config.example.toml

# ==============================================================================
# ParrotInk - Professional Voice-to-Text Configuration
# ==============================================================================
# This is the main configuration file for ParrotInk.
#
# HOW TO USE THIS FILE:
# 1. Commented-out lines (starting with '#') show the application DEFAULT values.
# 2. To change a setting, remove the '#' and modify the value to your preference.
# 3. Save the file and ParrotInk will automatically reload the new settings.
# ==============================================================================

[hotkeys]
# The key combination used to trigger dictation.
# Supports modifiers: ctrl, alt, shift, win.
# Examples: "ctrl+alt+v", "shift+f2", "win+v"
# hotkey = "ctrl+alt+v"

# If true, you must HOLD the hotkey while talking.
# If false (Toggle Mode), click once to start and click again to stop.
# hold_mode = false

[interaction]
# Automatically cancel and discard transcription if you click away from the active
# input field (anchor) while recording.
# cancel_on_click_outside_anchor = true

# Defines what is considered 'outside' for cancellation:
# 'control' - Cancel only if you click a DIFFERENT text box or UI element.
# 'window'  - Cancel if you click ANYWHERE outside the active application window.
# Options: "control", "window"
# anchor_scope = "control"

# Only applies to Toggle Mode (hold_mode = false).
# If true, pressing ANY key (other than the hotkey) while recording will stop it.
# stop_on_any_key = true

# If true, ParrotInk will automatically launch when Windows starts.
# run_at_startup = false

[interaction.sounds]
# Enable auditory feedback (beeps) when recording starts or stops.
# enabled = true

# Volume of the feedback sounds.
# Range: 0 (silent) to 100 (max volume).
# volume = 30

# Paths to custom WAV files for start/stop sounds.
# Ensure the path uses double backslashes (\\) or forward slashes (/).
# start_sound_path = "C:\\Windows\\Media\\Speech On.wav"
# stop_sound_path = "C:\\Windows\\Media\\Speech Off.wav"

[ui]

# Show a welcome popup on the first launch to explain the app's mechanics.
# show_onboarding_popup = true

[ui.floating_indicator]
# Enable the floating Acrylic HUD that shows real-time transcription progress.
# enabled = true

# If true, the HUD ignores mouse clicks (allowing you to click through it).
# click_through = true

# Transparency level of the HUD background when the app is idle.
# Range: 0.0 (transparent) to 1.0 (opaque).
# opacity_idle = 0.3

# Transparency level of the HUD background when actively recording speech.
# Range: 0.0 (transparent) to 1.0 (opaque).
# opacity_active = 0.8

# Distance in pixels from the bottom of the screen to position the HUD.
# y_offset = 60

# Maximum number of characters to display in the HUD preview before truncating.
# max_characters = 180

# UI refresh frequency in milliseconds. Lower is smoother but uses more CPU.
# Range: 5 to 500.
# refresh_rate_ms = 50

[audio]
# The name or partial name of the microphone to use for capture.
# If empty or "default", the system's default recording device is used.
# input_device = "default"

# Internal sample rate for microphone capture.
# 16000 is the industry standard for most ASR models.
# capture_sample_rate = 16000

# Duration of each audio buffer chunk in milliseconds.
# Lower values reduce latency but increase CPU usage.
# Range: 50 to 500.
# chunk_ms = 100

# Connection management strategy:
# 'warm' (recommended): Keeps connection ready but idle to eliminate handshake lag.
# 'on_demand': Connects only when hotkey is pressed (saves bandwidth, adds initial lag).
# 'always_on': Connection never closes (highest resource usage).
# connection_mode = "warm"

# How long to keep a 'warm' connection alive without activity before closing it (in seconds).
# Range: 30 to 1800.
# warm_idle_timeout_seconds = 300

# Automatically stop recording in 'Toggle Mode' after this many seconds of silence.
# Range: 5 to 3600.
# inactivity_timeout_seconds = 30

# Sensitivity for local voice activity detection.
# Lower values are more sensitive (detect quiet speech).
# Range: 0.001 to 0.05.
# voice_activity_threshold = 0.005

# Timeout for the initial server handshake in seconds.
# connection_timeout_seconds = 20.0

# Timeout for application cleanup during shutdown in seconds.
# shutdown_timeout_seconds = 10.0

# Maximum number of retry attempts for a failed network connection.
# max_retries = 3

# Initial delay in seconds before the first connection retry.
# initial_backoff_seconds = 1.0

# Maximum delay in seconds between connection retries.
# max_backoff_seconds = 60.0

[transcription]
# The active transcription service provider.
# Options: "openai", "assemblyai"
# provider = "openai"

