Advanced ASR Models

Settings

Transcription settings overview screenshot

Screenshot

What this page is for

Advanced ASR models are offline file routes, starting at Standard, in the ML Engine group in the model picker (e.g. Qwen3-ASR, Cohere, Parakeet). They sit beside Whisper and realtime models, but they target offline files—not low-latency live capture.

This page covers when to use each model.

Before you start

Mode: focused on offline file transcription today, not realtime-model mic latency.
Plan: Free cannot select these models; Standard gets SenseVoice and Moonshine; Pro (including Lifetime) gets all of them (SenseVoice, Moonshine, FireRed, Qwen3-ASR, Parakeet, Cohere).
Download: first use requires downloading the model in Settings; sizes are often larger than Tiny/Base Whisper.

Model overview

Model	Good for	Languages	Notes
Qwen3-ASR	Meetings, courses, Chinese-heavy media	Chinese, English, Japanese, Korean, and more	Multilingual with punctuation
Cohere Transcribe (multilingual)	International interviews, mixed sources	Many widely used languages	High-quality offline multilingual
Parakeet	English/European bulk work	English plus European languages	Relatively compact for English-first libraries

See the in-app model card for the authoritative language list.

Whisper vs realtime vs advanced ASR

Need	Lean toward
Live subtitles, low latency	Realtime models (or GPU Whisper realtime)
GGML ecosystem, links, watch folders	Whisper
Offline files, specific multilingual quality	Advanced ASR (ML Engine tab)

See Models & Engines Overview and Concept.

Parameters

Advanced ASR scenes use their own advanced fields—not Whisper VAD tables. See parameters by engine.

FAQ

Does transcription need the internet?
Inference is local; downloads and licensing may use the network.

Can I run advanced ASR and GPU Whisper together?
Pick one model per job; GPU packages primarily serve Whisper routes.

After upgrading?
Past transcripts stay as-is. New jobs can pick whichever ML Engine models the new plan includes.