Voz
Transcribe 10 minutes of audio in 2s on an iPhone, 14x faster than Whisper, accurate word timestamps.
Transcribe 10 minutes in 2 seconds.
Transcribe an audio or video file with Voz and get accurate start and end times for every word. 10 minutes takes 2s on an iPhone, on the device, so nothing is uploaded and nothing is billed per minute.
On Apple silicon, the model runs on the Neural Engine, so transcribing is lightning fast and consumes less power. On a Mac a back catalog of podcasts, interviews, and lectures runs through in one pass, and nothing is uploaded.
The timestamps are precise enough to edit on. Word starts land within 83ms of a forced aligner and word ends within 95ms, for precise transcript-based editing. Voz is an Apple Neural Engine optimized version of NVIDIA's Parakeet TDT 0.6B v3, with accuracy comparable to Whisper large-v3-turbo on most audio, with a download size of only 467MB.
14x faster than Whisper.
10 minutes of narration in 2s on an iPhone 17 Pro, 1.2s on an iPhone 18 Pro and 2.1s on an iPhone 15 Pro. On an M3 Ultra, 14x faster than whisper.cpp large-v3-turbo on the same 10-minute podcast, with a 7.40% word error rate over six public English sets where Whisper scores 7.00%.
Time to transcribe 10 minutes of audio.
| Device | Time | Realtime |
|---|---|---|
| M3 Ultra | 0.8s | 762x |
| iPhone 18 Pro | 1.2s | 492x |
| M4 Max | 1.3s | 453x |
| iPhone 17 Pro | 1.8s | 334x |
| iPhone 15 Pro | 2.1s | 283x |
| M1 Mac mini | 2.4s | 251x |
Word error rate on the Open ASR Leaderboard datasets, Voz with its own text normalizer, Whisper's figures from the leaderboard
| Dataset | Voz | Whisper large-v3-turbo |
|---|---|---|
| LibriSpeech test-clean | 2.19% | 2.13% |
| LibriSpeech test-other | 3.86% | 3.70% |
| GigaSpeech | 9.70% | 8.47% |
| SPGISpeech | 3.86% | 2.79% |
| Earnings-22 | 12.97% | 11.07% |
| AMI | 11.84% | 13.87% |
| Average | 7.40% | 7.00% |
Read the row that matches your audio. Clean, close-miked speech lands around 2-4% (LibriSpeech, SPGISpeech), podcasts and web video around 10% (GigaSpeech), and meeting rooms and phone calls 12-13% (AMI, Earnings-22), which is where the model beats Whisper. Per-language figures on 10 minutes of read speech each are on the model card, from 3.31% for Italian to 39.46% for Greek. The iPhone timings and the whisper.cpp comparison are our own runs and are not on the card yet.
Use cases
Transcribe 10 minutes in 2 seconds.
Transcribe any audio or video file
Record a podcast or a video and the full transcript is ready before the export finishes. Searchable text with a timestamp on every word, in any of 25 languages, without leaving the device.
Churn through a backlog
An archive of interviews, lectures, or meeting recordings becomes searchable in one pass on a Mac. A hundred hours runs through in 20 minutes, and nothing is uploaded.
Cut on a word
Every word carries a start and an end, so a selected range in the transcript is a cut in a video editor. Pair it with Clips for the shorts and Title to name them.
Inspiration
Ideas to build with Voz. Copy a prompt into your coding agent and go.
Build a podcast player that transcribes every episode and makes it searchable, on the device.
Build a meeting recorder that hands you a timestamped transcript before you leave the room.
Turn voice memos into searchable, editable text notes, offline.
Generate SRT subtitles for any video file on the Mac.
Add captions to a video player that run on the device.
Build an interview app for journalists where sources never leave the phone.
Build a full podcast studio: transcribe, clip, and title on the device.
Record, clean, transcribe, and clip: a whole creator pipeline, offline.
What the model does
- Word starts within 83ms and word ends within 95ms of a forced aligner, on average, at 80ms frame resolution.
- 10 minutes of audio in 2s on an iPhone 17 Pro, 1.2s on an iPhone 18 Pro, and 2.1s on an iPhone 15 Pro. On an M3 Ultra, 0.8s at 762x realtime. Single short clips run 50-62x, because every clip pays for a full 15s window.
- 7.40% word error rate averaged over six Open ASR Leaderboard sets (Whisper large-v3-turbo: 7.00%), 2.83% on half an hour of narration (Whisper: 2.72%), 11.84% on AMI meetings (Whisper: 13.87%).
- The whole graph runs on the Neural Engine with no CPU or GPU fallback; peak memory doesn't grow with the length of the recording.
- 25 languages: Bulgarian, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, German, Greek, Hungarian, Italian, Latvian, Lithuanian, Maltese, Polish, Portuguese, Romanian, Russian, Slovak, Slovenian, Spanish, Swedish, and Ukrainian.
- 467MB on disk, downloaded on demand and cached; the first load after a download takes 20s once while Core ML specializes, then 0.2s. Download during onboarding.
- Mono audio at any sample rate; the SDK resamples and downmixes. Longer audio is cut into 15s windows at pauses and joined on the words neighboring windows agree on.
