2025.1.0.1

⭐ Hybrid Recognition & Intent-based commands

Forget exact command phrases!
Speak your mind 🤯

Introducing improved, neural speech recognition running locally in DRE!

DRE now understands you much better with hybrid voice recognition system.

Instead of having to memorize every single voice command phrasing, you can freely command DRE in your own voice.
DRE has shown to understand more than 80% of ANY DRE/racing related voice prompts

⭐ Neural Voice Recognition

• DRE now supports a new speech recognizer: Local-running neural speech recognition
• This new model doesn’t rely on the classic Windows Speech Recognition where inputs are phonetically matched to an array of available exact phrases.
• Rather, in neural speech recognition, everything can be spoken as DRE now transcribes your voice input – so speak your mind!
• Full support for all 7 languages in DRE (EN, DE, FR, ES, IT, PT, BR)
• Choose between Neural model qualities in Sound -> Recognition -> Local Speech Model if the recognition is too good/poor:
  • The Speed vs Quality of the Local Recognition model
    • Fast & Alright: A small and lightening fast model that yields okay-ish results. Consider this when you have an older GPU.
    • Medium & Good: A medium-sized model which balances quality and speed. This is the perfect option for most use-cases and is the default
    • Slow & Great: A large and very accurate, but slightly slower model. Consider this when your GPU is newer and has the headroom for processing on top of running simulators.
• These models use between 50 – 500MB more RAM than the previous DRE process depending on the quality of the model. Using Classic recognition instead will not load the neural recognition model.

⭐ Intent-Based Commands

• Leveraging the new Neural Voice Recognition, it’s now possible to speak your mind to which DRE will attempt to match your intention to one or more commands.
• A special and custom trained local language model, Thyrring, takes care of expediting voice prompts to commands, as it looks at your semantic intent behind your words, using vector search.
• In the past, you would need to memorize the exact phrases to get the job done (or create custom ones)
• With Intent-Based commands you can naturally vary your speech to your natural flow
• Let’s take a few example
  • Exact phrase: Auto fuel now
  • New option examples: I need fuel to go to the end or Calculate the fuel automatically
• Some command types need you to be slightly more specific in order for them to trigger:
  • Chat with other drivers must use the exact phrasing like Chat with {DRIVER} ...
  • Note-taking command must use the exact phrasing like Remember to ...
  • Admin commands must include the word admin
• Though DRE has taken a big step upwards in cleverness, here are some things DRE cannot do:
  • Respond to anything you throw at it
    • DRE is guard-railed to focus on (sim)racing and DRE-related topics. When voice prompts are out of scope, you’ll hear DRE responding That's not something I can talk about and similar
  • Extract multi-intents from a single sentence
    • For example, prompting DRE to Set fuel buffer to 5 laps are in fact two calls: Changing Fuel Buffer Mode to Laps as well as setting the value to 500%
    • Instead, break the prompt into two sentences like Set the fuel buffer to Laps, and set the Fuel Buffer Laps to 500
  • Cook your eggs, but it is cooking!

⭐ Hybrid Voice Recognition

• Choose between voice input speech engines for optimized communication with DRE:
  • Classic: Windows Speech Recognition (WSR) (local built-in, fast and lightweight for direct phrase matches, but not precise otherwise) (previously this was the only option. Choose this to revert to the good(?) old times)
  • Neural: Automatic Speech Recognition (local model using CPU or GPU, ranging from medium to fast and more computationally expensive, but quite accurate)
  • Hybrid: Both WSR and Local Speech in use, combining their efforts by allowing the WSR to initially check for precise matches, before sending it off to the Local Speech recognizer (leveraging the best of both engines)
• Adjust this in Sound -> Recognition -> Speech Engine (Changing this value will restart DRE)
• If you don’t fancy the neural recognition to utilize your GPU, and instead use the slower CPU neural inference, you can change this by exiting DRE, editing the DRE Config.json NeuralRecognitionPreferredRuntimeOrder order, and restart DRE.
  • Most, if not all users, should keep this setting as-is, to true for GPU inference
  • As an example, on an older GPU like 1080Ti, Neural voice recognition creates only a smaller footprint and has not seen any effects to the in-sim performance

⚠️ Note! DRE uses Vulkan runtime by default which supports both NVIDIA and AMD GPUs.

