How to transcribe audio to text

Eight transcription tools compared on what an hour of audio really costs, a seven-step method, and the free-plan limits nobody explains.

Short answer: if you pay for Microsoft 365, upload the file to Word for the web and click Transcribe, up to 300 minutes a month. For free and private, run OpenAI Whisper on your own computer. For many hours a month, Notta costs the least per hour; for meetings, Otter; for a transcript you must quote word for word, Rev's human service.

We compared eight tools on their pricing and help pages line by line, worked out what an hour of audio really costs on each plan, and read what users report in public reviews. Below: the ranking, the method in seven steps, and the free-plan limits that the plan pages bury.

Romain CochardCEO of Hack'celeration
Updated 20268tools compared7method steps7traps and tips

The ranking is editorial: a sponsored placement is always labeled on its card, and no affiliate link changes a score.

The ranking

8 ways to transcribe audio to text, side by side

Transcription turns speech into written text. There are three ways to do it. Built-in tools (Word for Microsoft 365, Voice Memos on iPhone, YouTube captions) cost nothing extra but have monthly caps and few editing features. AI transcription apps (Notta, Otter, Happy Scribe, Sonix, Descript) upload your file, run a speech recognition model and return an editable transcript in minutes, with speaker labels and exports. Human services (Rev, GoTranscript) have a person type or correct the text, for hours or days of waiting and about $2 a minute.

Before you compare prices, know the units. Most apps sell minutes of audio per month: a one-hour recording uses 60 of them, and unused minutes are usually lost at the end of the month. Some sell hours (Sonix) or charge per minute for human work (Rev, Happy Scribe). Free plans add a second limit, the length of one recording or the number of files. What to demand from any tool: a free test on your own audio, speaker labels, an editor that plays the audio under the text, and the export you need (DOCX, TXT or SRT). Prices below are entry prices in US dollars, before tax. For a ranking of AI tools only, see our best AI transcription software.

Best forWhat it does bestFree planTeam sizeVisit
1Microsoft Word (Transcribe)Best if you already pay for Microsoft 365Free with a tool you already pay4.1/5Included in Microsoft 365—Microsoft 365 usersVisit →
2NottaBest paid allowance for the priceAbout $0.27 per hour at the cap4.1/5$8.17/mo (yearly)✓Several hours a monthVisit →
3OtterBest for meetings and live notes300 free live minutes a month4.0/5Free, Pro $8.33/mo (yearly)✓MeetingsVisit →
4Happy ScribeBest for many languages and subtitles150+ languages, SRT export4.0/5Free, Basic $8.50/mo (yearly)✓Multilingual and subtitlesVisit →
5RevBest when the transcript must be exactHuman transcription on demand3.9/5Free, human $1.99/min✓Transcripts you must quoteVisit →
6OpenAI WhisperBest free option for privacyRuns offline, no cap3.9/5Free (open source)✓Confidential audioVisit →
7SonixBest pay-as-you-go pricingNo subscription needed3.8/5$10/hour—Occasional filesVisit →
8DescriptBest for editing audio by editing textEdit the audio through the text3.4/5Free, Hobbyist $16/mo (yearly)✓Podcast and video editingVisit →

Prices marked yearly are monthly rates billed annually. Microsoft 365 Personal is often discounted for the first year: compare at the renewal price. Several guides also rank Trint, a newsroom tool priced for teams, Temi and Scribie, two low-cost per-minute services, and TurboScribe, an unlimited AI plan whose pricing page we could not open. For meeting notetakers such as Fireflies, see our Otter alternatives; for ElevenLabs Scribe, our ElevenLabs review.

Is your tool missing from this page? Email us: hello@hackceleration.com

1
Best if you already pay for Microsoft 365

Microsoft Word (Transcribe)

4.1/5

Word has a transcription tool built in, and it is the first result Google shows for this question. You open Word for the web, click Dictate, then Transcribe, upload an audio file, and Word returns the text split by speaker, labeled "Speaker 1", "Speaker 2" and so on. You rename a speaker once and the change applies everywhere, then add the whole transcript, or only the sections you need, to your document.

The limit is 300 minutes of uploaded audio a month with a Microsoft 365 subscription (30,000 with a Microsoft Copilot license), and Word reads .wav, .mp4, .m4a and .mp3 files. Microsoft 365 Personal costs $99.99 a year or $9.99 a month for one person (Microsoft regularly cuts the first year, so check the renewal price), and Family costs $129.99 a year. If you already pay for it, five hours of transcription a month cost you nothing extra. The catch nobody mentions: in desktop Word, Transcribe works only on Windows, and only for work accounts ("Commercial Tenants"). With a personal plan, or on a Mac, use Word for the web in Edge or Chrome. Transcribe handles 80+ locales (a language in a given country or region).

