Fast Conformer with Linearly Scalable Attention for Efficient Speech Recognition

Benchmark Model Rank Results
speech-recognition-on-aminvidia/parakeet-ctc-1.1b#28Word Error Rate (WER): 14.76
speech-recognition-on-aminvidia/parakeet-ctc-0.6b#38Word Error Rate (WER): 15.56
speech-recognition-on-aminvidia/parakeet-rnnt-1.1b#43Word Error Rate (WER): 15.96
speech-recognition-on-aminvidia/parakeet-rnnt-0.6b#45Word Error Rate (WER): 16.47
speech-recognition-on-aminvidia/stt_en_fastconformer_ctc_large#46Word Error Rate (WER): 17.72
speech-recognition-on-aminvidia/stt_en_fastconformer_transducer_large#48Word Error Rate (WER): 18.01
speech-recognition-on-ami-cleanednvidia/parakeet-ctc-1.1b#26Word Error Rate (WER): 12.24
speech-recognition-on-ami-cleanednvidia/parakeet-ctc-0.6b#31Word Error Rate (WER): 13.1
speech-recognition-on-ami-cleanednvidia/parakeet-rnnt-1.1b#45Word Error Rate (WER): 14.02
speech-recognition-on-ami-cleanednvidia/parakeet-rnnt-0.6b#48Word Error Rate (WER): 14.34
speech-recognition-on-ami-cleanednvidia/stt_en_fastconformer_ctc_large#50Word Error Rate (WER): 16.12
speech-recognition-on-ami-cleanednvidia/stt_en_fastconformer_transducer_large#51Word Error Rate (WER): 16.35
speech-recognition-on-earnings-22nvidia/parakeet-rnnt-1.1b#38Word Error Rate (WER): 12.94
speech-recognition-on-earnings-22nvidia/parakeet-ctc-1.1b#39Word Error Rate (WER): 12.98
speech-recognition-on-earnings-22nvidia/parakeet-ctc-0.6b#45Word Error Rate (WER): 13.47
speech-recognition-on-earnings-22nvidia/parakeet-rnnt-0.6b#46Word Error Rate (WER): 13.58
speech-recognition-on-earnings-22nvidia/stt_en_fastconformer_transducer_large#53Word Error Rate (WER): 17.71
speech-recognition-on-earnings-22nvidia/stt_en_fastconformer_ctc_large#54Word Error Rate (WER): 17.75
speech-recognition-on-gigaspeech-cleanednvidia/parakeet-rnnt-1.1b#18Word Error Rate (WER): 8.22
speech-recognition-on-gigaspeech-cleanednvidia/parakeet-rnnt-0.6b#23Word Error Rate (WER): 8.45
speech-recognition-on-gigaspeech-cleanednvidia/parakeet-ctc-1.1b#33Word Error Rate (WER): 8.74
speech-recognition-on-gigaspeech-cleanednvidia/parakeet-ctc-0.6b#37Word Error Rate (WER): 8.9
speech-recognition-on-gigaspeech-cleanednvidia/stt_en_fastconformer_transducer_large#51Word Error Rate (WER): 10.67
speech-recognition-on-gigaspeech-cleanednvidia/stt_en_fastconformer_ctc_large#52Word Error Rate (WER): 10.89
speech-recognition-on-gigaspeech-testnvidia/parakeet-rnnt-1.1b#18Word Error Rate (WER): 8.31
speech-recognition-on-gigaspeech-testnvidia/parakeet-rnnt-0.6b#24Word Error Rate (WER): 8.55
speech-recognition-on-gigaspeech-testnvidia/parakeet-ctc-1.1b#32Word Error Rate (WER): 8.88
speech-recognition-on-gigaspeech-testnvidia/parakeet-ctc-0.6b#36Word Error Rate (WER): 9.03
speech-recognition-on-gigaspeech-testnvidia/stt_en_fastconformer_transducer_large#53Word Error Rate (WER): 10.76
speech-recognition-on-gigaspeech-testnvidia/stt_en_fastconformer_ctc_large#55Word Error Rate (WER): 10.98
speech-recognition-on-librispeech-test-cleannvidia/parakeet-rnnt-1.1b#9Word Error Rate (WER): 1.15
