What hardware and devices do you support?

Operating System Hardware requirements macOS v14 Sonoma or later, Tahoe v26 recommended M1 or newer Windows 11 Intel/AMD/ARM (with GPU 2GB) Other Hardware Get in touch if you have different requirements

On-device speech-to-text for laptops.

Enterprise-grade Speech-to-Text running locally on Mac, Windows and other devices. Sub-second latency, no infrastructure costs and audio that never leaves the machine. Full speaker diarization and identification included.

✅ Runs on laptop hardware

✅ 55+ languages

✅ CoreML & DirectML optimized

[alt: Laptop with speech-to-text feature showing speaker labels and audio waveforms on a dark screen with green accents.]

Trusted by millions of users globally

Case study

Prosodica | Case study

Driving better conversations at scale

Leveraging speech recognition to track customer interactions, highlight key insights, and raise contact center performance

Case study

NCI | Case study

Redefining real-time captioning

Precise, low-latency transcription and translation, all delivered before your media even ends.

Case study

AI Media | Case study

Delivering 120X more with voice AI

Powering live content through AI-powered transcription, built on industry-leading voice AI

Case study

Adobe | Case study

Cloud-grade speech recognition on-device for Adobe Premiere

Run the most accurate on-device transcription locally; efficient enough for a laptop, powerful enough for professional work.

Case study

Prosodica | Case study

Driving better conversations at scale

Leveraging speech recognition to track customer interactions, highlight key insights, and raise contact center performance

Case study

NCI | Case study

Redefining real-time captioning

Precise, low-latency transcription and translation, all delivered before your media even ends.

Case study

AI Media | Case study

Delivering 120X more with voice AI

Powering live content through AI-powered transcription, built on industry-leading voice AI

Case study

Adobe | Case study

Cloud-grade speech recognition on-device for Adobe Premiere

Run the most accurate on-device transcription locally; efficient enough for a laptop, powerful enough for professional work.

Deploy enterprise STT on every laptop

Adobe and other ISVs build privacy-first, laptop-native applications with Speechmatics On-Device.

Tell us about your deployment and we'll follow up directly.

Why run offline STT on the laptop?

Reliable transcription regardless of network conditions, with data that stays private and features that match the cloud.

Shift compute to the edge

Offload transcription from cloud to the device
Predictable licensing model
Scale without scaling your server fleet

[alt: A "No Wi-Fi" indicator with a crossed-out Wi-Fi symbol in dark gray, signifying no internet connection available.]

No network required

Works when Wi-Fi drops
Critical for clinical and legal environments
Sub-second latency, no network dependency

[alt: Two-tier pink cake decorated with intricate white petal designs, topped with large white peony flowers on a white stand.]

Audio stays on-device

GDPR, HIPAA, and air-gapped ready
Compliance by architecture
Sell into regulated & government accounts

[alt: A vibrant, colorful pie chart with various segments in blue, green, yellow, and red, representing different data categories.]

Cloud-grade features

12-16% fewer errors than Whisper-based alternatives
Full speaker diarization and ID
55+ languages, real-time and batch

Shift compute to the edge

Offload transcription from cloud to the device
Predictable licensing model
Scale without scaling your server fleet

No network required

Works when Wi-Fi drops
Critical for clinical and legal environments
Sub-second latency, no network dependency

Audio stays on-device

GDPR, HIPAA, and air-gapped ready
Compliance by architecture
Sell into regulated & government accounts

Cloud-grade features

12-16% fewer errors than Whisper-based alternatives
Full speaker diarization and ID
55+ languages, real-time and batch

See it in action

How Adobe Premiere uses Speechmatics on-device

Adobe has partnered with Speechmatics since 2021 for speech-to-text in Premiere.
The new on-device model delivers within 5% of cloud accuracy, processed locally and never leaving the device.
12-16% fewer errors than Whisper-powered alternatives.

Engineered for laptop silicon

Built for the Mac and Windows hardware your users already have.

Runs on Mac

Optimized for CoreML

OSmacOS v14 Sonoma+

HardwareApple M1 or newer

RecommendedmacOS v26 Tahoe

AccelerationNeural Engine + GPU

Runs on Windows

DirectML optimized · GPU compute

OSWindows 11

HardwareIntel / AMD / ARM

Recommended2GB VRAM min

AccelerationDirectML GPU compute

Built for where your users already work

Laptop-first workflows

When your users need transcription that works everywhere their laptop goes. On a plane, in a courtroom, at the bedside.

