Legal Transcription Software for Mac That Never Uploads Your Audio

Photo by Pixabay on Unsplash

Search for legal transcription software and most of what comes back is not software. Rev, Verbit, Opus 2 and GoTranscript are transcription services. You send them the file, their people and their servers do the work, and you get a transcript back. That is a legitimate way to buy transcription. It is also a very different purchase from a program that runs on your own Mac and never sends the recording anywhere.

This page is about the second kind. It covers what legal transcription software has to do, which recordings a practice actually needs transcribed, where AI transcription is genuinely good enough and where it is not, and what changes when the speech engine runs on your machine instead of a vendor's.

Software or Service: Decide That First

The word "software" in the search results covers two products that behave nothing alike. Sorting this out first saves you from comparing a per minute rate against a subscription and wondering why the numbers do not line up.

A transcription service gives you turnaround tiers, human review options and, at the top end, a certified transcript. Verbit, Rev, Opus 2 and TranscribeMe sit here. You are buying labour and infrastructure. The trade is that the recording leaves your control and the price scales with every minute of audio you produce.

Transcription software gives you an engine. You run it, you get the transcript in minutes, and the cost does not move whether you transcribe one file a month or forty. Dragon, Express Scribe and VoicePrivate sit here. The trade is that nobody else reviews the output. That review pass is yours.

Firms that handle recorded material regularly tend to end up with both: software for the day to day intake calls, memos and internal recordings, and a service or a court reporter for the handful of transcripts that have to be official.

The Recordings a Practice Actually Needs Transcribed

Deposition audio is the example everyone reaches for, and it is usually not the biggest volume. Across the practices that buy transcription tools, the recurring files look more like this:

• Client intake and consultations. The first meeting, where the facts arrive fastest and nobody has time to type. A transcript here saves the reconstruction later.
• Depositions and examinations under oath. Your own working copy, separate from whatever the court reporter produces on their timeline.
• Hearings and motion arguments. What the judge actually said about scheduling, scope or a discovery dispute, before the official transcript arrives.
• Recorded statements. Insurance adjuster calls, witness statements, and recorded interviews in personal injury and defence work.
• Expert and witness prep calls. Long, dense, and full of terminology that you will want to search later.
• Internal case strategy meetings. The material you would least like sitting on somebody else's server.
• Dictated memos and case notes. Which is not transcription at all, it is dictation, and it is the higher volume habit in most practices.

That last one matters for what you buy. If you mostly want to speak and have text land in a document you already have open, that is legal dictation software, a different workflow with different pricing. If you have recorded files sitting in a case folder, that is transcription. VoicePrivate does both, but the buying decision is cleaner when you know which one is your actual bottleneck.

Photo by Rodolfo Quirós on Unsplash

Certified Transcripts and Working Transcripts Are Not the Same Purchase

No AI transcription tool, ours included, produces a certified transcript. This is the single most common mismatch in this category, so it is worth being blunt about.

A certified transcript carries an attestation. A court reporter or a certified transcriptionist signs that the record is accurate and complete, and that signature is what makes it usable as an official record for filing, appeal or impeachment. Verbit and Opus 2 sell that. Rev sells a human verified tier. Court reporters have sold it for a century. Software cannot: there is nobody to attest.

A working transcript is what the rest of the week runs on. It is the searchable text you read on the train, quote in an internal memo, use to find the four minutes of a six hour recording that matter, and hand to a paralegal with a note. It does not need a certification. It needs to exist quickly and stay confidential.

Buying software for the first job leads to disappointment. Buying a certified service for the second job means paying certification rates for text nobody will ever file. The deposition transcription breakdown walks through the cost comparison between ordering official transcription, using AI tools, and dictating your own notes, including when rush transcription is genuinely worth it.

Where the Audio Goes, and Why Rule 1.6 Makes It Your Problem

Upload a recording to a transcription vendor and you have introduced a third party to material your client expected to stay between you.

ABA Model Rule 1.6(a) bars revealing information relating to the representation without informed consent. Rule 1.6(c) adds a duty to make reasonable efforts to prevent inadvertent or unauthorized disclosure. State bar opinions on cloud vendors, in North Carolina, New York, California and Florida among others, converge on the same expectation: clicking accept on a terms of service is not diligence. You are supposed to know where the data goes, who can reach it, and how long it stays.