- Built on NVIDIA's Parakeet TDT 0.6B v3, released under CC BY 4.0. The weights are unchanged; the conversion, compression, and the Neural Engine runtime are ours.
Example app
Clipper
Built with Voz, Clips, Title
Generate short clips from a video podcast or a long recording, fully on device. A macOS app and a command-line tool over the same core.
Voz transcribes with a time on every word, Clips ranks the best spans, and Title writes a title and description for each.
macOS 26 or later, Apple Silicon
brew tap desert-ant-labs/tap
brew install --cask clipperGetting started
Add speech recognition to your iOS or macOS or web app in a few lines of code. Voz docs.
// Swift Package Manager
.package(url: "https://github.com/Desert-Ant-Labs/desert-ant-core.git", from: "3.5.0")
// target dependency
.product(name: "Voz", package: "desert-ant-core")import Voz
let voz = try await Voz()
let result = try await voz.transcribe(url)
result.text // the transcript
result.words.first?.start // 80ms resolution
result.realtimeFactor // seconds of audio per second of wall clockAdd Voz from Desert Ant Labs to this Swift project (iOS, macOS). What it does: On-device Speech to Text on the Neural Engine. SDK: Swift (iOS, macOS) Repo: https://github.com/Desert-Ant-Labs/desert-ant-core#readme // Swift Package Manager .package(url: "https://github.com/Desert-Ant-Labs/desert-ant-core.git", from: "3.5.0") // target dependency .product(name: "Voz", package: "desert-ant-core") Reference: - Model page: https://desertant.com/models/voz/ - Full catalog and other models: https://desertant.com/llms.txt Add the SDK, then follow its README for the exact API and current version. Do not invent API names or method signatures; confirm them against the README.
npm i @desert-ant-labs/voz onnxruntime-webimport { Voz } from "@desert-ant-labs/voz";
const voz = await Voz.load();
const result = await voz.transcribe(file); // File, Blob, ArrayBuffer, or samples
result.text;
result.words[0]; // { text, start, end }
result.realtimeFactor;Add Voz from Desert Ant Labs to this JavaScript / TypeScript project (Web, Node.js). What it does: On-device Speech to Text on the Neural Engine. SDK: JavaScript / TypeScript (Web, Node.js) Repo: https://github.com/Desert-Ant-Labs/desert-ant-core#readme npm i @desert-ant-labs/voz onnxruntime-web Reference: - Model page: https://desertant.com/models/voz/ - Full catalog and other models: https://desertant.com/llms.txt Add the SDK, then follow its README for the exact API and current version. Do not invent API names or method signatures; confirm them against the README.
Specs
- Speed
- 10 minutes of audio in 0.8s to 2.4s across Mac and iPhone, 251x to 762x realtime; per-device times in Performance.
- Accuracy
- 7.40% WER over six Open ASR Leaderboard sets; 2.83% long-form; word starts 83ms, ends 95ms mean error
- On-device size
- 467MB compiled Core ML, downloaded on demand
- Languages
- 25 European languages; accuracy from 3.31% (Italian) to 39.46% (Greek) on read speech
- Model
- NVIDIA Parakeet TDT 0.6B v3 (CC BY 4.0), converted to Core ML and compressed by Desert Ant Labs: log-mel frontend, conformer encoder, transducer decoder, all on the Neural Engine
- Platforms
- iOS, iPadOS, macOS, tvOS, visionOS (Core ML); browser and Node (WebAssembly, ONNX Runtime)
The Neural Engine speed is Apple only: on iOS, macOS, tvOS, and visionOS the runtime drives Core ML to keep the whole graph on the Neural Engine. The browser and Node run the same model on WebAssembly and ONNX Runtime, slower than Core ML, and there is no native Android build.
Voz doesn't know which of its 25 languages it is hearing, and a language it doesn't cover produces confident nonsense rather than an error, so pair it with Ear. Accuracy varies widely by language, word ends are the harder half of the timestamps, and 467MB is a real download to fetch during onboarding.
FAQ
What is Voz?
Transcribe 10 minutes of audio in 2s on an iPhone, 14x faster than Whisper, accurate word timestamps.
Does Voz run on device?
Yes. Voz runs on the device, with no server call, so the data stays with the user.
Which platforms does Voz support?
Voz ships as a native on-device SDK for Swift, JavaScript / TypeScript.
How much does Voz cost?
Each model is free up to 100k monthly active devices. Inference is unlimited. Contact us for custom licenses.
How accurate or fast is Voz?
10 minutes of narration in 2s on an iPhone 17 Pro, 1.2s on an iPhone 18 Pro and 2.1s on an iPhone 15 Pro. On an M3 Ultra, 14x faster than whisper.cpp large-v3-turbo on the same 10-minute podcast, with a 7.40% word error rate over six public English sets where Whisper scores 7.00%.
Does Voz run outside Apple platforms?
Yes, in the browser and in Node. The JavaScript package runs the same model on WebAssembly and ONNX Runtime, so a web app or a server transcribes without Apple hardware, on Windows, Linux, or an Android browser included. On Apple silicon the whole model runs on the Neural Engine, which is where it is fastest. A native Android build is not ready yet.