On-Device Speech Models
All 100% offline neural models supported by OpenDictate. Zero cloud latency.
Parakeet TDT 110M (int8)
Defaultnon-streamingDefault out-of-the-box model. Ultra-fast with a tiny memory footprint.
FastConformer 80ms (Streaming)
streamingUltra-low 80ms chunk streaming. Words appear live as you speak.
Parakeet Unified 0.6B (Streaming)
streaming560ms streaming chunks with high acoustic robustness for noisy environments.
Parakeet Unified En 0.6B
non-streamingHigh-accuracy ASR optimized for complex paragraphs and developer vocabularies.
Parakeet TDT 0.6B (Multilingual)
non-streamingMultilingual transcription with fast token duration modeling.
Whisper Turbo (Large v3)
non-streamingHighest accuracy with flawless casing, punctuation, and markdown formatting.
Whisper Tiny (en)
non-streamingLightweight Whisper variant capable of fast execution on minimal CPU hardware.
Whisper Base (en)
non-streamingBalanced speed and accuracy for everyday dictation.
Whisper Small (en)
non-streamingHigh-precision transcription for specialized terminology.
Whisper Medium (en)
non-streamingDeep neural accuracy for long technical transcripts and meetings.
Zipformer EN 20M (Live Captions)
utilityInternal live-caption engine generating partial words live while accuracy models compute.
Silero VAD v4
utilityUltra-fast neural speech boundary detector that automatically triggers dictation.
Zipformer 3.3M (Wake Word)
utilityHandsfree wake word spotting engine for voice-activated triggering.
OpenDictate automatically downloads and manages models on demand when selected in the app.