For transcription specifically, that means reading the vendor's retention terms, their subprocessor list, their human review process and their response policy for government requests. It is real work, and it repeats every time the vendor updates their terms.

Processing the file locally removes the question rather than answering it. There is no vendor in the transcription path, so there is no retention policy, no subprocessor list and no human reviewer to evaluate. Our full write up on Rule 1.6 and voice-to-text goes through the state bar guidance and the vendor questions in detail.

Legal Transcription Software and Services Compared

Named here as market context, from each vendor's own public description of how their product works.

The pattern is easy to miss in a list of nine vendors: almost everything marketed as legal transcription software is a hosted service. The two exits from that are a desktop dictation program and an on-device transcription app. Dragon covers dictation but does not transcribe recorded files, which is why attorneys comparing the two often land on a Dragon alternative for Mac rather than a straight replacement. If you want the ranked shortlist rather than the architecture argument, see best transcription software for lawyers in 2026.

Read This Before You Buy It for Depositions

VoicePrivate labels two speakers, not eight. Speaker labelling covers two speaker dictation, which fits an attorney and client, or an intake call, or a phone interview.

A deposition with counsel, opposing counsel, a witness and a court reporter is four voices, and you will get accurate text without reliable attribution of who said what. For a working transcript you skim alongside your own notes, that is often fine. If your workflow depends on machine speaker attribution across a room, a service with full diarisation is the honest recommendation and we would rather tell you now than process a refund in three weeks.

The same honesty applies to audio quality. Overlapping speakers, heavy accents and a recorder sitting at the far end of a conference table degrade every engine, local or hosted. AI transcription of a poor recording needs correction. That is true of the per minute vendors too, they just charge you for the correction pass.

Turnaround With No Queue in Front of You

Hosted transcription has a turnaround tier because your file is in a line behind other customers' files. That is what rush pricing buys: a better position in the queue.

Local processing has no queue. It has hardware. On Apple Silicon Macs the Neural Engine and Metal GPU do the work, and a long recording finishes in a fraction of its playback length. On Intel Macs and Windows PCs it still runs locally and takes longer. Either way, the file starts processing the moment you drop it in, at two in the morning the night before a hearing exactly as fast as at ten on a Tuesday.

The other half of this is the courthouse problem. After the one time model download on first run, VoicePrivate needs no connection at all. Deposition rooms, courthouse basements and hotel conference centres are where cloud transcription tools fail, and they fail at precisely the moment you were relying on them.

Legal Vocabulary Is Where Generic Engines Fall Down

Voir dire, lis pendens, nunc pro tunc, subrogation, res ipsa loquitur. None of these appear in consumer transcription training data at anything like the density they appear in your recordings, and a generic engine will produce a confident phonetic guess.

The Legal Edition layers 12,000+ legal terms over the base engine, covering torts, statutes, case law terminology and Latin phrases, plus citation handling that transcribes "v.", statute numbers and docket references as citations rather than as noise. Custom vocabulary handles the rest: client names, judge names, local court terminology, opposing firm names, the jurisdiction specific phrases no general model has seen.

Five deposition text shortcuts ship preloaded, so saying "slash depo" expands a full deposition header rather than making you type it. The legal dictation glossary lists the terminology the edition is tuned for, and the Legal Edition feature breakdown covers the processing stack in detail.

Transcribing a Recording, Start to Finish

Drag the file onto the window. No conversion step, no upload progress bar. VoicePrivate accepts MP3, M4A, WAV, FLAC, OGG and AAC, and file import is included on the paid plans.

When the transcript is ready, pick the export that matches where it is going:

• Plain text (.txt) for paralegal review or a paste into your case management system. Clio, MyCase and PracticePanther all accept text notes and file uploads.
• Markdown (.md) for structured case prep notes in Obsidian or Notion.
• JSON (.json) if someone at the firm can write a small import script into a custom field.
• SRT (.srt) and WebVTT (.vtt) for video depositions. Load the caption file into a player, review testimony against synchronised text, then clip the segment you need for a motion.