Standout features
  • Speaker labels you can rename in one click
  • 300 upload minutes a month with Microsoft 365
  • 80+ language locales
  • Transcript inserted straight into a Word document
+Pros
  • ✓Nothing extra to pay if you already have Microsoft 365
  • ✓Speaker separation included
  • ✓Familiar editor for the cleanup
−Cons
  • ✗Desktop app only on Windows work accounts
  • ✗Playback highlights sections, not single words
  • ✗Monthly cap, no top-up for personal plans
Verdict

The first thing to try if Microsoft 365 is already on your card. Use Word for the web, and keep files under five hours a month.

See how Word Transcribe works → →
2
Best paid allowance for the price

Notta

4.1/5

Notta records meetings, transcribes uploaded files and links, and adds AI summaries. Where it stands out is the allowance: Pro costs $8.17 a month billed $97.99 a year, for 1,800 transcription minutes a month, recordings up to 5 hours each and 100 file uploads. Used to the full, that is about $0.27 per hour of audio, against about $0.42 on Otter Pro billed yearly.

The free plan is where people get caught. It shows 120 minutes a month and 50 file uploads, but each conversation stops at 3 minutes: fine to test accuracy on a clip, useless for a full interview. Minutes do not carry over to the next month. Business costs $16.67 a month per seat, billed $199.99 a year, with unlimited transcription. Zapier and CRM integrations come with Business; to send transcripts to your notes or CRM automatically, see our Notta Zapier integration.

Standout features
  • 1,800 minutes a month on Pro
  • Recordings up to 5 hours on paid plans
  • Meeting bot, file upload and links in one app
  • AI summaries included
+Pros
  • ✓About $0.27 per hour at the monthly cap
  • ✓Long recordings accepted
  • ✓Clean web and mobile apps
−Cons
  • ✗Free plan cut at 3 minutes per conversation
  • ✗Pro price shown is the yearly rate
  • ✗Unused minutes lost each month
Verdict

The value pick once you transcribe more than a few hours a month. Test accuracy on the free plan, then pay yearly.

See Notta's plans → →
3
Best for meetings and live notes

Otter

4.0/5

Otter joins your calls, transcribes them live and writes a summary afterwards. For meetings it is the reference most guides start with. The free Basic plan includes 300 transcription minutes a month, with each conversation capped at 30 minutes, so roughly ten short meetings.

The trap for anyone who wants to transcribe a recording they already have: the free plan allows only 3 file imports in total, for the life of the account. The 300 minutes are for live recordings. Pro costs $8.33 per user a month billed yearly ($99.96 a year), or $16.99 month to month ($203.88 a year), for 1,200 minutes, 90 minutes per conversation and 10 imports a month. Students and teachers with a .edu address pay $79.99 a year. Reviews from 2025 and 2026 report two recurring problems: proper names and accents, and a yearly renewal charged to people who had stopped using it, so set a reminder before the renewal date. If meetings are your main need, our Otter alternatives page compares six AI notetakers, and the Otter Zapier integration pushes transcripts into other apps.

Standout features
  • Live transcription in Zoom, Meet and Teams
  • 300 free minutes a month
  • Custom vocabulary for names
  • Student discount on Pro
+Pros
  • ✓300 free minutes a month for meetings
  • ✓Good speaker separation in clean audio
  • ✓Summaries and action items
−Cons
  • ✗Only 3 file imports ever on the free plan
  • ✗Struggles with names and accents
  • ✗Annual plan renews automatically
Verdict

The right tool for meetings. For a stack of existing recordings, Notta or Sonix cost less.

See Otter's plans → →
4
Best for many languages and subtitles

Happy Scribe

4.0/5

Happy Scribe transcribes audio and video in 150+ languages and dialects, then exports to DOCX, TXT or SRT for subtitles. Its editor plays the audio under the text, which makes corrections fast, and every plan includes one glossary and one style guide, with unlimited ones on Business. When the AI is not enough, it sells human proofreading from $2.00 a minute ($1.90 on Business) in the same interface.

The free plan gives 10 minutes of uploaded files, a test, not a plan, and free MP4 exports carry a watermark. Basic costs $17 a month for 120 minutes a month, or $8.50 a month billed yearly ($102 instead of $204), but the yearly plan gives 1,440 minutes for the year rather than 120 a month: the same total, usable in one busy month. Pro costs $19 a month billed yearly or $29 monthly, for 600 minutes a month. Extra minutes cost $0.20 each, so $12 for an hour. For subtitles only, our subtitle generator ranking goes further.

