Transcribe audio to text

Four ways to turn a recording into text

Typing it out yourself costs nothing but time, usually four to six hours per hour of clear audio and more with several speakers. It makes sense for a short clip or a recording nobody else may hear.

Built-in tools are worth trying first if you already pay for them. Microsoft Word's Transcribe accepts WAV, MP4, M4A and MP3 uploads, separates speakers, and allows 300 minutes of uploaded audio a month on a Microsoft 365 subscription (30,000 with a Copilot licence), according to Microsoft's support page. Google Docs voice typing listens to a microphone only, so it cannot transcribe a file.

A human transcription service is the choice when a verbatim record must be right the first time, for example as evidence. Rev lists human transcription at $1.99 per minute on its pricing page (read 8 November 2026), which is about $120 for an hour of audio.

An AI transcription tool is the practical option for most recurring work: interviews, meetings, lectures, calls. It works on any file, separates speakers, keeps timestamps, and the time you spend goes into checking rather than typing.

The settings that decide how good the transcript is

Most bad transcripts are caused by audio or settings, not by the transcription engine. In Klang, four settings matter:

• Language: Automatic detects the language for each part of the recording, which is what you want when people switch between, say, Swedish and English. For a short clip, or a strong accent, choosing the language is more accurate.
• Number of speakers: Automatic works for most recordings. Set a fixed number from 1 to 10 when voices sound alike or many people talk.
• Format: Transcription gives readable paragraphs split at each change of speaker, for interviews and meetings. Subtitles gives short timed segments for video.
• Word lists and name lists (Pro): teach Klang product names, abbreviations and the spelling of people's names, and save them as a transcribe mode for the next upload.

Before you record: a five-point checklist

If you control the recording, these five things improve any transcript more than any tool choice:

1. Put the microphone close to the speakers: a phone on the table between two people beats a laptop at the end of the room.
2. Record in a quiet room, with fans, music and notifications off.
3. Ask people not to talk over each other; overlapping speech is where every transcriber loses words.
4. Start with names and roles on the record ("Anna, project lead") so renaming speakers afterwards takes seconds.
5. Keep the original file. Converting a poor MP3 to WAV makes it bigger, not clearer.

How to check a transcript before you rely on it

An AI transcript is a first draft. How much you check depends on what the text is for. A quick skim is enough for your own notes; anything you will quote, publish or file deserves this pass:

1. Rename the speakers first, so every later correction lands in the right place.
2. Check every name, number, date and amount against the audio. These are the errors that matter and the easiest to miss.
3. Listen again wherever two people spoke at once or the sound drops.
4. For a quote, play the sentence before and after it, so you keep the context.
5. Export only once the text is checked, so the file and the transcript in Klang agree.

Which export format for which job

Klang exports the same transcript in several formats, with options to include speaker names and timestamps. Word, Excel, SRT and VTT export are part of Pro.

• Word (.docx): speaker labels, timestamps and optional comments. For reports, interview write-ups and anything you will edit.
• Excel (.xlsx): one row per segment. For coding interviews or filtering by speaker in research.
• SRT and VTT: subtitle files for video editors and web players. Avid users have their own subtitle format.
• Plain text, HTML, CSV and JSON: for pasting elsewhere, publishing, or processing the transcript in your own tools.
• The original audio file, so the recording and its text stay together.

Transcribing audio for a team or organisation

When recordings contain customers, patients, employees or citizens, the question is less which tool is most accurate and more where the audio goes. Before a team starts uploading, check these, and ask the same of any tool you compare:

• Where audio is processed and stored. Klang processes and stores it in Europe and is ISO 27001:2022 certified.
• Whether recordings train AI models. Klang never uses your content for training, on any plan.
• A data processing agreement you can sign. On Klang's Business plan it is signed in the account, alongside SSO, an audit log and verified domains.
• Retention and deployment. Enterprise adds a custom retention period, a private cloud with your own encryption key, or installation in your own environment.
• Who can see what. Transcripts are shared through folders, so a recording is visible only to the people in that folder.

Supported files and languages

Audio: MP3, WAV, M4A, AAC, OGG, OPUS, AIFF and WMA. Video: MP4, WebM, AVI, MKV and MOV. Each file can be up to 10 GB, and you can upload several at once. Public YouTube videos can be imported by link.

Languages: Swedish, English, Norwegian, Danish, German, French, Spanish, Dutch, Italian, Portuguese, Turkish, Finnish, Greek, Polish, Russian, Ukrainian, Japanese, Korean and Chinese (Mandarin). With the language on Automatic, a recording that moves between them is transcribed in the language actually spoken.

Audio to text, four ways compared

Figures from Microsoft's Transcribe support page and Rev's pricing page, read 8 November 2026:

Other tools and prices change. Check the provider's own page before you decide.

Frequently asked questions