Speech control: More bright? OR more " catch mad " ?

Identify and understanding navigation instruction is accurate, can " one arrives directly " .

It is more difficult that speech controls window of air conditioning, car to wait.

Artificial intelligence has in the application on the car a lot of, speech control is a kind among them only. Nevertheless, still lose sight of in other application feel when be not being worn, the speech control that has gained ground gradually solemns is the spokesman of artificial intelligence. Go to 400 thousand yuan from 80 thousand yuan, horse of most valuable treasure boards to Chang'an below; On board mount speech controls a system. This period " special test " will detect it is good that speech is controlled after all use or bad to use.

About " speech control "

Alleged " speech is controlled " , it is the function that will control a car through speech. Speech control divides composition by two: Semantics identifying and car are controlled. The system is first " understand " speech instruction, machine language of speech instruction translate into, control traffic then.

Test purpose:

At present main supplier has speech control system two: One is Nuance, another is division big question flies. According to this two old systems, gentleman of car the world chooses two different models, what test speech control uses is convenient rate of gender, response and accuracy rate.

Test method:

Test car is on urban road with regular rate travel, the noise inside railroad car and shake with use daily consistent. Issue a statement by the driver that takes on driver's seat, observe and record the response of car. The error that to prevent different person different oral speech sound produces, whole test gives out speech to dictate by same individual, use language is unified for mandarin.

Test project 1: Navigation goes without giving thought to, can " one arrives directly "

Face the mobile phone navigation of change quickly, the car carries navigation by desolate not be any more. The operation is complex, without real time road condition, show insufficient clarity, map is updated too slow... the each chamfer place that the car holds navigation is worth our gush one gush. Even if passes An Jixing and Gbook, can a phone gives navigation tiring-room comes true, also cannot persuade a driver to abandon mobile phone navigation.

This kind of circumstance controls a system to appear till newest speech, just have so one man nods a hope to change. Gentleman of car the world sits on driver's seat, try early or late " navigation goes to such-and-such ground " , " I should go to such-and-such ground " , " I need to cheer " , the system can search destination and interest drop automatically, show on liquid crystal screen come. On search result page, still can have all sorts of operations, can say directly for instance " the 3rd " , " the 3rd page " and " cancel " .

Whole unit process of cargo bandling, no matter be Nuance, or division big question flies, the accuracy that identify and understands driver instruction is taller, it is to be able to accomplish what propagandist language place says really " one arrives directly " . Two systems that test a car react speed has speed to distinguish, this depends on the processing capability that the car carries ECU chip. But a bit can affirm, even if is the slowest, also be " the second replies " speed, did not give gentleman of car the world laggard and expatiatory sense. After the test, gentleman of car the world tells beside everybody, had speech control, can abandon really mobile phone navigation.

Test project 2: Phone missay, can change directly

Want to control implementation mobile phone to communicate through speech, want join mobile phone first, allow a system to read take address book. If not be oneself car, do not suggest so do, that meeting does not have any privacy but character. But to experience this one function, gentleman of car the world can sacrifice only oneself privacy, "Go ahead regardless " .

When the test, gentleman of car the world tries to sign up for a name directly first, for instance " phone some such-and-such " . Systematic identifying is accurate, response is quick. In address book, did not find corresponding name, meeting give out for a short while feedback. Gentleman of car the world tries to sign up for telephone number again, for instance " 139####2276 " . As long as language fast not be too fast, the system also can identify, show a number one by one on screen. Check special situation finally, number of newspaper skill plane 11 numbers, be when the system when join, gentleman of car the world changes instruction, for instance " will the last number, from ' 6 ' change ' 8 ' " . The system is met stop to be dialed for a short while, change new number, dial a telephone again.

Through speech control dials a telephone, still have a lot of fresh functions. For instance you are worn urgently change airline ticket, although address book of your mobile phone does not have customer service phone, but can say directly " phone of southern aviation customer service " , net of can automatic couplet searchs the system, join dial. Whole test process, do not need to relapse " affirm " , simplified flow, promoted the experience of the user.

Test project 3: Car controls meeting respecting mouth to do glossal dry

The test is over " navigation " and " phone " , reoccupy speech controls car function, the effect whether constant? Gentleman of car the world checks first " air conditioning " , speech input: "I am too cold " , " I am too hot " and " wind speed is too big " wait for 10 kinds of instructions. Let systematic identifying dictates and finish relevant air conditioning statement, 10 in have 2~3 only second success, when this or luck are better. Star-crossed when, 10 in did not succeed once. Check this one, gentleman of car the world rectifies an individual to become bad, him suspicion is mandarin not standard.

Is Nuance only such? But change division big question to fly system testing result also is same. According to introducing, "Navigation " , " phone " and " recreation " , it is " be born formerly " , the supplier offers, install namely have with the effect namely assure. "Air conditioning " belong to car function, need system and model undertake matching. Match degree intervene with the system truckload the time place of research and development is concerned, intervene earlier, the speed of semantics identifying and accuracy rate are jumped over tall.