speech-recognition-on-librispeech-test-cleannvidia/parakeet-rnnt-0.6b#16Word Error Rate (WER): 1.25
speech-recognition-on-librispeech-test-cleannvidia/stt_en_fastconformer_transducer_large#19Word Error Rate (WER): 1.31
speech-recognition-on-librispeech-test-cleannvidia/parakeet-ctc-1.1b#35Word Error Rate (WER): 1.52
speech-recognition-on-librispeech-test-cleannvidia/parakeet-ctc-0.6b#41Word Error Rate (WER): 1.57
speech-recognition-on-librispeech-test-cleannvidia/stt_en_fastconformer_ctc_large#42Word Error Rate (WER): 1.58
speech-recognition-on-librispeech-test-cleanparakeet-rnnt-1.1bWord Error Rate (WER): 1.46
speech-recognition-on-librispeech-test-othernvidia/parakeet-rnnt-1.1b#2Word Error Rate (WER): 2.1
speech-recognition-on-librispeech-test-othernvidia/parakeet-rnnt-0.6b#17Word Error Rate (WER): 2.64
speech-recognition-on-librispeech-test-othernvidia/parakeet-ctc-1.1b#31Word Error Rate (WER): 3.23
speech-recognition-on-librispeech-test-othernvidia/stt_en_fastconformer_transducer_large#35Word Error Rate (WER): 3.39
speech-recognition-on-librispeech-test-othernvidia/parakeet-ctc-0.6b#37Word Error Rate (WER): 3.41
speech-recognition-on-librispeech-test-othernvidia/stt_en_fastconformer_ctc_large#45Word Error Rate (WER): 3.64
speech-recognition-on-spgispeechnvidia/parakeet-rnnt-1.1b#20Word Error Rate (WER): 2.56
speech-recognition-on-spgispeechnvidia/parakeet-rnnt-0.6b#35Word Error Rate (WER): 2.88
speech-recognition-on-spgispeechnvidia/parakeet-ctc-0.6b#40Word Error Rate (WER): 3.47
speech-recognition-on-spgispeechnvidia/parakeet-ctc-1.1b#44Word Error Rate (WER): 3.53
speech-recognition-on-spgispeechnvidia/stt_en_fastconformer_transducer_large#49Word Error Rate (WER): 4.03
speech-recognition-on-spgispeechnvidia/stt_en_fastconformer_ctc_large#50Word Error Rate (WER): 4.25
speech-recognition-on-spgispeechparakeet-rnnt-1.1bWord Error Rate (WER): 3.11
speech-recognition-on-tedliumparakeet-rnnt-1.1bWord Error Rate (WER): 3.92
speech-recognition-on-voxpopulinvidia/parakeet-rnnt-1.1b#1Word Error Rate (WER): 5.12
speech-recognition-on-voxpopulinvidia/parakeet-rnnt-0.6b#14Word Error Rate (WER): 5.78
speech-recognition-on-voxpopulinvidia/stt_en_fastconformer_transducer_large#19Word Error Rate (WER): 6.04
speech-recognition-on-voxpopulinvidia/stt_en_fastconformer_ctc_large#21Word Error Rate (WER): 6.14
speech-recognition-on-voxpopulinvidia/parakeet-ctc-1.1b#22Word Error Rate (WER): 6.29
speech-recognition-on-voxpopulinvidia/parakeet-ctc-0.6b#30Word Error Rate (WER): 6.77
speech-recognition-on-voxpopuli-cleanednvidia/parakeet-ctc-0.6b#14Word Error Rate (WER): 3.19
speech-recognition-on-voxpopuli-cleanednvidia/parakeet-ctc-1.1b#16Word Error Rate (WER): 3.24
speech-recognition-on-voxpopuli-cleanednvidia/parakeet-rnnt-0.6b#20Word Error Rate (WER): 3.41
speech-recognition-on-voxpopuli-cleanednvidia/parakeet-rnnt-1.1b#25Word Error Rate (WER): 3.64
speech-recognition-on-voxpopuli-cleanednvidia/stt_en_fastconformer_ctc_large#30Word Error Rate (WER): 3.77
speech-recognition-on-voxpopuli-cleanednvidia/stt_en_fastconformer_transducer_large#44Word Error Rate (WER): 4.31