Assistants & voice / Automotive
- Input → outputSpeechtoSpeech
- ModelLFM2.5-Audio-1.5B
- Runs onIn the car
- What it doesTool calls · 8 vehicle domains
Hear. Understand.Assist. On the road.
An automotive assistant with local intelligence for vehicle control.
Hosted browser demo · microphone access is optional · vehicle controls are simulated
Problem
Drivers say ‘It’s freezing in here’ or ‘Close the windows and play some jazz’ rather than a sequence of commands from a menu. Rigid voice systems struggle with everyday phrasing, follow-ups, and multiple actions in one turn. Cloud-dependent assistants also rely on coverage and send cabin audio off the vehicle, while the head unit has a fixed compute and memory budget.
Solution
A fine-tuned LFM2.5-Audio-1.5B takes speech in, interprets the request, emits vehicle tool calls, and produces the spoken reply. The same model handles climate, media, navigation, and other controls, including compound requests and conversational follow-ups. Local deployment keeps the experience available beyond coverage, retains cabin audio in the vehicle, and avoids a cloud API charge for every request.
Works beyond coverage.
Local inference keeps the assistant available in tunnels, garages, and rural roads.
Fits your voice experience.
Adapt the model to your vehicle’s context and controls, brand wake word, everyday phrasing, and languages.
Control the running cost.
On-device inference removes the cloud API charge from each driver request.