A podcaster wants show notes, a journalist wants the quotes from an interview, a student wants the lecture in writing. Each has a recording and needs the words in it as text. There are four ways to get there: type it yourself, pay a person, use the dictation built into your computer, or use a transcription program. This guide covers what each one asks of you, and how to check the result whichever you choose.
Method 1: Type It Yourself
Play a few seconds, pause, type what you heard, and go back when you missed a word. It costs nothing but your time, and you hear every word yourself.
- Time: Longer than the recording itself, since you stop and go back, and longer still with fast speech, several speakers or a poor microphone. To know your own pace, time yourself on five minutes of the recording and multiply.
- Cost: Your time.
- What helps: A player that slows the audio down and skips back a few seconds from the keyboard, and headphones.
- Suits: Short clips, or a passage you will quote word for word.
Method 2: Pay a Person
A transcription service sends your file to a trained transcriptionist and returns the text. Some services also offer transcription by machine; the ones that offer it by a person charge by the minute of audio, and some offer it in English only. Which services offer it, and in which languages, is in Audio Transcription Software Compared, each fact from the service’s own pages.
- Time: Set by the service, and by the turnaround you pay for.
- Cost: By the minute of audio: a weekly podcast or a steady stream of interviews adds up.
- What you get: A person has listened to every word, which is what difficult audio, or a record where every word matters, calls for.
- Suits: Legal or medical records, and recordings a program struggles with.
Method 3: Use the Dictation Built Into Your Computer
Your computer probably came with dictation: you speak, and it types. It is made for a voice speaking to it, not for a recording.
- Google Docs voice typing types what you say into a document, in Chrome, Edge or Safari; its spoken punctuation commands (“period”, “new paragraph”) work in English only (Google Docs Help). It does not take an audio file.
- Apple Dictation enters text anywhere you can type on a Mac, with no time limit, and stops after 30 seconds without speech (Mac User Guide). It does not take an audio file either.
- Microsoft Word’s Transcribe is the exception: with a Microsoft 365 subscription it transcribes an uploaded .wav, .mp4, .m4a or .mp3 file and labels its speakers, up to 300 minutes of uploaded audio a month (Microsoft Support).
To use voice typing or Dictation on a recording, you would play it aloud to the microphone, in real time, and the result has no timestamps and no speaker labels.
Method 4: Use a Transcription Program
A transcription program takes the file and returns the text, usually with timestamps, often with the speakers told apart, and in formats such as captions or a Word document. Under it is a speech recognition model trained on a large amount of recorded speech.
How good the text is depends on the model and on the audio. One speech recognition model, trained on 680,000 hours of audio, was measured on many test sets, and its authors report that, compared with human transcribers, it approaches their accuracy and robustness. They also report what goes wrong: accuracy uneven across languages, and, on long recordings, text repeated in a loop or unrelated to the audio (Radford et al., 2022).
- Time: The program’s, not yours: you send the file and come back to the text.
- Cost: Many have a free plan, and paid plans charge in different ways: by the hour of audio, by the person, or by the minute. Compare what an hour of your audio costs.
- What you get: A draft. It is best on clear audio from a close microphone, and worse with noise or people talking over each other.
- Suits: Podcasts, interviews, lectures, meetings and subtitles, as long as someone reads the result.
Check the Transcript, Whichever Method You Use
A person can mishear, and so can a program. Before you quote, publish or file a transcript:
- Names and terms. People’s names, places, brands and the vocabulary of your field: whoever does not know them has to guess.
- Numbers. Fifteen and fifty sound alike; check every figure, date and amount against the audio.
- The negatives. A lost “not” reverses a sentence.
- Who said it. If the transcript labels speakers, check the label on every quote you use.
- Words nobody said. A program can add text, or repeat a passage, especially in silence or noise (Radford et al., 2022, §6). If a sentence surprises you, listen to that moment.
How to Transcribe a Recording with Verbato
Verbato is a transcription program of the fourth kind. It also writes documents from the transcript: notes, summaries, letters, the documents your work asks for.
- Drop a file, or record in your browser.
- Or paste a link.
Verbato's app. The content was written to show it.
- The mark beside a section.
- The line it was written from, played from its moment.
Verbato's app. The content was written to show it.
- Create a free account at verbato.io. No card is asked for.
- Send the recording. Upload an audio or video file, paste a link, or record in your browser. You can also send a voice note by Telegram, or a recording by email.
- Press Start. Verbato transcribes 80+ languages. The kind of recording sets the documents written from it; your account has a usual one, and you can change it on the recording’s page while it processes. On a paid plan it can tell the speakers apart.
- Check the transcript. Press any segment to play the audio from that moment. On a paid plan you can correct a segment, or find and replace a word throughout.
- Export it as plain text, Word, PDF, SRT, VTT or JSON, on every plan.
You can try it on your own recording before paying anything. What each plan includes (the hours, speaker identification, the longest recording) is on the pricing page.
Which Method Should You Use?
- A clip of a few minutes, or a passage you will quote: type it yourself.
- A record where a person must have heard every word: pay for transcription by a person.
- Notes you are speaking, not a recording: the dictation built into your computer.
- Recordings of any length, often: a transcription program, and read the result before you rely on it.
For the programs and services side by side, read Audio Transcription Software Compared.