Standout features
  • 150+ languages and dialects
  • DOCX, TXT and SRT exports
  • Human proofreading in the same app
  • Yearly minutes usable at any time of the year
+Pros
  • ✓Widest language coverage here
  • ✓Subtitles and transcripts from one upload
  • ✓AI and human options side by side
−Cons
  • ✗Free plan limited to 10 minutes of files
  • ✗Watermark on free video exports
  • ✗Minutes cost more than Notta's
Verdict

Pick it for multilingual work and subtitles. Take the yearly plan if your volume comes in bursts.

See Happy Scribe's plans → →
5
Best when the transcript must be exact

Rev

3.9/5

Rev sells both: AI transcription inside a subscription, and human transcription by the minute. The free plan includes 45 AI minutes a month for one user. Essentials costs $25.49 per seat a month billed $305.90 a year, with 5,000 AI minutes per user a month and a discount on human work (3% on monthly billing, 10% on yearly).

Human transcription starts at $1.99 a minute, so $119.40 for a one-hour interview. That is the price of being able to quote a recording in court, in print or in a thesis without re-listening to every line. Legal formatted transcripts are priced by the page and sold separately. GoTranscript, which several guides rank, sells the same kind of human service.

Standout features
  • AI and human transcription in one account
  • 5,000 AI minutes a month on Essentials
  • Legal formats priced by the page
  • Mobile app for dictation
+Pros
  • ✓Human option for audio the AI gets wrong
  • ✓Large AI allowance on paid plans
  • ✓Discount on human work for subscribers
−Cons
  • ✗Human transcription adds up fast
  • ✗Paid plan priced per seat
  • ✗Free allowance small for files
Verdict

The safe choice when an error costs more than the transcript. For routine notes, an AI plan is enough.

See Rev's pricing → →
6
Best free option for privacy

OpenAI Whisper

3.9/5

Whisper is OpenAI's speech recognition model, and its code and model weights are released under the MIT license: you can run it on your own computer, offline, for as many hours as you like, and nothing leaves the machine.

You install Python, ffmpeg and the package, then run one line, for example whisper interview.mp3 --model turbo. The model size sets the trade-off: tiny needs about 1 GB of graphics memory, turbo about 6 GB and runs about 8 times faster than large, which needs about 10 GB. On a laptop without a strong graphics card it still works, only slower. What you give up: speaker labels without an extra tool, an editor that plays the audio under the text, and support. Developers who want the same thing as an API can compare providers in our Whisper alternatives.

Standout features
  • MIT license, code and weights
  • Runs offline on your computer
  • Models from tiny to large
  • Text, SRT and VTT output
+Pros
  • ✓Free with no monthly cap
  • ✓Audio never leaves your machine
  • ✓Accurate on clear speech with the larger models
−Cons
  • ✗Command line, no editor
  • ✗Speaker labels need an extra tool
  • ✗Needs a decent graphics card to be fast
Verdict

The best free route if you can open a terminal. Use turbo for speed, large for difficult audio.

See Whisper on GitHub → →
7
Best pay-as-you-go pricing

Sonix

3.8/5

Sonix charges $10 per hour of audio with no subscription, which is the simplest way to pay for an occasional interview or lecture. If your volume is steady, Core costs $25 a month for 5 hours ($5 an hour) or $275 a year, and any extra hour on a plan is billed at $10.

The editor syncs text and audio, and Sonix also translates transcripts. One pricing detail matters for teams: an extra seat costs $25 a month and does not add hours, because the included hours belong to the workspace. Two people sharing Core still share 5 hours.

Standout features
  • $10 an hour with no subscription
  • Text synced to audio in the editor
  • Translation of transcripts
  • Plans from $25 a month
+Pros
  • ✓No commitment for occasional files
  • ✓Clear hourly pricing
  • ✓Good editor for corrections
−Cons
  • ✗Free trial limited to 30 minutes
  • ✗Extra seats add no hours
  • ✗Costs more than Notta at volume
Verdict

Good for occasional files. Past five hours a month, a subscription elsewhere costs less.

See Sonix's pricing → →
8
Best for editing audio by editing text

Descript

3.4/5

Descript is an audio and video editor built on the transcript. You import a recording, it transcribes it, and deleting words in the text deletes them in the audio. For a podcast or a YouTube video that is faster than any timeline. If all you need is text, it is more tool than you need, and our Descript alternatives list lighter editors.

