dikt

Voice dictation for the terminal. Record, transcribe, copy — zero npm dependencies.

Uses Mistral's Voxtral for speech-to-text.

Install

npm install -g dikt

Requires sox for audio recording (not needed for --file):

# macOS
brew install sox

# Ubuntu/Debian
sudo apt install sox

# Arch
sudo pacman -S sox

Optional dependencies for --file mode:

• ffmpeg — enables compression, chunked transcription of long files, and broader format support
• yt-dlp — enables transcribing audio from URLs (YouTube, podcasts, etc.)

Setup

On first run, dikt will prompt you for your Mistral API key and model preferences:

dikt setup

Config is stored in ~/.config/dikt/config.json.

Usage

dikt

This opens an interactive TUI where you can record, transcribe, and copy text.

Keys

Key	Action
Space	Start / stop recording
c / Enter	Copy transcript to clipboard
a	Toggle auto-copy
h	Cycle through history
r	Re-transcribe last recording
Esc	Cancel recording
s	Re-run setup
?	Show keybindings
q	Quit

Single-shot mode

# Print transcript to stdout
dikt -q

# Output JSON
dikt --json

# Pipe to another tool
dikt -q | claude

# Wait longer before auto-stopping
dikt -q --silence 5

Stream mode

Continuously transcribe, emitting chunks on pauses:

dikt --stream

# Stream as JSON Lines
dikt --stream --json

# Stream as continuous flowing text
dikt --stream -n

# Stream continuously until Ctrl+C
dikt --stream --silence 0

File mode

Transcribe an existing audio file (wav, mp3, m4a, flac, ogg, webm, aac, wma, and more):

dikt --file meeting.wav

# Save to a file (.json auto-enables JSON output)
dikt --file meeting.wav -o transcript.json
dikt --file meeting.wav -o transcript.txt

# With JSON output
dikt --file recording.mp3 --json

# Transcribe from a URL (requires yt-dlp)
dikt --file https://youtube.com/watch?v=VIDEO_ID
dikt --file https://youtube.com/watch?v=VIDEO_ID -o transcript.txt

Speaker identification & timestamps

# Speaker labels
dikt -q --diarize

# Timestamps
dikt -q --timestamps segment
dikt -q --timestamps word
dikt --file lecture.mp3 --timestamps segment

# Combined with JSON
dikt -q --json --diarize

Options

Flag	Description
--file <path|url>	Transcribe audio file or URL (via yt-dlp)
-o, --output <path>	Write output to file (.json auto-enables JSON)
--stream	Stream transcription chunks on pauses
--json	Output JSON (single-shot or stream)
-q, --quiet	Record once, print transcript to stdout
--silence <seconds>	Silence duration before auto-stop (default: 2.0)
--pause <seconds>	Pause duration to split stream chunks (default: 1.0)
--language <code>	Language code, e.g. en, de, fr (default: auto)
--timestamps <granularity>	Add timestamps: segment or word
--diarize	Enable speaker identification
-n, --no-newline	Join stream chunks without newlines
--no-color	Disable colored output
--no-input	Fail if config is missing (no wizard)
--setup	Run setup wizard
--update	Update to latest version
--version	Show version
-h, --help	Show help

Update

dikt update

Environment variables

Variable	Description
DIKT_API_KEY	Override API key
DIKT_MODEL	Override model (default: voxtral-mini-latest)
DIKT_LANGUAGE	Override language (default: auto)
DIKT_TEMPERATURE	Override temperature
DIKT_CONTEXT_BIAS	Override context bias

License

MIT