Best Meeting Transcription Software 2026: 5 Tested Picks

TL;DR: For on-device transcription with speaker diarization and no minute caps, Everpaper tops our list (full disclosure: Everpaper is our product). Otter.ai is the mature cloud standard with the broadest platform coverage. Fireflies.ai has the most generous free tier. Granola produces polished notes without a bot in the call. ScreenPipe is the source-available choice for developers who want raw local capture.

How we evaluated

We scored each tool on the things that actually decide whether a transcript is usable:

• Transcription accuracy mechanics. Which engine runs, where it runs, and whether the model is named or a black box. On-device engines (Parakeet, Whisper, MOSS) behave very differently from cloud ASR.
• Speaker identification. Diarization quality: does the tool label "Speaker 1 / Speaker 2," real names, or just "Me / Them"?
• Minute caps and storage limits. What stops working when you hit the ceiling.
• Export formats. Can you get txt, markdown, SRT, VTT, or JSON out, or are you stuck copy-pasting from a web app?
• Offline behavior. Does anything work without a network connection?

Pricing below was checked 2026-08-04 against each vendor's public pricing page.

1. Everpaper: best for private, on-device transcription

Everpaper is our product, so read this section knowing that. It's a private, on-device daybook for macOS, and its meeting feature, The Minutes, records and transcribes entirely on your Mac.

What we like. Three named speech engines, all local. The default is Parakeet (parakeet-tdt-0.6b-v3), which runs on the Apple Neural Engine, so transcription doesn't fight your meeting app for CPU. Whisper is available as an alternative, and MOSS (MOSS-Transcribe-Diarize, Apache 2.0) does transcription and speaker diarization in a single model, so there's no separate diarization pass, no cloud round-trip. MOSS is a ~2 GB download, happiest on 16 GB machines, and the docs are honest about its edges: word-level timestamps within a sentence are approximate, and a voice that goes quiet for a long stretch can re-enter under a new speaker label.

Recording captures microphone and system audio as separate time-aligned tracks, which is what makes speaker labeling meaningful. The live panel shows a speaker-labeled transcript as the meeting runs, with level meters and Mute/Pause/Stop. Hide my voice and Hide any speaker (⌘4) redact speakers from the transcript, and the setting persists.

Everpaper has no minute caps. Transcription never touches the network, whether the recording is a 4-hour workshop or a 15-minute standup.

Export is where the everpaper-cli shines. Ships inside the app bundle at /Applications/Everpaper.app/Contents/MacOS/everpaper-cli, shares the app database, and exports meetings to txt (default), markdown, srt, vtt, and json. The transcribe group transcribes any audio or video file on-device, with --diarize (threshold 0.6, up to 6 speakers), --language, and --processing raw|clean flags. SRT and VTT exports mean you can caption video directly from a meeting recording. Import Transcript (1.11.0) goes the other way: bring a .vtt from Zoom, Teams, or Otter in as a real meeting, no audio needed. A Lexicon of custom words (1.15.x) fixes recurring name and jargon errors at the source.

Drawbacks. macOS 14.2+ only, and Apple Silicon required, since the Neural Engine speech model means Intel Macs are out. No Windows, Linux, iOS, Android, or web version. It's in private beta with limited spots. AI features like the live Ask tab and auto-summaries wait until you configure an AI provider; transcription itself never needs one. Auto-summary ("Run marked prompts automatically") is off by default precisely because it sends the transcript to your provider.

Pricing. In private beta with limited spots, with a founding-license option for early supporters, not a monthly subscription. No public price. Join the waitlist at everpaper.app.

Best for: managers and ops leads on Apple Silicon Macs who want private, speaker-labeled transcripts, subtitle-grade export formats, and no monthly minute math. See the meeting docs and CLI docs.

2. Otter.ai: best for teams already living in Zoom/Meet/Teams

Otter is the incumbent. Founded in 2016, it claims 35M+ users as of December 2025, and it shows: the product is polished, the integrations are deep, and everyone on your team has probably seen the bot join a call.

