Configuration

VoicePad stores its configuration in a YAML file. You can edit this file directly or change settings from the Settings tab in the interface.

Config File Location

The global config file lives at:

~/.config/voicepad/voicepad.yaml

On Windows, this expands to C:\Users\<YourName>\.config\voicepad\voicepad.yaml.

This file is created automatically when you save settings from the interface. If the file is missing or invalid, VoicePad shows onboarding again and rebuilds the configuration from your selections.

Settings

recordings_path: ~/.config/voicepad/data/recordings
markdown_path: ~/.config/voicepad/data/markdown
model_cache_path: ~/.config/voicepad/models
vad_model_path: ~/.config/voicepad/models/vad
logs_path: ~/.config/voicepad/logs
log_level: INFO

input_device_index: null
recording_prefix: recording

global_hotkey: ctrl+alt+v

transcription_model: turbo
transcription_device: auto
transcription_compute_type: auto
language: en
beam_size: 1
transcription_vad_filter: false
initial_prompt: "Hello. This is a transcription with proper punctuation, capitalization, and grammar."
text_postprocessing_enabled: false
no_speech_threshold: 0.6
hallucination_silence_threshold: 2.0
hallucination_max_repetitions: 3
min_audio_duration_s: 0.5
min_fresh_speech_duration_s: 0.25
trim_trailing_silence_rms_threshold: 0.01
trim_trailing_silence_frame_ms: 20

silence_threshold_ms: 1000
min_chunk_s: 15.0
max_chunk_s: 29.0
overlap_s: 0.5
stream_poll_interval_s: 0.3
stream_context_chars: 200

dedup_prev_tail_words: 50
dedup_full_duplicate_threshold: 0.8
dedup_min_overlap_words_for_partial: 3
dedup_partial_lead_words: 5

vad_threshold: 0.5
vad_min_speech_duration_ms: 250
vad_speech_pad_ms: 30
vad_model_filename: silero_vad_v6.onnx
vad_model_url: https://raw.githubusercontent.com/SYSTRAN/faster-whisper/master/faster_whisper/assets/silero_vad_v6.onnx
vad_download_chunk_size: 8192

local_agreement_mic: false
local_agreement_file: true

model_warmup_enabled: true
model_warmup_duration_s: 0.5
model_warmup_language: en
model_warmup_beam_size: 1
model_warmup_vad_filter: false

Global Hotkey Format

The global_hotkey setting uses natural hotkey strings like ctrl+shift+space, alt+v, or f9. The hotkey works system-wide on Windows, allowing you to start/stop recording from any application.

Settings visibility

Not all settings are shown in the Settings tab interface. Advanced runtime tuning such as inference thresholds, VAD behavior, overlap deduplication, warm-up behavior, text post-processing, final-chunk speech gating, model download locations, and the full advanced model list can be changed by editing the YAML file directly.

Notes on Advanced Settings

text_postprocessing_enabled Keeps raw model text by default. When set to true, VoicePad re-enables its lightweight text cleanup pipeline.
min_fresh_speech_duration_s Controls how much new VAD-confirmed speech must exist in the final tail of a recording before VoicePad sends that tail to Whisper. This helps avoid hallucinated last-word or last-phrase endings.

For details on individual settings, see:

Whisper Models: choose the right model for your hardware
Input Device: select which microphone to use
Output Paths: change where recordings and transcriptions are saved
GPU Acceleration: NVIDIA GPU requirements and performance

Changing Settings

The easiest way to configure VoicePad is through the interface:

Open VoicePad: uvx voicepad
Go to the Settings tab
Change any value
Press Save

Changes take effect immediately. If you change the transcription model, VoicePad reloads it automatically.