Excels in scenarios requiring privacy, offline capability, or handling sensitive data in regulated environments.

Video editing & captioning

Local video editing tools with real-time transcription baked in. No uploads, no waiting, no cloud dependency.

Healthcare & legal scribes

Ambient scribe and dictation tools that keep working when hospital Wi-Fi drops. Patient and client data never leaves the device.

Regulated industries

Financial services, insurance, and legal. Sectors where compliance by architecture beats compliance by policy.

Government & law enforcement

Transcription where on-device processing simplifies security clearance and removes cross-border data transfer concerns.

Edge AI assistants

Deploy to devices, assistants, and enterprise systems without server costs or bottlenecks. Real-time, always available.

Note-taking & meetings

Meeting transcription that works in air-gapped environments, on flights, and in sensitive boardrooms.

Not all on-device STT is created equal

Speechmatics' on-device model matches its cloud model. Open-source alternatives ship stripped-down server models instead.

[alt:A stylized horizontal bar graph comparing multiple values, with one central bar highlighted in a darker color to indicate a specific benchmark or result.]

Accuracy

Cloud-parity accuracy

Our 2026 on-device model delivers the same accuracy as our cloud STT. Same architecture, distilled and quantized for laptop silicon with hardware-native acceleration.

[alt:A digital transcript interface showing a dialogue between three speakers discussing a weather forecast, with text bubbles clearly separated by speaker labels.]

Features

Full speaker diarization & identification

Includes our speaker diarization and speaker identification, with 12-16% fewer errors than Whisper-based alternatives.

[alt:A realistic globe centered on the continents of Africa and Europe, floating against a green background with abstract curved lines.]

Scale

Adobe runs it in production

Adobe Premiere has run Speechmatics on-device since 2021, alongside ISVs in legal, healthcare, and media production.

[alt:A laptop device displaying an audio waveform visualization on its screen, with a location pin icon floating above to symbolize local processing or location services.]

Future

Built for on-device deployment

Speechmatics On-Device is a first-class deployment target, built for laptops from the ground up, not a stopgap until connectivity returns.

Offline speech-to-text FAQs

Pricing is based on your deployment volume and use case. Speak to our sales team for a tailored quote and volume-based discounts.

Operating System	Hardware requirements
macOS v14 Sonoma or later, Tahoe v26 recommended	M1 or newer
Windows 11	Intel/AMD/ARM (with GPU 2GB)
Other Hardware	Get in touch if you have different requirements

Our 2026 on-device model has parity with the Speechmatics Standard cloud model. Same underlying architecture, distilled and quantized for local hardware. Includes full speaker diarization and speaker identification, with 12-16% fewer errors than Whisper-based alternatives.

The on-device model is compiled with hardware-native acceleration paths. On Mac, it leverages CoreML to run on the Neural Engine and GPU. On Windows, it uses DirectML for GPU compute. This means maximum throughput within a minimal resource envelope. No platform-specific code from your side.

Requires approximately 1 CPU core, an AI accelerator (Neural Engine on Mac, GPU on Windows), and ~800MB of system memory. No external GPU or dedicated inference hardware needed. Standard business and consumer laptops handle it comfortably. Contact us to learn more about specific deployment configurations.

On-device speech-to-text for laptops.

Trusted by millions of users globally

Driving better conversations at scale

Redefining real-time captioning

Delivering 120X more with voice AI

Cloud-grade speech recognition on-device for Adobe Premiere

Driving better conversations at scale

Redefining real-time captioning

Delivering 120X more with voice AI

Cloud-grade speech recognition on-device for Adobe Premiere

Deploy enterprise STT on every laptop

Why run offline STT on the laptop?

See it in action

Engineered for laptop silicon

Engineered for laptop silicon

Runs on Mac

Runs on Windows

Built for where your users already work

Built for where your users already work

Video editing & captioning

Healthcare & legal scribes

Regulated industries

Government & law enforcement

Edge AI assistants

Note-taking & meetings

Not all on-device STT is created equal

Not all on-device STT is created equal

Cloud-parity accuracy

Full speaker diarization & identification

Adobe runs it in production

Built for on-device deployment

Offline speech-to-text FAQs

How much does it cost?

What hardware and devices do you support?

How does on-device accuracy compare to cloud?

What about CoreML and DirectML optimization?

What are the resource requirements?