Tech, ExplainedIndustrial tech and mobility
Part

In-car voice assistant

Ask for a charging stop in Hindi.

INSIDE: DIGITAL COCKPITHOW IT WORKS · FINDING A CHARGER BY VOICEThe driversays: nearestcharging point.The microphonerecords about3 secondsSpeechrecognitionconverts thesound intothe text ofthe requestThe assistantworks out theintent: find acharger near meIt asks the mapfor chargingstations within10 kmIt reads outthe answer:nearest is 6km away, 2 of4 chargers freeThe driver saysyes. ThenavigationstartsALSO TALKS TOSpeech-to-text (ASR) models↗ GENERATIVE AI APPLICATIONSINSIDE: DIGITAL COCKPITHOW IT WORKS · FINDING A CHARGER BY VOICEThe driver says: nearestcharging point. Themicrophone records about3 secondsSpeech recognitionconverts the sound intothe text of the requestThe assistant works outthe intent: find acharger near meIt asks the map forcharging stations within10 kmIt reads out the answer:nearest is 6 km away, 2of 4 chargers freeThe driver says yes. Thenavigation startsALSO TALKS TOSpeech-to-text (ASR) models↗ GENERATIVE AI APPLICATIONSINSIDE: DIGITAL COCKPIT ·FINDING A CHARGER BY VOICEThe driver says: nearest charging point.The microphone records about 3 seconds↓Speech recognition converts the soundinto the text of the request↓The assistant works out the intent: finda charger near me↓It asks the map for charging stationswithin 10 km↓It reads out the answer: nearest is 6 kmaway, 2 of 4 chargers free↓The driver says yes.The navigation startsALSO TALKS TOSpeech-to-text (ASR) models↗ GENERATIVE AI APPLICATIONS
The driver's spoken request goes from the cabin microphone through speech recognition to the assistant, which acts on the map, the phone or the cabin settings and reads out the answer.
opens a page inside this one a page in the same scene a page in another scene a concept page a label only, with no page

An in-car voice assistant is software that lets the driver ask the car for something by speaking. The driver can ask for a phone call, a route or a charging stop without taking their hands off the wheel or their eyes off the road. Newer assistants understand Hindi, Tamil and other Indian languages as well as English.

The assistant works in three steps. First, a microphone in the cabin records what the driver says. Second, speech recognition turns the sound into written text. Third, the assistant works out what the driver wants, and acts on it. It may start a call, ask the map for a route or change the cabin temperature. It then reads the answer aloud.

A car is a hard place for this. The engine, the horns and the air conditioner make noise, and several people may talk at once. Indian drivers also mix languages in one sentence, for example Hindi and English. The assistant has to handle all of this.

Some of the work happens on the car's own chips, so that it keeps working in a tunnel with no mobile signal. For instance, a driver on a hill road near Shimla says, in Hindi, that she wants the cabin at 22 degrees. The car's own chips understand this short command, and the air conditioning changes. No mobile signal is needed. A longer request, such as finding a restaurant, needs the carmaker's cloud.

Newer assistants also use foundation models, large AI models that can answer a question asked in everyday words. These can, for example, explain what a warning light means. The sound is first turned into text by a speech-to-text model, and the text is then given to the larger model.

The assistant is different from a phone-call system such as conversational IVR. A caller phones a bank and a computer answers. The in-car assistant is inside the vehicle, and it controls the car's own functions as well.

The assistant is one of the four parts of the digital cockpit. It sends its commands to the infotainment system for navigation and music. Voice assistants for cars are built by a few specialist suppliers, with engineering teams in India.

Tech, Explained is a part of ingrid.fyi. Every page explains one piece of technology in plain words.