Speech to text

Related: AI video transcriber · subtitles for an AI drama · AI text to speech.
How speech to text works
The audio is cut into short slices and turned into a picture of its frequencies over time. A model trained on many thousands of hours of speech with matching transcripts reads that picture and predicts the most likely words, using the whole sentence to decide between words that sound alike, such as “their” and “there”. Modern systems then add punctuation and capitals, and some mark who is speaking and when each word starts.
Voice to text or audio to text: live dictation and file transcription
| Voice typing (live) | Transcribe audio to text (files) | |
|---|---|---|
| Input | Your voice into a microphone | A recording: meeting, interview, podcast, voice note |
| Speed | Words appear as you speak | Minutes of audio in a fraction of the time |
| Speakers | One | Often several, labelled |
| Output | Text in the app you are typing in | A transcript, a document or a subtitle file |
| Typical use | Messages, notes, writing hands-free | Notes, quotes, captions, search |
Transcribe audio to text accurately
- Record close to the microphone in a quiet room; this matters more than the tool.
- Set the language and region instead of relying on detection.
- Add names, brands and jargon to the tool’s custom vocabulary if it has one.
- Read the transcript against the audio at a few points, then check every number and name.
- Keep the timestamps: they let you jump from a quote back to the moment it was said.
Speech to text free: what free tools usually limit
Built-in dictation on phones and computers is free for voice typing. Free audio to text converters usually cap the minutes per month or the length of each file, limit export formats, or leave out speaker labels. Tools that run on your own computer keep recordings private; online tools upload the audio, so read how long they keep it and whether it trains their models.
Where speech to text still goes wrong
- Names and rare words: spelled the way they sound.
- Numbers: “fifteen” and “fifty” are easy to swap.
- Crosstalk: two people speaking at once loses words from both.
- Music and noise: speech under loud music drops out.
- Accents and mixed languages: switching language mid-sentence confuses many tools.
Speech to text for subtitles
Subtitles need more than a transcript: the text is split into short lines and each line is timed to the moment it is spoken, then saved as an SRT or VTT file or burned into the picture. When the words are already written, as in a scripted video, the better approach is to keep the script’s text and use speech recognition only to find the timing.
A transcript checklist
- Language and region set.
- Names, numbers and jargon checked.
- Speakers labelled the way you need.
- Timestamps kept.
- Permission to record, transcribe and publish.
| Transcribing speech | Subtitles from a script | |
|---|---|---|
| Words come from | What the recogniser hears | The script as written |
| Names and spelling | Need checking | Exactly as written |
| Timing comes from | The audio | The voices in the video |
| Best for | Recordings without a script | Videos made from a script |
FAQ
Does SceneMixer have a speech to text tool?
Not as a separate tool for your recordings: SceneMixer takes text, pictures and videos, not audio files. Inside the product, speech recognition is used to time subtitles and to read the lines of a clip you remake.
Are SceneMixer subtitles a transcription?
No. The words are your script’s lines, so names come out exactly as you wrote them; SceneMixer recognises the speech only to find when each line is said.
Can I get an SRT file from SceneMixer?
Yes, for a whole episode: tick Add subtitles when you export it on SceneMixer, then use Download SRT.
How does SceneMixer read the lines of a video I upload?
On the Remake tab every shot of your clip is listened to by a speech recogniser, and SceneMixer quotes what it heard into the new storyboard instead of guessing from lip movement.
What if SceneMixer cannot make out a line?
It is left out rather than invented. Compare the storyboard with your clip and add the missing line to that shot in SceneMixer’s editor.
Do subtitles cost credits on SceneMixer?
No. Subtitles cost no credits on SceneMixer; they are available on accounts that have made at least one purchase.
Subtitles that spell names right
Write the script, render the episode, and export subtitles timed to the voices.
Updated November 20, 2026