What we like. The capture model is a cloud bot: OtterPilot joins Zoom, Google Meet, or Microsoft Teams as a visible participant via calendar sync, records, transcribes in real time, captures slides, and emails a summary after. There's also a desktop app (Mac and Windows), iOS and Android apps, and a Chrome extension, so coverage is the widest on this list. Speaker identification works on every tier, with taggable speakers on Pro and up. Cross-meeting AI Chat lets you query your entire meeting archive, and Pro adds Salesforce, HubSpot, and Zapier integrations. That combination suits sales and customer-success teams that want transcripts flowing into a CRM.

Drawbacks. Everything runs in Otter's cloud (AWS, US-based). There is no offline mode at all; no network, no transcript. The June 2026 privacy policy states Otter may train "proprietary AI technology on de-identified audio recordings and on transcriptions," and it discloses data-labeling providers. That's the core trade-off. There's also an active consolidated federal class action, In re Otter.AI Privacy Litigation (N.D. Cal.), alleging recording without all-party consent; a motion-to-dismiss hearing was held May 20, 2026. Accuracy complaints persist in reviews: speaker misattribution with generic "Speaker 1/2" labels that need manual correction, and hit-or-miss results even in clean audio. And the minute caps bite: Capterra reviewers mention the stress of tracking usage against monthly limits. Deleted conversations still count toward your monthly cap.

Pricing (checked 2026-08-04). Basic: free, 300 transcription minutes/month, 30 minutes per conversation, 3 lifetime file imports. Pro: $16.99/user/month monthly or $8.49/user/month billed annually, 1,200 minutes/month, 90 minutes per meeting, 10 imports. Business: $30/user/month monthly or $24/user/month billed annually, up to 4 hours per meeting, unlimited imports. Enterprise: custom.

Best for: organizations that need a bot in the call, CRM plumbing, and cross-meeting search, and that can accept cloud processing and its consent implications. We compare it directly in Otter.ai vs Everpaper and list more options in best Otter.ai alternatives.

3. Fireflies.ai: best free tier and multilingual teams

Fireflies (founded 2016, by Krish Ramineni and Sam Udotong) is the other big cloud bot, and its pitch is breadth: 100+ transcription languages and an unusually usable free plan.

What we like. The free tier offers unlimited transcription and unlimited AI summaries (rate-limited) with 400 minutes of storage per team, the most generous no-cost option here for a team that wants to try a bot notetaker. The bot ("Fred") joins Zoom, Meet, Teams, and Webex, and there are desktop, iOS, Android, and Chrome extension paths plus file upload. Speaker identification/diarization is built in. Security posture is genuinely strong for a cloud tool: SOC 2 Type II, GDPR, HIPAA with a BAA on Enterprise, and a public commitment ("We don't train on it by default unlike other AI companies") plus a stated 0-day data retention policy with vendors. API access exists on every tier, including a documented GraphQL API, which ops teams building meeting-data pipelines will appreciate.

Drawbacks. The auto-join behavior is the top complaint across reviews: the bot appears uninvited in sensitive calls via calendar sync and is reportedly hard to fully disable. Accuracy degrades with heavy accents, crosstalk, and specialized vocabulary. The AI-credits system (20/20/30/50 credits by tier) confuses people; Trustpilot reviews mention unexpected charges and hard-to-track usage. Storage is the real cap: 400 minutes per team on Free, 8,000 per seat on Pro. There's a December 2025 BIPA class action (Cruz v. Fireflies.AI Corp., C.D. Ill.) alleging voiceprint collection without consent, worth knowing if your org has Illinois participants. Nothing works offline.

Pricing (checked 2026-08-04). Free: $0, unlimited transcription, 400 min storage/team, 20 AI credits. Pro: $10/seat/month annual ($18 monthly), 8,000 min storage/seat. Business: $19/seat/month annual ($29 monthly), unlimited storage, conversation intelligence. Enterprise: $39/seat/month, annual only, SSO/SCIM, HIPAA BAA, private storage.

Best for: multilingual teams, orgs that want HIPAA on a notetaker, and anyone who needs a real free tier to pilot with. Also covered in best AI meeting assistants 2026.

4. Granola: best note quality without a bot

Granola takes a different approach: no bot joins your call. The desktop app captures your microphone and system audio locally, streams it to cloud transcription providers, and merges a transcript with whatever rough notes you typed. It became a favorite among founders and VCs for a reason.

