Edge Container for Speech Recognition

A Low-Latency Streaming On-Device Automatic Speech Recognition System Using a CNN Acoustic Model on FPGA and a Language Model on Smartphone

This paper presents a low-latency streaming on-device automatic speech recognition system for inference. It consists of a hardware acoustic model implemented in a field-programmable gate array, coupled with a software language model running on a smartphone. The smartphone works as the master of the automatic speech recognition system and runs a three-gram language model on the acoustic model output to increase accuracy. The smartphone calculates and sends the Mel-spectrogram of an audio stream with 80 ms unit input from the built-in microphone of the smartphone to the field-programmable gate array every 80 ms. After ~35 ms, the field-programmable gate array sends the calculated word-piece probability to the smartphone, which runs the language model and generates the text output on the smartphone display. The worst-case latency from the audio-stream start time to the text output time was measured as 125.5 ms. The real-time factor is 0.57. The hardware acoustic model is derived from a time-depth-separable convolutional neural network model by reducing the number of weights from 115 M to 9.3 M to decrease the number of multiply-and-accumulate operations by two orders of magnitude. Additionally, the unit input length is reduced from 1000 ms to 80 ms, and to minimize the latency, no future data are used. The hardware acoustic model uses an instruction-based architecture that supports any sequence of convolutional neural network, residual network, layer normalization, and rectified linear unit operations. For the LibriSpeech test-clean dataset, the word error rate of the hardware acoustic model was 13.2% and for the language model, it was 9.1%. These numbers were degraded by 3.4% and 3.2% from the original convolutional neural network software model due to the reduced number of weights and the lowering of the floating-point precision from 32 to 16 bit. The automatic speech recognition system has been demonstrated successfully in real application scenarios.

Influence of Highly Inflected Word Forms and Acoustic Background on the Robustness of Automatic Speech Recognition for Human–Computer Interaction

Automatic speech recognition is essential for establishing natural communication with a human–computer interface. Speech recognition accuracy strongly depends on the complexity of language. Highly inflected word forms are a type of unit present in some languages. The acoustic background presents an additional important degradation factor influencing speech recognition accuracy. While the acoustic background has been studied extensively, the highly inflected word forms and their combined influence still present a major research challenge. Thus, a novel type of analysis is proposed, where a dedicated speech database comprised solely of highly inflected word forms is constructed and used for tests. Dedicated test sets with various acoustic backgrounds were generated and evaluated with the Slovenian UMB BN speech recognition system. The baseline word accuracy of 93.88% and 98.53% was reduced to as low as 23.58% and 15.14% for the various acoustic backgrounds. The analysis shows that the word accuracy degradation depends on and changes with the acoustic background type and level. The highly inflected word forms’ test sets without background decreased word accuracy from 93.3% to only 63.3% in the worst case. The impact of highly inflected word forms on speech recognition accuracy was reduced with the increased levels of acoustic background and was, in these cases, similar to the non-highly inflected test sets. The results indicate that alternative methods in constructing speech databases, particularly for low-resourced Slovenian language, could be beneficial.