You if have an NVIDIA GPU, consider downloading and installing CUDA Toolkit for slightly faster inference. Choosing Custom (Advanced) installation and selecting only the CUDA -> Runtime components should be sufficient. More info

⭐ Chat View

• Modern chat-styled interaction with DRE
• This new view allows you to easily follow along with your communication with DRE
• Four options to interact with DRE:
  1. Speak using voice as usual, using PTT, VoiceActivated, or other mic input modes
  2. Hold the microphone button down in the Chat View to mimic PTT
  3. Type in your query
  4. Click any of the suggestion chips below the input area. These are randomized naively from the current available and active voice commands.
     

⭐ Classic Recognition Min Confidence: Offset slider

• Percent points increase on minimum confidence required for the Classic recognizer in Hybrid recognition mode (Classic + Neural).
• This has been implemented to prevent DRE from being too trigger-happy from the classic speech recognizer, to allow the Neural recognizer to take action, or wait for true local speech recognized confirmations.
• Higher values give more responsibility to the Neural recognizer.
• This offset is only in effect for Classic recognitions with Speech Engine set to ‘Hybrid’
• Defaults to 30% points, adjust this in Sound -> Recognition -> Min Confidence: Classic Recognizer Offset

⭐ Interfaced DRE settings with Voice Recognition

• You can now manipulate the DRE settings using your voice (or text chat) to do the following:
  • Request general info about the setting
  • Request the current value
  • Locate the setting in the UI (DRE switches to the View and tab)
  • Set a new value (supports direct value, incrementing, comma delimited string settings are not supported)
• For example, you can say:
  • Enable Delta Wave for me please -> activates Delta Wave Signal
  • I don't want to hear Pit Speeding Signal anymore -> Deactivates Pit Speeding Signal
  • What is Cool Lapping all about? -> Summarizes what the Cool Lapping feature does
  • Find Overtake alert for me -> Locates and shows the Overtake Alert setting in the DRE UI
  • Lower ambience volume by 15 -> Decreases the Ambience Volume setting 15%
• This is enabled at all times for Neural recognition, but if you want the Classic recognizer to be able to use these, enable Sound -> Recognition -> Load User Config grammar for Classic (requires DRE restart)

🔄 Dynamic Command availability

• Commands are now only enabled and available for the speech recognizers when they meet certain criteria.
• For example
  • any command that affects the car engine (starting, killing it) will only be enabled once in-car.
  • any camera commands are only enabled once out of the car, in replay-mode
• Hover any greyed-out command in Commands page to see when it will be active
• 

⭐ Multi-Sim: Le Mans Ultimate

The official World Endurance Championship simulator, Le Mans Ultimate, has been added to DRE.

DRE now supports the following simulators:

• iRacing (Primary)
• Le Mans Ultimate (LMU) (Secondary)
• Assetto Corsa Competizione (ACC) (Secondary)

DRE supports multiple simulators and will primarily track iRacing’s features and available commands.

Other simulators are supported only for the subset of features common with iRacing, without unique features of other simulators.
Support for Assetto Corsa Evo is the next in line

DRE automatically installs the required plugin and configuration files for LMU, making it easy to get started racing with DRE 🙌

🔄 Fueling Notes for Le Mans Ultimate

LMU uses absolute, total fuel values for Hypercars and LMGT3 cars when setting fuel levels for the upcoming pit stop.

• Instead of specifying how much fuel you want to ADD to the car, you specify how much NEEDED when leaving the pit stop.