What we like. Note quality is the headline. Nat Friedman called it "the best meeting notes of anything I've tried so far," and long-term users report it replaced their entire note-taking workflow. Because it listens to system audio, it works with any meeting platform on desktop (Zoom, Meet, Teams, anything) plus in-person meetings via the mobile apps. No participant sees a bot, no awkward announcements. The human-in-the-loop model (you jot, AI fleshes out) gives more control over the final artifact than fully automatic summaries. Audio is not stored at all: Granola transcribes in real time and keeps only the transcript and your notes, which is a meaningfully smaller data footprint than bot tools that archive recordings.

Drawbacks. Transcription is cloud (Deepgram and AssemblyAI are named on the security page, with OpenAI and Anthropic for summarization), so nothing works offline. Speaker identification is the weakest on this list: desktop transcripts default to "Me" (mic) and "Them" (system audio), and named speaker tags exist only for Google Meet (browser extension) and Zoom (macOS). If who-said-what matters, this hurts. No audio playback means you can't verify the transcript against the recording. Granola trains its own models on anonymized user data by default on consumer tiers (opt-out in Settings; off by default on Enterprise). Export options are thin; a meetingnotes.com teardown scored it 66/100 citing exactly that. And you can't import pre-recorded audio files.

Pricing (checked 2026-08-04). Basic: free, AI meeting notes with limited meeting history (exact cap not stated on the pricing page). Business: $14/user/month, unlimited notes and history, Attio/Notion/Slack/HubSpot/Affinity/Zapier integrations, MCP, API. Enterprise: $35/user/month, SSO, org-wide auto-deletion, model training off for the team.

Best for: individuals and small teams who care about the quality of the written notes more than the raw transcript, and who dislike bot-visible tools. Head-to-head in Granola vs Everpaper.

5. ScreenPipe: best for developers building their own pipeline

ScreenPipe is the outlier: not a meeting tool at all, but a 24/7 local recorder for screen and mic with a local search API and an agent framework ("pipes") on top. Written in Rust, backed by Y Combinator, and positioned as a workflow-memory layer for AI agents.

What we like. Genuinely local-first. Data lands in ~/.screenpipe/db.sqlite plus a data/ folder on your machine, with a local API at localhost:3030. Capture, OCR, local Whisper transcription (whisper-tiny through whisper-large-v3-turbo, ~100 languages), and search all work fully offline; network is needed only for cloud transcription (Deepgram, screenpipe-cloud), cloud AI providers, team sync, and license checks. No vendor minute caps in local mode; your disk is the limit. It records whatever the machine hears and displays, so it's meeting-platform agnostic with no bot. The MCP server hooks into Claude Desktop, Codex, Cursor, and VS Code, and pipes let you schedule agents (for example: a meeting-ended trigger that writes a summary into a Markdown note) that can run on local models via Ollama. Default-on PII removal redacts emails, phone numbers, and card numbers.

Drawbacks. This is a toolkit, not a meeting product. The official documentation reviewed does not state whether speaker identification or diarization is supported. There is no finished meeting-centric workflow; you assemble one with pipes or code. The official troubleshooting docs tell users to lower capture FPS and use whisper-tiny on older machines, which tells you CPU and battery are real concerns for always-on capture. In June 2026 the source license moved from MIT to a source-available commercial license, requiring paid licenses for commercial use of the source; that source-license question remains separate from the clearly priced Free, Basic, and Business product plans.

Pricing (checked 2026-08-04, supplied ScreenPipe pricing page). Free is $0 with a 7-day Business trial, local capture, AI search, meeting notes, 50+ integrations, and 10 AI credits/month. Basic is $25/month or $250/year, with 150 credits and unlimited user-installed scheduled tasks and history. Business is $50/seat/month or $500/seat/year, with 400 credits, frontier Claude and GPT models, cloud sync, automations, provider-key support, and team workspace.

Best for: developers and technical ops people who want raw local capture plus an API, and are happy to build the meeting layer themselves. Full breakdown in ScreenPipe vs Everpaper.

Honorable mentions

Honorable mentions: Fathom, Read.ai, Notta, and Krisp.

