Recurrent Neural NetworkRNN
Definition
A neural network architecture that processes sequential data by maintaining a hidden state across time steps.
Recurrent neural networks were a foundational architecture for speech recognition before transformers. RNNs process sequences one element at a time, maintaining a hidden state vector that carries information from previous time steps. Variants like LSTMs and GRUs introduced gating mechanisms to better capture long-range dependencies.
While transformers have largely replaced RNNs in state-of-the-art ASR, RNN-based models remain relevant in production systems where streaming (real-time) output is critical. The RNN-Transducer architecture, used in many on-device speech systems, combines an RNN-based audio encoder with a prediction network for low-latency transcription.