Furthermore, LMU (and WEC) uses a combination of Virtual Energy (VE) and Fuel Ratio to calculate the fuel levels.
When you command DRE to add a certain amount of fuel, DRE attempts to fulfill this request by:

• Setting an immediate NEEDED fuel amount in the car based on the current fuel level in the car. Coming to the real pit stop later on, however, this value will no longer be valid.
• Coming to the actual pit stop time; When heading into the pit lane before the pit stall, DRE now once again sets the earlier requested fuel amount to be added.
  • This time, the values will remain valid for your remaining session as fuel levels in the car won’t change much between pit-in and your pit stall.
  • If you auto-fueled in between, DRE will ignore re-setting your previous set pit stop fuel amount
• DRE Auto-Fuel has its own internal calculations of the fuel NEEDED to finish the session, so when DRE triggers Auto Fueling it uses the FUEL NEEDED and won’t have to re-set the fuel levels when getting to pit-in.

The way DRE calculates the fuel to add:

• If the fuel to set is specified as the fuel NEEDED (the fuel to leave the pits with after the stop), the fuel is simply set
• Otherwise, the fuel to add is added onto the existing fuel level
• It then converts from the fuel to ADD to the fuel NEEDED
• To interface with LMU, the fuel NEEDED value is now converted to VE by dividing the fuel NEEDED by the Fuel Ratio currently set
• If the VE calculated for the fuel NEEDED is within bounds of the current Fuel Ratio set (is 100% or less), the VE is simply set
• Otherwise, DRE attempts to raise the Fuel Ratio up until its maximum limit to meet your request as well as setting the VE to 100%

⭐ New Binds

• View: Log – bind a control, or speak the command Show DRE log to show the DRE Log view
• View: Chat – bind a control, or speak the command Show DRE chat to show the DRE Chat view
• Simulator Text Chat activation
  • An in-simulator text chat activation binding has been added to support typing text chat messages in LMU Multiplayer
  • This binding defaults to the key "T"
  • Call this activation verbally by Activate chat

⭐ Low Fuel in Qualification

• Added toggle to enable/disable Low Fuel alerts during qualification
• Disabled by default, allowing your qual runs to out fuel should you be that tight on your margins 🤨

🔄 Analytics

• LMU transmits real-time Tire Wear, which is now reflected in the Tires page on every new lap
• Fixed: Tire Wear graph would not always follow the lowest wear values on the Y-axis
• Session Laps is now titled "Laps Left" for clarity
• Laps Left box has had a small ‘~’ added to the total race laps for you when not in a lap-based race being the overall leader

🔄 Log view is now hidden by default

• With the introduction of the Chat view, the Log view has less of its original merits
• The Log view is now hidden by default, to allow focus on the Chat
• You can restore the Log by enabling Settings -> General -> Log Shortcut in Side Menu
• Alternatively, you can show the log via Settings -> General -> Show Log

🔄 Fuel carry-over

• Session changes now better carry over the previous session’s fuel numbers
• Without fuel data for the specific track/car combo, DRE now trusts the actual driven laps way more than the synthesized car fuel rate
• This should align much better with actual fuel numbers in the opening parts of the qual and race sessions

🔄 Fuel Only command

• The fuel amount to set in the fuel only command has until now been to maximize fuel filling to the tank capacity.
• However, with the introduction of the maximize fuel command, the Fuel Only command will now instead set fuel to the estimated amount needed to finish the race, or the next pit stop in the strategy.

✅ Fuel Rates

• Fuel rates for various track conditions and tire compound mixes did not multiply correctly onto the default dry track + dry compound fuel rate
• This caused weird fuel rates for non-dry running

✅ Session Laps

• Fixes and improvements to the session laps duration algorithms
• Lap-based multi-class races
  • Laps Left in multi-class + lap-based races are now correctly affected by the difference between the expected overall leader at checkered and drivers’ lap times
  • This also takes pit stops time losses into account
• Time-based races
  • Races defined by time now better take into account the expected overall leader at the end
  • The logic now looks at the moment after the time has expired when the leader is expected to cross the S/F to take the checkered

✅ Debris alerts waited politely for your busy-state to end

• Now, Debris warnings will fire immediately, interrupting any current DRE speech

✅ Overtake alerts could start out with a "clear" call

• This occurred when the cars entered and exited their overtake zone in rapid succession.
• Now, a 250ms blocking period has been introduced for clear calls after the overtake has started

✅ Bugs caught by the internal bug tracker