---

LFS-STT - Live for Speed speech to text

lfs-stt is a InSim plugin for Live For Speed that enables speech-to-text chat using OpenAI’s Whisper model. Players can record audio in-game, have it transcribed, and send messages directly into the server chat. Multiple chat channels can be configured with custom prefixes — especially useful on cruise servers or organized events.

Features

• Record messages in-game via configurable binds
• Speech-to-text using Whisper
• Message preview before sending
• Cycle through multiple chat channels (message prefixes)
• Configurable UI position, scale, and timing settings
• Optional GPU acceleration for faster transcription (highly recommended as running on CPU is very slow)

---

In-Game Setup

Set up the following InSim binds in LFS:

Command	Description
/o stt talk	Toggle recording on/off
/o stt accept	Accept the message in preview and send it to the server on the selected channel
/o stt nc	Select the next chat channel (cycles back to the first channel after the last one)
/o stt pc	Select the previous chat channel

---

Configuration

All plugin settings are managed via a TOML configuration file (config.toml).

Example config.toml

# ================================
# InSim connection settings
# ================================

# InSim host
insim_host = "127.0.0.1"

# InSim port
insim_port = "29999"

# ================================
# Model / AI settings
# ================================

# Path to the whisper model directory or file
# By default the plugin ships with a small English-only model, which is sufficient for most use cases.
# Download other models from https://huggingface.co/ggerganov/whisper.cpp
model_path = "models/small.en.bin"

# Whether to use GPU acceleration to run the speech-to-text model
# Requires Nvidia GPU and CUDA installed
use_gpu = false

# ================================
# Timing settings
# ================================

# How long message previews stay visible
message_preview_timeout_secs = 20

# Maximum message recording duration
recording_timeout_secs = 10

# ================================
# UI layout settings
# ================================

# UI scale factor
ui_scale = 5

# Vertical UI offset (0–200)
ui_offset_top = 170

# Horizontal UI offset (0–200)
ui_offset_left = 10

# Button ID offset (0–230)
# Use if buttons are conflicting with other InSim plugins
btn_id_offset = 50

# ================================
# Advanced settings
# ================================

# When true, last recorded message is saved to debug.wav
debug_audio_resampling = false

# Logging verbosity
# Valid values: error, warn, info, debug, trace
debug_log_level = "info"

# ================================
# Chat channels
# ================================
# You must define at least ONE chat channel.
# Each channel needs a non-empty display name.
# Add as many channels as you want.

[[chat_channels]]
# What you see in the UI
display = "/say"
# Message prefix sent to InSim
prefix = ""

[[chat_channels]]
display = "^5!local"
prefix = "!l"

---

Usage

1. Download the latest release
2. Configure InSim binds in LFS as described above.
3. Launch LFS and launch lfs-stt.exe.
4. Press your talk bind to start recording, press it again to stop. Press accept to send the transcribed result.
5. Use nc / pc binds to switch between chat channels.

---

Tips

• Default channels: The plugin comes with /say and !local configured by default, but you can change them or add more by adding more [[chat_channels]] blocks.
• GPU usage: Enable use_gpu = true only if your system supports it — otherwise CPU works fine.
• UI customization: Adjust ui_scale, ui_offset_top, and ui_offset_left to avoid overlapping with other InSim plugins.
• Logging: debug_log_level can help troubleshoot issues — set to debug or trace during testing.

---

Contributing

• PRs and bug reports welcome.
• Disclaimer - I'm using this project as an opportunity to learn Rust.

---