language field repeats the setting, and says Detected automatically when the model chose.

Text language offers Same as spoken or English (translated). English (translated) uses the model’s own speech translation, so speech in a listed language comes out as English subtitles or text in one pass. It translates into English only; for other target languages, or for a spoken version, a dubbing tool is the next step.

Model download, processor and recording length

Accuracy offers Fast or Accurate. Fast uses the small model; Accurate uses a larger one and takes longer to run. Both run on your device’s processor, on a computer or a phone.

A phone transcribes up to 10 minutes per run, and a computer up to 60 minutes. Long recordings are worked through in blocks of about a minute, each cut at the quietest moment nearby so no word is split, and the progress line moves with every block. Cut a longer file into parts with the free Audio Trimmer.

What leaves your device during transcription

The recording is read from your device, turned into text in this browser tab and handed back as a file. It is not uploaded for transcription.

The page still uses the network. It loads its own code, downloads the speech model from a public model host on first use and the small voice model from Imagera, and can send an error report when a run fails. Those reports carry the file’s size and type, never its name or its sound.

Next steps for a finished transcript

A separate subtitle file only shows where the player or app supports it. The paid subtitle burner prints an SRT into the picture, so the words show everywhere. The paid dubbing studio goes further and speaks the whole video in another language.

The transcript can also become a script. The paid voice studio reads text aloud in a cloned or designed voice, and the free Audio Trimmer cuts the recording around the lines you want to keep.

How do you use the audio to text?

1. Choose the recording

   Pick a MP3, M4A, WAV, MP4, MOV or WebM file from your device: a voice memo, a podcast episode or a video with speech.

2. Set language, file type and accuracy

   Leave Spoken language on Detect automatically or name the language, choose SRT subtitles, VTT subtitles or Plain text, and pick Fast or Accurate under Accuracy.

3. Transcribe and check it

   Press Transcribe. The first run downloads the model, then progress moves block by block. Read the text and the Words and Subtitles counts before you save.

4. Save your result

   2 free downloads, 8 more after sign-in.

When is the free audio to text enough?

Audio to Text: questions and answers