Core Algorithm


VoiceAI's original voice dynamic hybrid neural network, a voice neural network technology that dynamically mixes traditional speech processing algorithms, greatly enhances the processing of speech signals in different application scenarios, especially in the areas of noise reduction, dereverberation, and echo cancellation. It has superior robustness over traditional algorithms.


VoiceAI's pioneering speech deep processing neural network technology, a deep optimized speech neural network technology based on DNN, RNN and CNN, makes the voiceprint recognition in the noise environment far superior to the traditional method, and only requires a single model. It can meet the application in different scenarios, and the recognition rate is higher than 99.5%.

Technology Engine

  • placeholder
    VoiceDNA Voiceprint 1:1/1:N comparison

    Voiceprint 1:1 verification engine(determine if it is the same person) and 1:N big data cloud search engine(Return speaker recognition result and approximation ranking in cloud database containing massive personnel sample library)

  • placeholder
    VoiceKEY Content and identity dual verification

    The dual verification engine of speaker identity (voice recognition) and specific speech content (voice recognition) further reduces the risk of counterfeiting and supports the operation of the cloud and local terminal devices.

  • placeholder
    VoicePLUS Intelligent Voice Command Recognition

    Voice command recognition engine, including voice wakeup + voiceprint recognition + command word recognition, can run in CPU and memory limited mobile or embedded systems

  • placeholder
    VoiceALL Real-time Speech Recognition

    The speech content is automatically recognized to text in real-time(Speech to text), which supports real-time continuous voice input and recognition input for daily speech at 16k and 8k sampling rate. The industry-specific vocabulary can be customized according to customer's needs to improve recognition accuracy.

  • placeholder
    VoiceSCAN Living detection algorithm

    Prevent recording and other attack techniques (such as voice change, voice synthesis, etc.), prevent success rate of more than 90%, effectively reduce the risk of identity counterfeiting and misappropriation