The free plan includes 60 minutes of media a month and 100 AI credits, once, with exports at 720p and without watermark-free export. Hobbyist costs $16 a month billed yearly or $24 monthly for 10 media hours a month. Credits go fast: a 2026 review describes 800 credits spent cleaning filler pauses on one short reel. Our Descript review details the AI credits, and the Descript pricing breakdown compares every plan.

Standout features
  • Edit audio by editing the transcript
  • Filler word removal
  • Free plan with 60 minutes a month
  • Studio Sound cleanup
+Pros
  • ✓Unique text-based editing
  • ✓One tool from transcript to published episode
  • ✓Good for multi-track podcasts
−Cons
  • ✗Overkill for plain transcripts
  • ✗Free AI credits are one-time
  • ✗Paid plans cost more per hour of text
Verdict

Worth it if the transcript is a step towards an edited episode. For text alone, pick another tool.

See Descript's plans → →
The method

How to transcribe audio to text in 7 steps

Most bad transcripts come from bad audio or the wrong tool for the file, not from the AI. Here is the order we recommend, from the reason you need a transcript to the file you archive. If you are recording a show, our guide to starting a podcast covers microphones and recording apps.

01

1. Decide what the transcript is for

Write down what you will do with the text, because it decides the tool. Notes from a meeting can live with a few errors: an AI tool is enough. A quote for an article, a legal record or a research interview you will analyze line by line needs either a careful review or a human transcriptionist. Subtitles need an SRT file with timings, not a Word document. Five minutes of thinking here avoids paying twice.

Time
5 minutes
Output
AI, AI plus review, or human
02

2. Check what you already have

Before you sign up anywhere, look at the tools on your devices. With Microsoft 365, Word for the web transcribes 300 minutes of uploaded audio a month. On an iPhone 12 or later, Voice Memos shows a transcript of each recording (tap More, then View Transcript, or Copy Transcript to paste it elsewhere), in English and nine other languages, though not in every country. For a YouTube video, the automatic captions already are a transcript. Google Docs voice typing, often suggested, only listens to your microphone: it cannot read an audio file, and playing a recording into the mic loses a lot of words.

Cost
$0 with tools you already have
Limit
300 min/month in Word
03

3. Prepare the audio

Accuracy depends on the recording more than on the tool. Speech recognition loses words to background noise, mispronunciations, accents and people talking over each other. If you are recording, put the phone or microphone close to the speakers, one per person for a podcast, and avoid cafés. If the file already exists, convert it to a common format (MP3, M4A or WAV all work in Word and Whisper) and cut the long silences at the start and the end: you pay for them in minutes.

Formats
MP3, M4A, WAV, MP4
Biggest gain
A closer microphone
04

4. Pick the tool for the length and the volume

Match the free plan to your file before you upload. A 3-minute voice note fits Notta's free plan; a 25-minute call fits Otter's free plan, which stops at 30 minutes per conversation; a 2-hour lecture needs Word (300 minutes a month), Whisper, or a paid plan. For a one-off hour, Sonix at $10 is simpler than a subscription. For several hours a month, Notta Pro is the cheapest per hour. For audio you will quote, Rev's human service at $1.99 a minute.

One file
Sonix, $10 an hour
Every month
Notta Pro, $97.99 a year
05

5. Turn on speaker labels and add your vocabulary

Before you start the transcription, set the language of the recording instead of letting the tool guess, and turn on speaker detection if there is more than one voice. Word, Otter, Happy Scribe and Rev all separate speakers; with Whisper, speaker labels take an extra tool. If the tool has a custom vocabulary or glossary, add the names of people, products and places: that is where AI transcripts fail most often.

Fixes most
Names and jargon
Set
Language and speakers
06

6. Review with the audio playing

Read the transcript while the audio plays, at 1.25x or 1.5x speed, and correct as you go. Editors that highlight the word being spoken (Happy Scribe, Sonix, Descript, Otter) make this much faster than Word. Check every number, name and negation against the audio: "can" and "can't" are the classic swap. For a one-hour interview, a careful review of a good AI transcript takes far less time than typing it yourself.

Speed
1.25x to 1.5x
Check
Names, numbers, negations
07

7. Export in the right format and file it

Export DOCX or TXT for reading and quoting, SRT or VTT for subtitles, and keep the timestamps if you may need to find a passage again. Name the file with the date and the speakers. Then delete the recording from the cloud tool if it was confidential: on a cloud service the audio stays in your account until you remove it.

Text
DOCX, TXT
Subtitles
SRT, VTT
Traps and tips

What the plan pages don't tell you

Seven details from the pricing and help pages, and from user reviews, that change which tool is cheapest for you. If you also edit the audio afterwards, our podcast software ranking covers the editors.

