Stateful Conformer with Cache-based Inference for Streaming Automatic Speech Recognition

Benchmark Model Rank Results
speech-recognition-on-ami-cleanednvidia/nemotron-speech-streaming-en-0.6b#13Word Error Rate (WER): 8.84
speech-recognition-on-ami-cleanednvidia/nemotron-3.5-asr-streaming-0.6b#36Word Error Rate (WER): 13.42
speech-recognition-on-earnings-22nvidia/nemotron-speech-streaming-en-0.6b#31Word Error Rate (WER): 12.02
speech-recognition-on-earnings-22nvidia/nemotron-3.5-asr-streaming-0.6b#49Word Error Rate (WER): 14.95
speech-recognition-on-gigaspeech-cleanednvidia/nemotron-speech-streaming-en-0.6b#10Word Error Rate (WER): 7.91
speech-recognition-on-gigaspeech-cleanednvidia/nemotron-3.5-asr-streaming-0.6b#46Word Error Rate (WER): 9.86
speech-recognition-on-librispeech-test-cleannvidia/nemotron-speech-streaming-en-0.6b#61Word Error Rate (WER): 1.88
speech-recognition-on-librispeech-test-cleannvidia/nemotron-3.5-asr-streaming-0.6b#99Word Error Rate (WER): 2.83
speech-recognition-on-librispeech-test-othernvidia/nemotron-speech-streaming-en-0.6b#67Word Error Rate (WER): 4.34
speech-recognition-on-librispeech-test-othernvidia/nemotron-3.5-asr-streaming-0.6b#98Word Error Rate (WER): 6.78
speech-recognition-on-spgispeechnvidia/nemotron-speech-streaming-en-0.6b#28Word Error Rate (WER): 2.69
speech-recognition-on-spgispeechnvidia/nemotron-3.5-asr-streaming-0.6b#39Word Error Rate (WER): 3.27
speech-recognition-on-voxpopuli-cleanednvidia/nemotron-speech-streaming-en-0.6b#5Word Error Rate (WER): 2.4
speech-recognition-on-voxpopuli-cleanednvidia/nemotron-3.5-asr-streaming-0.6b#40Word Error Rate (WER): 4.24