For video heavy litigation the caption export is the part that pays for itself. Scrubbing a three hour video deposition for one exchange is miserable. Reading the captions and jumping to the timestamp takes a minute.

Photo by Pixabay on Unsplash

Finding Things Again Six Months Later

The cloud platforms sell search hard, and they are right that it matters. A transcript you cannot search is a document you will never reopen.

VoicePrivate keeps a local transcription history with Quick Recall search across everything you have transcribed and dictated. It runs against the local database on your machine, so the index is not a copy of your case material sitting in a vendor's search cluster. Nothing is uploaded to make search work.

What it does not do is cross reference findings across a document production or flag contradictions between witnesses. Those are litigation analytics products and they are a different category with a different price tag. This is search over your own transcripts.

The Audit Log Your Firm's Security Review Will Ask For

Legal Edition writes every transcription into a tamper evident log. Entries are chained with SHA-256 checksums, so each one is linked to the one before it and an edit to a past entry breaks the chain and becomes detectable.

Each entry records timestamp, event type, engine, app version, OS version, source and duration, along with the app's no outbound design assertion for the transcription path. A built in verification check validates the linkage. The whole log exports as JSON in one click.

This is the part that matters when a client's outside counsel guidelines or your own firm's IT review asks what happens to recorded client material. Instead of forwarding a vendor's security page, you can produce a signed chain showing what was transcribed, when, by which local engine. Encrypted backup and restore covers moving that database to a new machine.

To be clear about what this is not: it is an internal record, not a third party certification. VoicePrivate is not making a compliance certification claim here. The log is evidence you can produce, and its value is that it comes from your own device rather than from a vendor's dashboard.

Live Dictation Into Clio, Word and Outlook

Transcription handles the recordings you already have. Dictation handles the writing you have not done yet. VoicePrivate types directly into whatever Mac or Windows app has focus.

Dictate case notes straight into a Clio matter field, draft a motion in Word, answer a client in Outlook, without touching the keyboard and without the text passing through a cloud service on the way. A global hotkey starts it from any app, and silence detection stops it when you pause.

Per-app transcription modes let you set behaviour by application: long form draft settings in Word, short note settings in your practice management system, different custom vocabulary priorities depending on whether the day is litigation or transactional. It reduces the setup friction that makes attorneys abandon dictation tools in week two.

For the workflow side of that, including microphone choice and practice area patterns, see the complete guide to legal dictation software and the ranked field in best dictation software for lawyers in 2026.

Cleaning Up a Transcript With AI Command Mode

Raw transcript text is rarely what you want to read. AI command mode takes a plain language instruction and rewrites the text on your Mac, using the on-device AI, without sending the transcript to a hosted model.

The instructions that earn their keep in practice:

• "Remove all filler words and false starts" before a deposition transcript goes out for review.
• "Summarise this into bullet points by topic" for a case prep digest.
• "Extract every date and deadline mentioned" from a long intake recording.
• "Format this as a Q and A with Q: and A: prefixes" to restructure examination testimony.

This is the same capability the meeting AI tools advertise as summaries and action items. The difference is where the text goes to be rewritten. AI grammar polish is available on macOS with on-device AI.

Client Work That Is Not in English

The engine handles 99 languages, with automatic detection or manual selection.

Immigration practices, community legal services and firms serving multilingual populations take intake recordings, witness statements and client calls in Spanish, Mandarin, Vietnamese, Korean, Tagalog and dozens of others. Producing a working transcript locally first means you know what is in the recording before you decide whether it needs a certified human translation, and you are not paying per minute to find out.

The 12,000+ legal term set is tuned for English legal terminology. For other languages, custom vocabulary is the tool for terms the engine should get right.

What It Costs and What It Does Not Cost

$84 per year, or $9.99 per month. The Team plan is $68 per seat per year for 2 to 5 floating seats. There is no free tier and no free trial. Every plan carries a 14 day money back guarantee, which is how you evaluate it: buy, test it on real recordings, and get a full refund inside two weeks if it does not fit. Details on the Legal Edition pricing page.