Have only " air conditioning " such? Speech of test personnel attempt is controlled " car window " and " scuttle " , the result also is to be over almost be defeated. Gentleman of car the world says: "Inside the car too frowsty " , "I want to bask in bask in sunshine " , the system cannot identify an instruction, need to be inputted afresh. Nevertheless, gentleman of last car the world discovers the problem comes from " semantics identifying " . Then gentleman of car the world the injunction with more clear and immediate make known to lower levels: "Open scuttle " , " air conditioning key is low 1 ℃ " , " window of the car after opening " . Although not be one-time success completely, but successful rate rises 30%~50% . But even if is every 2~3 second success, the user experiences also be extreme difference, speed moves an operation far slow at the hand. Of course, cannot deny speech pilot progresses accordingly, compared with previously, ascensive range is quite distinct. Plant of car main engine tries hard again, it is better that the effect still is met.

Test project 4: Long-range control often makes mistake

Want to realize long-range control, need makes up in the mobile phone an APP, undertake all sorts of test and verify again. In principle, a car can be bound only decide number of a mobile phone. So, mobile phone number changes after, want to remember be being bound to car solution. Otherwise, car is stolen, do not know why.

Test personnel has bound decided mobile phone to have a test with plant of car main engine directly, this thinking won't appear problem, the result lets a person be surprised. Stand outside the car, gentleman of car the world gives orders " open pair of Shan Heming flute " , car does not have any reaction. The car searchs in the parking lot, often want to use this one function. Press again, mobile phone APP hints directly " an operation is half-baked " . Put out APP, enter afresh again, still be identical clew.

So give orders to examine " car condition " meeting how? Fortunately, do not have this " hinder sb " . On the mobile phone, jun Nengqing clear examines car the world oily quantity, embryo pressing and engine oil use a case, have the 100 kilometers that drive recently average oil bad news, have car even " check-up reports " , undertake scan and giving a mark to engine, transmission and batholith.

Gentleman of car the world ever was in the car academy of brand of a China, see newest speech controls a system: Can be started and shut engine, can control cruise system, accuracy rate is adjacent 100% . Regrettablly, this one newest system should arrive to just can embark to be produced in the quantity next year on the car.

Titbits: Additional kind of test

Somebody is afraid, as speech control identifying is led taller and taller, if sit in the ursine child of back row to increase chaos to you, can bring about " operate by accident " , affect safety of drive a vehicle then? To this, gentleman of car the world had a test technically. Gentleman of car the world sits in back row, bind on safety belt, with normal speech and language fast issue a statement, speech control system is done not have any reaction, this and microphone install advanced platoon to matter. However, if gentleman of car the world does not bind safety belt, discharged two seat once upon a time body of the explore intermediate, stand by speech to control a system to issue a statement, speech controls a system or can give out so answer. Change character, if rebuke,child collect makes trouble to in front of is to be able to bring a few issues.

Nevertheless, personnel of flying research and development of division hair dispatch tells car the world gentleman, itself of speech control system has powerful " error correction " function, meeting screen affects the instruction of safety of drive a vehicle in great quantities. Cite a case, between travel process, release " shut engine " instruction, the system calculates identify and understand, also won't carry out. Personnel of this research and development still expresses, before long hind, the speech inside the car controls a system to be able to identify the instruction of driver's seat only, and the statement that other place issues will not be approved.

Extraordinary verdict:

Current speech control has 1. to promote considerably before relatively, the input that need not follow the prescribed order again dictates well and truly, need not according to rigorous logic layer upon layer advance, basically can accomplish natural communication, simple for, you are OK " word saying a person " , speak place to want, other give the system will do.

2. is in " navigation " , " phone " , " recreation " wait for a function to go up, speech pilot semantics identifies ability strong, accuracy rate is high. But in " air conditioning " , " scuttle " wait for the control that involves car function to go up, estimation is to stem from safe consideration, speech pilot semantics identifies ability to lose a bit, input the instruction of accurate clarity only, the system just can be carried out.

3. we think, so far, speech is controlled or as key-press the operation comes quickly. We had been calculated, with adjusting air conditioning temperature is exemple, the hand moves operation key-press and knob, time is in 1 second less than. Identical operation, speech control needs 30 much seconds, still do not succeed certainly.

4. government is publicized, speech controls integral accuracy rate to be between 80%~90% , when demonstrating, can achieve even 100% . Use actually, check through relapsing, integral accuracy rate is insufficient 30% , still need to be perfected further. As a whole, need not regrettablly, was being used still is to be able to let a person catch mix madly abandon finally. (civil / mighty force of week of reporter of entire media of graph Guangzhou daily)