Local AI transcription for Windows

Sound in.
Notes out.

SoundWiz turns recordings, meetings and interviews into accurate, timestamped transcripts — entirely on your own PC. No cloud. No uploads. No subscriptions leaking your audio.

Download for Windows — Free See how it works

v1.5.0·Windows 10 / 11, 64-bit·CUDA & Vulkan GPU acceleration·Whisper models by OpenAI, via whisper.cpp

100%
local processing — audio never uploaded
99
languages, auto-detected
8
Whisper models, tiny to large-v3
6
export formats, no lock-in
Features

Everything you need to go from audio to notes

Professional transcription workflow, built as a native Windows app — fast, focused and private.

Private by design

Transcription runs entirely on your machine with whisper.cpp. No accounts, no uploads, no one listening in — your recordings simply never leave the building.

Speaker labels

Interviews and meetings are automatically split by voice, with every line tagged as Speaker 1, Speaker 2 and so on — readable dialogue, not a wall of text.

99 languages, auto-detected

Set the language or leave it on Auto — SoundWiz detects what's being spoken and transcribes accordingly. English, Spanish, Mandarin, German and 95 more.

GPU acceleration

NVIDIA CUDA and Vulkan backends are detected automatically, with tuned CPU builds as fallback. The right model for your hardware is recommended on first run.

Any audio or video

MP3, WAV, M4A, FLAC, OGG, MP4, MKV, MOV and more — built-in FFmpeg decoding handles whatever you drop in, and long recordings are processed in smart chunks.

Clean exports

Save as TXT, timestamped TXT, Markdown, SRT subtitles, DOCX documents or audit JSON with file hashes. Open standards, no lock-in, ready for your editor.

How it works

Three steps. No accounts, no queues.

From raw recording to finished notes in under a minute for most files.

1

Drop in your audio

Drag any recording onto the window — a meeting, lecture, interview or voice memo. SoundWiz reads its length and recommends a model for your hardware.

2

Choose model & language

Pick from eight Whisper models, or accept the recommended one. Leave language on Auto or pin it to one of 99 supported languages.

3

Export your notes

Review the timestamped, speaker-labeled transcript and export it as TXT, Markdown, SRT, DOCX or JSON — then get back to work.

Whisper models

Pick the model that fits the job

Every official OpenAI Whisper model is one click away, downloaded straight from Hugging Face. Bigger models are more accurate; smaller ones are faster.

ModelSizeBest forEdition
tiny78 MBFast preview passesFree
tiny.en78 MBEnglish-only previewFree
base148 MBFree default — good speed and accuracyFree
base.en148 MBEnglish-only defaultFree
small488 MBShort recordingsPro
medium1.5 GBRecommended for interviewsPro
large-v3-turbo1.6 GBBest speed / quality balancePro
large-v33.1 GBHighest quality, heavyPro

Models are stored locally and reused across jobs. After the initial download, SoundWiz works fully offline.

Privacy

Your audio never leaves your machine.

SoundWiz is built for people who work with sensitive recordings: journalists protecting sources, lawyers and clinicians bound by confidentiality, researchers under NDA, and businesses that simply refuse to paste meeting audio into a browser tab.

Transcription runs on-device with whisper.cpp. There is no cloud component, no account system and no upload step — the only network access is the one-time model download you choose and trigger.

  • All processing happens locally, on your CPU or GPU
  • No accounts, no sign-up, no tracking of your audio
  • Works fully offline once a model is installed
  • Projects and transcripts stored on your own disk
soundwiz — session log
──────────────────────────────
› load model local large-v3-turbo
› decode audio local ffmpeg 42.1s
› transcribe local cuda · 28 chunks
› diarize speakers local 2 speakers
› export local srt + docx
──────────────────────────────
audio bytes uploaded: 0
Pricing

Start free. Upgrade when your recordings get longer.

The free edition is a real tool, not a trial — many users never need anything else.

Free

$0
For voice memos, lectures and quick notes.
  • Files up to 20 minutes
  • tiny & base models
  • 99 languages, auto-detect
  • All export formats
  • 100% local processing
Download free
Pro

Pro

$29.90 one-time
For interviews, research and full-day recordings.
  • Files of any length
  • All 8 models, including large-v3-turbo and large-v3
  • Hardware-aware model recommendations
  • Audit JSON with SHA-256 file hashes
  • Everything in Free, still fully local
Get Pro

Secure checkout by Creem, our Merchant of Record — cards, PayPal, Apple Pay & Google Pay. Exact price incl. taxes is shown before payment. Covered by our 14-day money-back guarantee.

Sales, invoicing, taxes and refunds for Pro licenses are handled by Creem (creem.io) as Merchant of Record. License keys are delivered by email and activated inside the app — see our Terms, Privacy Policy and Refund Policy.

FAQ

Frequently asked questions

Is my audio uploaded anywhere?
No. Transcription runs entirely on your PC using whisper.cpp. The only network traffic SoundWiz ever makes is downloading the Whisper model file you explicitly choose, from Hugging Face. After that, you can disconnect from the internet and keep transcribing.
Which file formats are supported?
Anything FFmpeg can decode: MP3, WAV, M4A, AAC, FLAC, OGG, OPUS, WMA, MP4, MKV, MOV, AVI and more. Video files are handled the same way — SoundWiz extracts the audio track automatically.
Do I need an NVIDIA GPU?
No. If you have an NVIDIA GPU, SoundWiz uses CUDA for the fastest results. Vulkan is supported for other GPU vendors, and optimized CPU builds are used otherwise. On first run the app recommends the largest model your hardware can handle comfortably.
What are the limits of the free edition?
The free edition transcribes files up to 20 minutes long and includes the tiny and base models. Pro removes the duration limit and unlocks all eight models, including large-v3-turbo and large-v3 for the best accuracy.
How accurate are the transcripts?
With a modern model like large-v3-turbo, accuracy on clear audio is excellent — but any AI transcription can make mistakes: hallucinated words, mixed languages or drifting timestamps. SoundWiz marks every export as a draft. For anything important, a human review pass is expected, and the timestamped formats are designed to make that fast.
How do I buy a Pro license, and how is payment handled?
Click "Get Pro" — you'll be taken to a secure hosted checkout run by Creem (creem.io), our Merchant of Record. Creem handles payment (cards, PayPal, Apple Pay, Google Pay), invoicing, and VAT/GST/sales taxes worldwide. Your license key is delivered by email immediately after purchase; enter it in the app under Settings → License to unlock Pro. We never see your payment card details.
What if Pro isn't right for me?
Every Pro purchase is covered by a 14-day money-back guarantee. Email support with your order number or use the link in your Creem receipt — refunds are issued to the original payment method within 5–10 business days, and your license reverts to the Free edition.
Can I use the transcripts in my editor or player?
Yes. Export as Markdown or DOCX for documents, plain TXT for notes, SRT for subtitles in video editors and players, or audit JSON (with SHA-256 hashes of the source audio) when provenance matters.

Ready when your audio is.

Download SoundWiz free and turn your first recording into notes in the next five minutes. No account required.

Download SoundWiz for Windows

v1.5.0 · Windows 10 / 11, 64-bit · ~25 MB installer · models downloaded on demand