cactus transcribe

Real-time speech-to-text transcription

Overview

The cactus transcribe command provides speech-to-text capabilities using on-device models. Supports both live microphone input and audio file transcription.

Syntax

cactus transcribe [model] [flags]

Arguments

• [model] - Optional transcription model name. Default: parakeet-1.1b

Flags

—file

Transcribe an audio file instead of live microphone input:

cactus transcribe --file <path/to/audio.wav>

Supported formats: WAV, MP3, FLAC, OGG

—precision

Set the quantization precision level:

cactus transcribe --precision INT4|INT8|FP16

Default: INT4

Options:

• INT4 - 4-bit quantization (smallest size, fastest)
• INT8 - 8-bit quantization (balanced)
• FP16 - 16-bit floating point (highest quality)

—token

Provide a HuggingFace API token for gated models:

cactus transcribe <model> --token <your-hf-token>

—reconvert

Force reconversion of the model from source weights:

cactus transcribe --reconvert

Examples

Live Microphone (Default)

# Start live transcription with default model
cactus transcribe

Live Microphone Mode

When run without --file, transcription starts from your default microphone:

┌─────────────────────────────────────────────┐
│ Cactus Transcribe - parakeet-1.1b           │
│ Listening... (press Ctrl+C to stop)        │
└─────────────────────────────────────────────┘

[00:00:03] Hello, this is a test of the
           transcription system.

[00:00:08] It's working really well and
           capturing my speech accurately.

File Transcription Mode

With --file, the audio is processed and transcription is displayed:

cactus transcribe --file podcast-episode.mp3

┌─────────────────────────────────────────────┐
│ Transcribing: podcast-episode.mp3           │
│ Duration: 45:32                             │
└─────────────────────────────────────────────┘

Processing... ████████████████████ 100%

[Transcript]

Welcome to today's episode. We're going to
discuss the latest developments in...

Default Model

The default transcription model is parakeet-1.1b, which provides:

• Fast real-time transcription
• Good accuracy for general speech
• Low memory footprint (~300MB INT4)

Available Models

Supported speech-to-text models:

• parakeet-1.1b (default)
• moonshine-base
• moonshine-tiny

See the Model Library for full specifications.

See Also

Download Command

Pre-download transcription models

Test Command

Run STT benchmarks