How accurate is Audio to Text, and what makes a transcript worse?

It is good on clear speech from one person and noticeably worse with background noise, people talking over each other, strong accents, and names or technical words it has not seen. Treat the output as a draft to correct; the Text tab is editable for exactly that reason.

Which languages can Audio to Text transcribe?

The English-only model handles English. The multilingual models also cover Spanish, French, German, Portuguese, Hindi, Japanese, Chinese and Russian, and can detect the language automatically, though naming it explicitly gives better results.

How long does Audio to Text take to transcribe an hour of audio?

Roughly the length of the recording on a typical laptop with the small model, and less with the smaller ones. The first run also downloads the model once. The page shows each step as it happens: decoding, downloading and listening.

How do I get subtitles for a video as SRT or VTT with Audio to Text?

Drop the video file in, transcribe it, open the Subtitles tab, set the line length and cue length you want, and download SRT or VTT. Load the file next to your video in a player or editor; if the timing is a little off, shift all the timings first.

Is my recording uploaded by Audio to Text, and what is downloaded once?

The recording is decoded and recognised on your device and is never uploaded. What is downloaded, once, is the recognition model itself from a public model host, which your browser then caches so later runs start immediately and can work offline.

Is there a maximum file length for Audio to Text?

There is no set limit, but decoding needs the whole file in memory, so very large video files can fail. For long videos, extract the audio first and transcribe that, or split the recording into parts.

ArrayKit

Audio to Text Converter

Turn a recording, or the sound of a video, into an editable transcript and SRT or VTT subtitles. Recognition runs on your device and the file is never uploaded.

Audio to Text Converter

Turn a recording, or the sound of a video, into an editable transcript and SRT or VTT subtitles. Recognition runs on your device and the file is never uploaded.

Recordings are often interviews, calls and meetings. The audio is decoded and recognised on this device and is never uploaded; only the model itself is downloaded, once.

Convert the subtitles to another format

About Audio to Text

Audio to Text turns a recording, or the soundtrack of a video, into text you can edit and subtitles you can use. Speech recognition runs on your device in a background worker: the file is decoded, a model is loaded, and the result comes back as timed segments. From those the tool builds an editable transcript with search, timestamped paragraphs, JSON, and subtitle cues cut at sentence ends within the limits you set for characters per line, lines and seconds on screen. You can correct the wording of any cue, shift every timing by a fraction of a second, and download SRT or WebVTT. The model is fetched once and cached by your browser; the recording itself never leaves your computer.

Features

Reads audio files and the audio track of video files
Three on-device model sizes, from a quick English-only one to a slower multilingual one
Editable transcript with search that highlights every match
Subtitle cues limited by characters per line, lines and seconds on screen, cut at sentence ends first
Edit the words of any cue, and shift every timing earlier or later
Download SRT or WebVTT, plain text, timestamped paragraphs or JSON
Word count, duration, speaking rate and cue count
Progress for each step: decoding, the one-time model download, and recognition

How to use the Audio to Text

Drop an audio or video file onto the page
Choose a model and the spoken language
Press Transcribe and wait while the file is decoded, the model loads and the speech is recognised
Correct the text or the cues, then copy the text or download the SRT or VTT

Example

Input

interview.mp3 — first 10 seconds
Max 42 characters per line · 2 lines · 6 s per cue

Output

1
00:00:00,000 --> 00:00:03,400
Welcome back to the show. Today we are
talking about how maps get made.

2
00:00:03,400 --> 00:00:07,100
Our guest has surveyed coastlines
on four continents.

3
00:00:07,100 --> 00:00:09,800
Let's start at the beginning.

Cues break at sentence ends first, and no line runs past the limit you chose.

Common errors & troubleshooting

The browser says it could not decode the file. — Try MP3, WAV, M4A or OGG. Some video containers, such as MKV and AVI, cannot be read by a browser at all; extract the audio to MP3 or WAV first and transcribe that.
Names, numbers or technical terms are wrong. — That is where recognition is weakest. A larger model helps a little, and setting the spoken language explicitly helps more; read the result against the recording either way.
Subtitle lines are too long, or each cue stays up too long. — Open the Subtitles tab and lower the characters per line or the maximum cue length. The cues are rebuilt at once, and the new limits are applied everywhere.
Subtitles appear slightly early or late in the video player. — Use Shift all timings to move every cue by a fraction of a second. A negative value makes them earlier and stops at the start of the video.

Frequently asked questions

How accurate is Audio to Text, and what makes a transcript worse?: It is good on clear speech from one person and noticeably worse with background noise, people talking over each other, strong accents, and names or technical words it has not seen. Treat the output as a draft to correct; the Text tab is editable for exactly that reason.
Which languages can Audio to Text transcribe?: The English-only model handles English. The multilingual models also cover Spanish, French, German, Portuguese, Hindi, Japanese, Chinese and Russian, and can detect the language automatically, though naming it explicitly gives better results.
How long does Audio to Text take to transcribe an hour of audio?: Roughly the length of the recording on a typical laptop with the small model, and less with the smaller ones. The first run also downloads the model once. The page shows each step as it happens: decoding, downloading and listening.
How do I get subtitles for a video as SRT or VTT with Audio to Text?: Drop the video file in, transcribe it, open the Subtitles tab, set the line length and cue length you want, and download SRT or VTT. Load the file next to your video in a player or editor; if the timing is a little off, shift all the timings first.
Is my recording uploaded by Audio to Text, and what is downloaded once?: The recording is decoded and recognised on your device and is never uploaded. What is downloaded, once, is the recognition model itself from a public model host, which your browser then caches so later runs start immediately and can work offline.
Is there a maximum file length for Audio to Text?: There is no set limit, but decoding needs the whole file in memory, so very large video files can fail. For long videos, extract the audio first and transcribe that, or split the recording into parts.

Related tools

Audio to PDF Transcript — Transcribe a recording on your device and export it as a PDF, SRT or VTT.
Subtitle Converter — Convert SRT, WebVTT, SBV and ASS subtitles, and fix timing that drifts.
Voice Recorder — Record from your microphone, trim the take and save it as WAV or the original format.
AI Text Summarizer — Summarize long text into a short summary — the AI model runs in your browser, nothing is uploaded.
Audio Converter — Convert audio between MP3, WAV, FLAC, M4A and OGG with bitrate control.
Extract Audio from Video — Extract the audio from a video as MP3, WAV or M4A, in your browser — nothing uploaded.
Video Trimmer — Cut a clip from a video by start and end time, in your browser — nothing uploaded.
PDF OCR — Recognise the text in a scanned PDF and make it searchable.

Audio to Text Converter

Audio to Text Converter

About Audio to Text

Features

How to use the Audio to Text

Example

Common errors & troubleshooting

Frequently asked questions

Related tools

More Video tools