A vehicle-mounted speech recognition method and system based on a pulse neural network
By constructing a spiking neural network based on reparameterized spiking convolution self-attention, the accuracy and robustness issues of in-vehicle speech recognition in complex acoustic environments are solved, realizing an efficient in-vehicle speech recognition method suitable for end-to-end speech recognition in vehicles.
Patent Information
- Application Number
- CN202610570248.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-04-28
- Publication Date
- 2026-07-17
AI Technical Summary
Existing in-vehicle speech recognition technologies lack accuracy and robustness in complex acoustic environments, especially in in-vehicle scenarios where noise is complex and reverberation is significant. Existing spiking neural networks consume a lot of memory during training and lack targeted adaptation.
We employ a spiking neural network based on reparameterized spiking convolutional self-attention, combined with a multi-level spiking convolutional encoder, a Transformer decoder, and a CTC-assisted classification branch. Through in-vehicle voice data preprocessing, feature extraction, and absolute position encoding, we construct a speech recognition method for in-vehicle scenarios, reducing memory usage during training and enhancing the ability to extract local information.
It improves the accuracy and environmental robustness of in-vehicle speech recognition, reduces training costs and computational load, and is suitable for efficient in-vehicle applications.
Smart Images
Figure CN122417013A_ABST