Building Accurate Low Latency ASR for Streaming Voice Search
This work addresses the need for accurate, low-latency ASR in real-time voice search applications, particularly for Hinglish, representing an incremental improvement over existing streaming methods.
The paper tackled the problem of low accuracy in streaming automatic speech recognition (ASR) for voice search by developing modified LSTM, attention, and CTC models, achieving a word error rate of 3.69% without end-of-speech detection and reducing latency by approximately 1300 ms (46.64%) compared to a baseline.
Automatic Speech Recognition (ASR) plays a crucial role in voice-based applications. For applications requiring real-time feedback like Voice Search, streaming capability becomes vital. While LSTM/RNN and CTC based ASR systems are commonly employed for low-latency streaming applications, they often exhibit lower accuracy compared to state-of-the-art models due to a lack of future audio frames. In this work, we focus on developing accurate LSTM, attention, and CTC based streaming ASR models for large-scale Hinglish (a blend of Hindi and English) Voice Search. We investigate various modifications in vanilla LSTM training which enhance the system's accuracy while preserving its streaming capabilities. We also address the critical requirement of end-of-speech (EOS) detection in streaming applications. We present a simple training and inference strategy for end-to-end CTC models that enables joint ASR and EOS detection. The evaluation of our model on Flipkart's Voice Search, which handles substantial traffic of approximately 6 million queries per day, demonstrates significant performance gains over the vanilla LSTM-CTC model. Our model achieves a word error rate (WER) of 3.69% without EOS and 4.78% with EOS while also reducing the search latency by approximately ~1300 ms (equivalent to 46.64% reduction) when compared to an independent voice activity detection (VAD) model.