Hutchison ago
You usually see at the mall's voice assistant, it looks very simple, in fact, behind this little voice assistant, is a very complex technical support. To-back technique are: the front end of Noise Cancellation, efficient audio codec technology, the dual mode Bluetooth technology, the DMA protocol transplantation. Mobile side audio codecs, noise reduction, speech recognition technology. A wealth of entertainment and use of resources (maps, music, audio, etc.). The whole process down, is a complex project. Trying to play up the whole industry chain, it really was not able to play up a company, the industry is in need of many companies to work together.
Difficulties terminal device
- Noise reduction technology distal
front end noise reduction technology, there is typically a software algorithm, generally comprises a single microphone, dual microphones, multi-microphone noise reduction algorithm, echo cancellation algorithm, the AGC algorithm. The three of use is one of the mating process. What specific setting parameters, how fit, with properties of this product are strongly correlated. Students have this question can be more exchanges.
- Audio coding technology
Since Bluetooth spp ble bandwidth and are unable to transmit uncompressed audio, there is a relatively high do not guarantee the quality of wireless transmission. Thus, audio transmission requires a certain codec. In the DMA length recording, the audio codec used only two: SBC and OPOS. Because SBC is an older codec algorithm, regardless of the compression ratio or compression quality, and opus can not be compared, so you want to hear a clear voice assistant, it is the current mainstream opus coding. opus is an open source codec technology, whether it is the compression ratio, anti-jitter performance, or reduction of sound quality, is unique of its kind in the current audio codec.
opus The only downside is the relatively high consumption of MIPS, if you want to port it to the headset or headphones on a small platform similar to, or takes a lot of effort, a lot of chips are because they can not transplant or transplant opus opus can not be optimized , resulting in audio codec efficiency is particularly low or unavailable. This is when we fumble for a long time, done a lot of optimization algorithm only do better.
- Bluetooth support dual-mode
DMA voice piece, though spp ble and can be used. If you want to use on the Mac, only ble, because apple is not a third-party company to develop spp interface. However, if you want to use the words ble on android, android manufacturers due too scattered, each ble to do all differ greatly, it will cause a lot of compatibility problems. This is why the current mainstream solutions are android + spp apple + ble model, this requires your chip supports dual-mode Bluetooth and a ble. Currently the vast majority of headphones chip, dual-mode support of good indeed very small. This we have quite a few detours, behind only to find a suitable chip to be this thing. - Low-cost technology
to make products of it, for the money. So, the cost is the focus of the majority of equipment manufacturers first consideration. To be able to make a small degree DMA + solutions, many companies use more than two master chip. Costs will need more than $ 5, the price determines the range of commercial products can not be large. Based on this, we put everything into one chip, a chip to get a product, this greatly reduces the cost. However, also spent a lot of energy. After all, put everything into a single chip, this level of software engineers is a great test.
Technical difficulties of background app
Effective voice recognition technology
for all previous efforts of all, is to last for speech recognition. Domestic manufacturers a lot to do speech recognition, why do we choose Baidu it? There are several reasons, Baidu ai is the largest piece of domestic investment, the first investment, commercial widest range of Internet companies. Their speech recognition technology do particularly well, and also to develop a wide range of third-party manufacturers. So, choose Baidu. We also chosen to make a lot of the actual study. Later, other products made out, we prove our original choice was right.A wide range of consumer content
of speech recognition technology, if not for the support behind the content, customers just ask a few minor problems, it is estimated nothing attractive. So, app how much content is the key to the success of the product. Baidu can be said that in this case the original capital. Maps, music, audiobooks, basically the mainstream manufacturers have signed the contract. You can call QQ music, you can call the Himalayas. These resources, most manufacturers can not really mobilized.