# Performance profile.
# 'fast'     - Prioritizes speed (faster finalization).
# 'balanced' - Good balance between speed and sentence completion.
# 'accurate' - Waits longer for natural pauses to ensure better context.
# Options: "fast", "balanced", "accurate"
# latency_profile = "balanced"

# Microphone environment optimization.
# 'headset' - Optimized for close-proximity microphones.
# 'laptop'  - Optimized for built-in or room microphones (far-field).
# 'none'    - Raw audio signal.
# Options: "headset", "laptop", "none"
# mic_profile = "none"

# If true, the app will type partial (interim) results as you speak.
# If false (Default), it will only type the final, high-confidence sentence.
# Note: The HUD floating indicator will show partials regardless of this setting.
# partial_results = false

# --- OpenAI Configuration ---

[providers.openai.core]
# Models allowed: "gpt-4o-mini-transcribe", "gpt-4o-transcribe"
# transcription_model = "gpt-4o-mini-transcribe"

# BCP-47 language code (e.g., "en", "es", "fr", "ja").
# language = "en"

# How long to keep a single OpenAI session open before refreshing (in seconds).
# session_rotation_seconds = 3300

[providers.openai.advanced]
# If true, the manual values below will override the 'latency_profile' settings.
# override = false

# Contextual instructions to improve ASR accuracy (jargon, names, etc.).
# prompt = ""

# Noise reduction filter level. Options: "near_field", "far_field", "off".
# noise_reduction = "off"

# Server-side VAD sensitivity. Range: 0.0 to 1.0.
# vad_threshold = 0.6

# Audio padding included BEFORE speech detection (in milliseconds).
# prefix_padding_ms = 300

# Silence duration required to finalize a segment (in milliseconds).
# silence_duration_ms = 500

# --- AssemblyAI Configuration ---

[providers.assemblyai.core]
# Region host for AssemblyAI. Options: "us", "eu".
# region = "us"

# Language code for multilingual streaming.
# language_code = "en"

# The transcription model to use.
# Options:
# "u3-rt-pro" (Default) - Latest model with instructional prompting support.
# "universal-streaming-english" - Legacy English-only model.
# speech_model = "u3-rt-pro"

# Natural language instructions to guide the transcription (e.g. "Use medical terms").
# Note: Mutual exclusive with 'keyterms_prompt'. If both are set, 'prompt' wins.
# prompt = ""

# Server-side VAD sensitivity. Range: 0.0 to 1.0.
# vad_threshold = 0.4

# Time in seconds of inactivity before session auto-termination. 0 = disabled.
# Range: 0 or 5 to 3600.
# inactivity_timeout_seconds = 0

[providers.assemblyai.advanced]
# If true, the manual values below will override the 'latency_profile' settings.
# override = false

# If true, the provider adds punctuation and capitalization to final text.
# format_text = false

# List of terms to boost recognition for (max 100 terms, 50 chars each).
# Example: keyterms_prompt = ["ParrotInk", "Gemini"]
# keyterms_prompt = []

# Confidence threshold for detecting end of sentence. Range: 0.0 to 1.0.
# end_of_turn_confidence_threshold = 0.4

# Minimum silence needed to finalize a turn (in milliseconds).
# min_end_of_turn_silence_when_confident_ms = 400

# Maximum silence allowed before a turn is forced to finalize (in milliseconds).
# max_turn_silence_ms = 1000

# If true, detects spoken language (Requires multilingual model).
# language_detection = false

[logging]
# Enable logging to a local file in %LOCALAPPDATA%\ParrotInk\Logs.
# This is highly recommended for troubleshooting as the app has no console window.
# file_enabled = true

# Detail level for the log file.
# "error"   - Only log critical failures (Recommended for production).
# "info"    - Log major application events.
# "verbose" - Log everything for deep debugging.
# Options: "error", "info", "verbose"
# file_level = "error"

# Maximum size of a single log file in bytes. Default is 5MB.
# file_max_bytes = 5242880

# Number of archived log files to keep before deleting the oldest.
# file_backup_count = 5

# Custom path for the log file. If empty, uses the default AppData location.
# file_path = ""

[test]
# FOR DEVELOPERS ONLY: Use local mock servers instead of real API endpoints.
# enabled = false
# openai_mock_url = "ws://localhost:8081"
# assemblyai_mock_url = "ws://localhost:8081"