What is Whisper?

Whisper is OpenAI's open-source automatic speech recognition (ASR) model, released in 2022. It supports multilingual transcription and is widely used for accented speech, technical terminology, and noisy environments.

Whisper comes in multiple sizes and variants. Larger models can improve accuracy but require more compute, while turbo variants are designed to improve the speed and accuracy tradeoff.

OpenAI Whisper API vs Groq Whisper

OpenAI hosts Whisper through its API. Groq hosts whisper-large-v3-turbo on its custom LPU (Language Processing Unit) infrastructure. End-to-end performance still depends on the selected model, audio length, network, provider load, and any cleanup the app adds.

DimensionOpenAI Whisper APIGroq Whisper (large-v3-turbo)
LatencyVaries by model, audio, network, and loadDesigned for low-latency inference; actual timing varies
ModelOpenAI-managed Whisper modelwhisper-large-v3-turbo
AccuracyTest with your language and audioTest with your language and audio
InfrastructureProvider-managedGroq LPU
StreamingNoNo
Language supportModel and provider dependentModel and provider dependent
PricingCheck current provider pricingCheck current provider pricing

Why latency matters for dictation

A noticeable pause after you stop speaking can break your train of thought. The useful measurement starts when you stop speaking and ends when clean text is ready to use.

On iPhone and Android, tap Mic, speak, then tap Stop. The useful latency begins when you tap Stop and ends when the cleaned text is ready to review.

AiType uses Groq-hosted Whisper because low-latency transcription helps keep that complete workflow responsive. AiType typically completes transcription and AI cleanup in about 250 ms. Actual timing varies by device, connection, language, and text length.

What about local / on-device Whisper?

You can also run Whisper locally with tools such as whisper.cpp. Local processing can keep audio on the device, but speed, battery use, and model quality depend on the hardware and model. AiType currently uses cloud transcription and adds a separate AI cleanup pass.

Why AiType uses Groq

The bottom line

Whisper is the model, and Groq is the inference platform AiType uses to run it. The product typically completes transcription and cleanup in about 250 ms, but your result will vary with the full workflow.

Test AiType's dictation workflow

14-day free trial. Mac, Windows, iPhone, Android.