Speech to text: how it works

What is speech to text?

Speech to text (also called speech recognition) is software that turns spoken words into written text. The model listens to the sound, recognises the words and adds punctuation. The same technology sits behind the microphone on your phone keyboard and behind services that write out whole recordings, but they are used for different things.

Dictating or transcribing a recording

Writing out a recording is called transcription. What that involves is explained on what is transcription.

Dictating on your phone and computer

For live dictation you usually need nothing extra, it is already built into your device:

• iPhone and iPad: tap the microphone on the keyboard. If it is not there, turn on Dictation in Settings > General > Keyboard.
• Android: in the Google keyboard (Gboard), tap the microphone above the keys.
• Windows: in a text field, press Windows key + H to start voice typing.
• Mac: turn on Dictation in System Settings > Keyboard and start it with the shortcut shown there.
• Google Docs and Word: Google Docs has Voice typing under Tools (in Chrome), Word in Microsoft 365 has a Dictate button.

These features are made for one person dictating calmly in short pieces. For a conversation between several people, a long recording or a lot of jargon they are less suitable: there are no speakers and you cannot replay what exactly was said. Depending on your device and settings, the speech is processed on the device itself or on the manufacturer's servers.

When a transcript works better

• It lasts longer than a few minutes. You do not want to dictate an interview, lecture or meeting live.
• Several people are talking. A transcript labels every turn, see speaker recognition.
• You want to be able to replay. Click a word and you hear the audio from that point.
• There are names and jargon. You add them to your vocabulary once.
• The recording already exists. You simply upload a voice memo, a WhatsApp message or a Teams recording.
• The conversation is confidential. At Murmel everything is processed and stored in the Netherlands.

Speech to text with Murmel

1. 1
   Open the StudioLog in at app.murmel.nl on your laptop or phone. On an iPhone use Safari, on Android Chrome. There is no app to install.
2. 2
   Press Record and talkYou see a live waveform, so you know the sound is coming in. Already have a recording, such as a voice memo? Choose Upload instead.
3. 3
   Stop and read the textWhen you stop, Murmel converts the recording. An hour of audio is usually ready within a minute, with speakers and punctuation.
4. 4
   Correct and exportFix what is needed and export to Word, PDF, text or subtitles. See export.

Recording on your phone

Open the Studio in your phone's browser and press Record. Keep the screen on and stay in the browser: if the phone locks, the browser may pause the microphone. A browser recording lasts up to about two hours. For longer sessions, record with your phone's recorder app and upload the file. More tips at record in the browser, voice memo to text and WhatsApp voice message to text.

A model of its own for Dutch

Murmel uses murmel-flash, a speech model of its own trained from scratch on Dutch speech, with English as a second language. It is built for the way Dutch people actually talk: accents, informal speech and names. Recordings are processed at Leafcloud in Amsterdam and stored at Scaleway in the Netherlands. How the model was measured is described in the Murmel v5 benchmark.

Prepaid credits cost €1.45 per hour of audio incl. VAT, from €6.05 and valid for 6 months. Basic costs €10.89 per month incl. VAT for 12 hours. See pricing.

Frequently asked questions