Offline speech recognition control chip module intelligent Chinese and English speech recognition control home solution details6291 platform is based on MIPS chip, runs Linux system, and supports sing
Offline speech recognition control chip module intelligent Chinese and English speech recognition control home solution details
6291 platform is based on MIPS chip, runs Linux system, and supports single wheat / dual wheat remote speech recognition. The running algorithms include: local speech recognition based on deep neural network algorithm, speech enhancement, speech noise reduction, sound source location, beamforming and local speech synthesis.
The single wheat version of the 6291 platform uses a high-sensitivity digital microphone for recording, and uses steady-state and dynamic noise filtering algorithms to dynamically adjust the recording volume. After the recording noise is reduced, the corpus data with high signal-to-noise ratio is sent to the recognition engine for recognition through speech enhancement, which ensures the high accuracy and anti noise ability of recognition at different distances (speaking from a distance).
On the basis of the single microphone version, the dual microphone version of 6291 platform adds dual microphone sound source positioning (180 degrees) and beamforming. Using the spatial filtering characteristics of the microphone, the directional pickup beam is formed by positioning the angle of waking people, and the noise outside the beam is suppressed to ensure high recording quality, and then the recording data is identified locally, It is suitable for voice control in real and complex scenes.
The 6291 platform supports both Mandarin Chinese recognition and dialect recognition. Users can customize learning and training wake-up words and control instructions. The content of learning and training is not limited to languages and speaking content. After successful learning and training, the speech brain supports not only the existing Mandarin recognition, but also the dialect recognition of learning and training.
The 6291 platform has launched a store version, which is suitable for voice recognition and voice control in noisy environment. The products with integrated voice brain 6291 can also be demonstrated and operated in stores.
Technical features:
1. Based on the new DNN deep neural network recognition algorithm, it supports local recognition in the off-line state, and the recognition instructions can reach 10000.
2. Based on the secondary development of voice brain, there is no need for developers to write code for wake-up words, voice recognition content and voice brain parameter configuration. The threshold of secondary development is lower, and the voice brain can be quickly integrated into the product.
3. Far field recognition, long-distance pickup, recognition distance of more than 5m.
4. Recognition rate up to 98%
