Training and Inference Efficiency of Encoder-Decoder Speech Models

Open paper
Benchmark Model Rank Results
speech-recognition-on-aminvidia/canary-1b-flash#19Word Error Rate (WER): 12.11
speech-recognition-on-aminvidia/canary-180m-flash#23Word Error Rate (WER): 13.66
speech-recognition-on-ami-cleanednvidia/canary-1b-flash#19Word Error Rate (WER): 10.58
speech-recognition-on-ami-cleanednvidia/canary-180m-flash#25Word Error Rate (WER): 12.1
speech-recognition-on-earnings-22nvidia/canary-1b-flash#32Word Error Rate (WER): 12.03
speech-recognition-on-earnings-22nvidia/canary-180m-flash#36Word Error Rate (WER): 12.5
speech-recognition-on-gigaspeech-cleanednvidia/canary-1b-flash#16Word Error Rate (WER): 8.19
speech-recognition-on-gigaspeech-cleanednvidia/canary-180m-flash#36Word Error Rate (WER): 8.87
speech-recognition-on-gigaspeech-testnvidia/canary-1b-flash#16Word Error Rate (WER): 8.26
speech-recognition-on-gigaspeech-testnvidia/canary-180m-flash#35Word Error Rate (WER): 8.96
speech-recognition-on-librispeech-test-cleannvidia/canary-1b-flash#11Word Error Rate (WER): 1.2
speech-recognition-on-librispeech-test-cleannvidia/canary-180m-flash#36Word Error Rate (WER): 1.52
speech-recognition-on-librispeech-test-othernvidia/canary-1b-flash#11Word Error Rate (WER): 2.52
speech-recognition-on-librispeech-test-othernvidia/canary-180m-flash#38Word Error Rate (WER): 3.41
speech-recognition-on-spgispeechnvidia/canary-1b-flash#2Word Error Rate (WER): 1.75
speech-recognition-on-spgispeechnvidia/canary-180m-flash#11Word Error Rate (WER): 2.04
speech-recognition-on-voxpopulinvidia/canary-1b-flash#3Word Error Rate (WER): 5.38
speech-recognition-on-voxpopulinvidia/canary-180m-flash#20Word Error Rate (WER): 6.11
speech-recognition-on-voxpopuli-cleanednvidia/canary-180m-flash#23Word Error Rate (WER): 3.57
speech-recognition-on-voxpopuli-cleanednvidia/canary-1b-flash#38Word Error Rate (WER): 4.17

Rank counts only results with a code link.