The number that moves is not the subscription, it is the per minute alternative. Transcription services price by audio length, human tiers cost more than AI tiers, and rush turnaround costs more again. Rates vary by vendor and tier, so get a current quote rather than trusting a figure on a page like this one. The structural point stands regardless of the rate: a service bills by volume, software does not. A practice that records two intake calls a week is running a meter every week. A practice that records two a year is not, and for them a per minute service is genuinely the cheaper answer.

There is no per minute charge here and no cap on how much you transcribe.

What VoicePrivate Does Not Do

No native plugin for Clio, LexisNexis or Thomson Reuters. Stating that plainly rather than implying a capability that does not exist.

Live dictation types directly into those applications, which covers most of the daily use. But moving a finished transcript into a case management system is a two step process: transcribe, then import the file. That is real friction, and for a firm whose workflow depends on automatic sync it may be the deciding factor. It is the cost of keeping the audio off a vendor's servers, and it is a fair trade to decline.

Also not on offer: certified transcripts, speaker diarisation beyond two speakers, court reporter services, or an official record for filing.

How to Verify Any Vendor's Privacy Claims, Ours Included

Do not take a transcription vendor's word for where your audio goes. Four checks that actually settle it:

VoicePrivate's answers: processing is local, audio is never transmitted or stored externally, no account is required, and the app does not collect usage data. The architecture is described on the privacy page, the feature set on the features page, and common technical questions in the FAQ.

Frequently Asked Questions

What is legal transcription software?

Legal transcription software turns recordings from a law practice into written text, either automatically with a speech engine or by giving a human transcriptionist playback controls. It is distinct from a legal transcription service, where you send the file to a vendor and people or servers you do not control produce the transcript. Software you run yourself keeps the file on your machine. A service does not.

What software do legal transcriptionists use?

Professional legal transcriptionists commonly use Express Scribe as a playback and foot pedal interface alongside manual typing. Firms that outsource use services such as Rev, Verbit, GoTranscript, TranscribeMe or Opus 2. Attorneys doing their own dictation have historically used Dragon Legal. On-device tools such as VoicePrivate are a newer category that runs the speech engine locally instead of sending audio to a vendor.

Is AI legal transcription accurate enough for court use?

For a working transcript that you read, search and cite internally, current AI transcription is usually good enough after a review pass. For an official record filed with a court, it is not a substitute. Certified transcripts come from a court reporter or a certified transcription service that attests to accuracy. Treat AI output as a draft, not as a certified record.

Does legal transcription software work offline?

Cloud services do not, since the upload is the product. VoicePrivate does. After the one time model download on first run, transcription and dictation work with the network off entirely, which matters in courthouses and deposition rooms with unreliable connections.

Does macOS have a built-in transcription feature?

macOS includes Dictation under System Settings, Keyboard, Dictation, which converts live speech to text and runs on-device on Apple Silicon Macs on macOS 13 and later. It does not transcribe pre-recorded audio or video files, does not export SRT, WebVTT, JSON or Markdown, and carries no legal vocabulary. Apple also does not ship a standalone file transcription app for Mac as of 2026; Voice Memos added basic transcription on iPhone in iOS 18, but there is no Mac equivalent. For recorded files you need a dedicated tool.

How much does legal transcription software cost compared with a service?

Transcription services charge per minute or per hour of audio, so cost scales with volume and rush turnaround adds a premium. VoicePrivate is a flat subscription at $84 per year or $9.99 per month, with a Team plan at $68 per seat per year for 2 to 5 floating seats. There is no free tier and no free trial. Every plan carries a 14 day money back guarantee.

Where to Go Next

If you are choosing between dictating and transcribing, start at the legal dictation software hub, which covers the billable hours maths, practice area patterns and equipment. Attorneys evaluating the whole workflow rather than one file type should read the dictation software for lawyers overview.

Coming from a Nuance product, the Dragon Legal Anywhere alternative comparison covers the migration questions specifically. For transcription outside legal work, the Mac transcription software guide compares options across professional categories, and Mac transcription for legal teams is the short version of this page.

VoicePrivate ships in five editions: General, Healthcare, Legal, Finance and Insurance, each with its own vocabulary set. The Legal Edition is on paid plans only, with the 14 day money back guarantee on every one.