How to choose

Work through four questions:

1. Where may the audio go? If client confidentiality, legal privilege, or company policy rules out cloud processing, your shortlist is Everpaper and ScreenPipe, full stop. Everyone else on this list ships audio or transcripts to a server.
2. How important is speaker attribution? If you need to know who committed to what, prioritize real diarization: Everpaper (MOSS or --diarize), Otter, or Fireflies. Granola's "Me/Them" labels won't survive a dispute about who approved the budget.
3. What's your meeting volume? Add up monthly minutes. A manager with 25 hours of meetings a month blows through Otter's Pro cap (1,200 minutes) and needs Business or an uncapped local tool. Fireflies caps storage, not transcription, which suits high-volume teams differently.
4. What happens to the transcript next? If transcripts feed subtitles, a wiki, or an automation pipeline, check export formats first. SRT/VTT/JSON export (Everpaper's CLI) or a real API (Fireflies, ScreenPipe) saves you from copy-paste workflows.

FAQ

Is there free meeting transcription software?

Yes, with strings attached. Fireflies.ai's free tier offers unlimited transcription with 400 minutes of storage per team (checked 2026-08-04), the most usable free plan here. Otter.ai's free Basic plan gives 300 transcription minutes per month with a 30-minute per-conversation cap. Granola's free tier has limited meeting history. ScreenPipe's CLI is free for personal use and fully local. Everpaper is in private beta, not a free product. In every cloud case, the free tier is subsidized by limits, upsell pressure, or data terms. Read the privacy policy before you point a free bot at sensitive calls.

Can meeting transcription work offline?

Yes, but only with on-device tools. Everpaper transcribes fully offline on Apple Silicon using Parakeet, Whisper, or MOSS. Recording, transcription, speaker diarization, search, and export never touch the network. ScreenPipe also works offline for capture and local Whisper transcription. Otter, Fireflies, and Granola all require a connection: their transcription runs in the vendor's cloud, so no network means no transcript.

Which meeting transcription software is most accurate?

There is no universal winner; audio quality, accents, crosstalk, and specialist vocabulary all matter. Otter and Fireflies reviewers report errors in difficult audio. Everpaper offers three named local engines and a Lexicon for recurring names and jargon. Granola names Deepgram and AssemblyAI as its transcription providers. For multilingual meetings, Fireflies explicitly advertises 100-plus languages.

Do these tools identify speakers?

Everpaper, Otter, Fireflies, and Granola document speaker labeling, with different approaches. Everpaper's MOSS model does transcription and diarization in one pass on-device, and its CLI offers --diarize with up to 6 speakers. Otter and Fireflies label speakers in the cloud, with taggable speakers on Otter's paid tiers. Granola defaults to "Me" and "Them" and only names speakers on Google Meet and Zoom via extensions. The official ScreenPipe documentation reviewed does not state whether speaker identification or diarization is supported.

Is it legal for a bot to record my meeting?

It depends on jurisdiction. All-party-consent states and countries require everyone on the call to agree to recording, and two of the tools on this list face active litigation on exactly this point: Otter (a consolidated federal class action alleging recording without all-party consent) and Fireflies (a BIPA suit alleging voiceprint collection without consent). Bot-free local tools sidestep part of the issue because no third party joins or stores the call, but consent law still applies to you as the recorder. When in doubt, announce the recording.

Can I transcribe an audio or video file I already have?

Everpaper's CLI transcribes any audio or video file on-device (transcribe group, with --diarize and --language flags), and its Import Transcript feature ingests .vtt files from Zoom, Teams, or Otter as meetings. Otter allows file imports: 3 lifetime on the free plan, 10/month on Pro, unlimited on Business. Fireflies supports audio/video upload on all tiers. Granola cannot import pre-recorded files.

Methodology

We evaluated tools in August 2026 against five criteria: transcription engine mechanics, speaker identification, minute and storage caps, export formats, and offline behavior. Everpaper claims come from its shipped documentation (we built it). Competitor claims come from each vendor's public pricing, security, and help pages as linked in our research notes, plus named third-party reviews and court records where relevant; every published price carries a "checked 2026-08-04" stamp. Rankings weigh privacy architecture and exportability heavily because our audience is managers and ops leads handling confidential conversations; a team that prioritizes CRM integrations above privacy may rank the cloud tools differently.

---

Everpaper is in private beta with limited spots. On-device meeting transcription with Parakeet, Whisper, and MOSS, speaker-labeled live transcripts, and CLI export to txt, markdown, SRT, VTT, and JSON. No account, no minute caps, no audio leaving your Mac. Join the waitlist at everpaper.app, and read the meeting docs and CLI docs.