01

Notta's 120 free minutes are 3-minute clips

Notta's free plan advertises 120 minutes a month, but each conversation is cut at 3 minutes. That is 40 short clips, not a two-hour interview, and exporting a transcript only starts on Pro. Use the free plan only to test accuracy on your accent and your audio, then decide.

Free cap
3 min per conversation
02

Otter's free plan imports 3 files, ever

The 300 free minutes on Otter are for live recordings. Imported files are limited to 3 for the life of the account, playback runs at 1x only, and the history keeps your 25 most recent conversations. If your recordings already exist, start with Word, Whisper or a paid plan, and keep Otter for meetings.

Free imports
3 lifetime
03

Word Transcribe on a personal plan: use the web version

In desktop Word, Transcribe is reserved for Windows work accounts. With Microsoft 365 Personal or Family, or on a Mac, open Word for the web in Edge or Chrome: the feature is there, with the same 300 minutes a month.

Browsers
Edge, Chrome
04

Yearly minutes can be spent in one month

Happy Scribe's yearly Basic plan gives 1,440 minutes for the year instead of 120 a month. If your work comes in bursts (a field study, a conference season), the yearly plan lets you spend most of it in a single month, for $102 instead of $204 on monthly billing.

Basic yearly
1,440 min a year
05

On Sonix, a second seat adds no hours

A Sonix seat costs $25 a month and the hours belong to the workspace, so a second person shares the same 5 hours on Core. For a team, compare the cost per hour, not per seat.

Extra seat
$25/month, 0 hours
06

Pay for humans only on the passages that matter

Human transcription at $1.99 a minute is $119.40 an hour. If only 10 minutes of a long interview will be quoted, transcribe the whole file with AI, then send just those minutes to a human or check them yourself against the audio.

Human, 1 hour
$119.40 at Rev
07

Set a reminder before a yearly renewal

Yearly plans renew automatically. Reviews from 2026 describe people charged for another year of a transcription tool they had stopped using. Put the renewal date in your calendar the day you subscribe.

When
The day you subscribe
FAQ

Transcribing audio to text, your questions

  • Can Google Docs transcribe an audio file?
    No. Google Docs voice typing only listens to your microphone, in Chrome, Edge or Safari: there is no way to upload a file. Playing a recording through your speakers into the mic works badly, as words are lost to room noise. For a file, use Word for the web if you have Microsoft 365, or Whisper for free.
  • How do I transcribe a voice memo on my iPhone?
    On an iPhone 12 or later, open the recording in Voice Memos, tap More, then View Transcript. Choose Copy Transcript to paste the whole text into Notes or an email, and use the search field to find a recording by a word spoken in it. Transcription covers English and nine other languages, and is not available in every country.
  • Can AI transcription tell different speakers apart?
    Yes, most tools label speakers automatically, but they need clean audio to do it well. Word, Otter, Happy Scribe and Rev all separate voices; with Whisper you add a separate diarization step. Overlapping speech is where labels break: on a recorded group call, rename the speakers after a first pass and check the passages where two people talk at once.
  • Is it legal to record and transcribe a conversation?
    It depends on where the people are. In the United States, federal law allows recording when one of the parties consents, while some states, including California, ask for the consent of everyone on the call. In the UK and the EU, transcripts that identify people are personal data under data protection law. The simple rule: say you are recording at the start, and keep the transcript only as long as you need it.
  • How long does it take to transcribe an hour of audio?
    With an AI tool, usually minutes rather than hours: Word warns that a transcript can take up to about the length of the audio file, depending on your internet speed. A human transcript at Rev is delivered in 12 hours or less, and its Rush option, at $1.25 more per audio minute, speeds delivery up by as much as 5 times. Budget the review on top: listening to a one-hour recording at 1.5x takes 40 minutes, before any correction.
  • Is AI transcription safe for confidential recordings?
    Only if you check where the audio goes. Most cloud apps keep your files in your account until you delete them; Microsoft says Word's Transcribe service does not store the audio or the results once the transcript is done, though the audio file itself sits in your OneDrive. Healthcare data in the US needs a provider that signs a HIPAA business associate agreement. For patient, legal or source interviews, Whisper running on your own computer keeps the audio local.
  • How do I transcribe a YouTube video?
    Start with YouTube's own automatic captions, which already are a transcript in many languages, then correct them. Their quality varies with accents, mispronunciations and background noise, so review names and numbers. For a clean document, download the audio of your own video and run it through Word, Whisper or Happy Scribe.