Speech Recognition: HCLG&DNN with computing in memory architecture？
Beyond Moreer
1 reply
Adopt the NN acoustic models and HCLG language models . With the hybrid structure, on the one hand , this model can fully use this open source for training and, on the other hand, contribute to the hybrid structure, customers can easily achieve the customized keywords recognition. Meanwhile , computing in memory architecture will improve operation efficiency 100 times.
Replies
Beyond Moreer@lucas_wang6
- With computing in memory structure, the memory wall has been broken, computing and memory exists in one unit, improving AI computing and lower power consumption. WTM2101 is the CIM chip which has been applied to intelligent voice and intelligent health. Speech recognition is the significant function in WTM2101 and has been widely used in smart wearable devices.
Share