Granite-speech: open-source speech-aware LLMs with strong English ASR capabilities

Benchmark Model Rank Results
speech-recognition-on-amiibm-granite/granite-speech-3.3-8b#6Word Error Rate (WER): 8.43
speech-recognition-on-amiibm-granite/granite-speech-3.3-2b#7Word Error Rate (WER): 8.53
speech-recognition-on-ami-cleanedibm-granite/granite-speech-3.3-8b#6Word Error Rate (WER): 7.7
speech-recognition-on-ami-cleanedibm-granite/granite-speech-3.3-2b#7Word Error Rate (WER): 7.8
speech-recognition-on-earnings-22ibm-granite/granite-speech-3.3-8b#8Word Error Rate (WER): 9.07
speech-recognition-on-earnings-22ibm-granite/granite-speech-3.3-2b#10Word Error Rate (WER): 9.73
speech-recognition-on-gigaspeech-cleanedibm-granite/granite-speech-3.3-8b#27Word Error Rate (WER): 8.58
speech-recognition-on-gigaspeech-cleanedibm-granite/granite-speech-3.3-2b#39Word Error Rate (WER): 9.03
speech-recognition-on-gigaspeech-testibm-granite/granite-speech-3.3-8b#27Word Error Rate (WER): 8.65
speech-recognition-on-gigaspeech-testibm-granite/granite-speech-3.3-2b#39Word Error Rate (WER): 9.1
speech-recognition-on-librispeech-test-cleanibm-granite/granite-speech-3.3-8b#8Word Error Rate (WER): 1.11
speech-recognition-on-librispeech-test-cleanibm-granite/granite-speech-3.3-2b#12Word Error Rate (WER): 1.21
speech-recognition-on-librispeech-test-otheribm-granite/granite-speech-3.3-8b#12Word Error Rate (WER): 2.52
speech-recognition-on-librispeech-test-otheribm-granite/granite-speech-3.3-2b#20Word Error Rate (WER): 2.84
speech-recognition-on-spgispeechibm-granite/granite-speech-3.3-2b#43Word Error Rate (WER): 3.5
speech-recognition-on-spgispeechibm-granite/granite-speech-3.3-8b#45Word Error Rate (WER): 3.54
speech-recognition-on-voxpopuliibm-granite/granite-speech-3.3-8b#6Word Error Rate (WER): 5.44
speech-recognition-on-voxpopuliibm-granite/granite-speech-3.3-2b#12Word Error Rate (WER): 5.64
speech-recognition-on-voxpopuli-cleanedibm-granite/granite-speech-3.3-2b#50Word Error Rate (WER): 4.52
speech-recognition-on-voxpopuli-cleanedibm-granite/granite-speech-3.3-8b#52Word Error Rate (WER): 4.54