Finstreder: Simple and fast Spoken Language Understanding with Finite State Transducers using modern Speech-to-Text models

Open paper
Benchmark Model Rank Results
intent-classification-on-slurpFinstreder (Conformer)#3Accuracy (%): 53.11
intent-classification-on-slurpFinstreder (Quartznet)#4Accuracy (%): 43.15
slot-filling-on-slurpFinstreder (Conformer)#3F1: 0.395
slot-filling-on-slurpFinstreder (Quartznet)#4F1: 0.313
spoken-language-understanding-on-fluentFinstreder (Conformer + AMT, character-based)#1Accuracy (%): 99.8
spoken-language-understanding-on-fluentFinstreder (Quartznet + AMT)#3Accuracy (%): 99.7
spoken-language-understanding-on-fluentFinstreder (Conformer)#5Accuracy (%): 99.5
spoken-language-understanding-on-fluentFinstreder (Conformer, character-based)#6Accuracy (%): 99.4
spoken-language-understanding-on-fluentFinstreder (Quartznet)#7Accuracy (%): 99.2
spoken-language-understanding-on-fluentAmazon Alexa#10Accuracy (%): 98.7
spoken-language-understanding-on-snips-smartlightsFinstreder (Conformer, character-based)#5Accuracy (%): 89.0
spoken-language-understanding-on-snips-smartlightsFinstreder (Conformer)#6Accuracy (%): 88.0
spoken-language-understanding-on-snips-smartlightsFinstreder (Quartznet)#9Accuracy (%): 84.8
spoken-language-understanding-on-snips-smartspeakerFinstreder (Conformer, character-based)#2Accuracy-EN (%): 87.9Accuracy-FR (%): 86.5
spoken-language-understanding-on-snips-smartspeakerFinstreder (Conformer)#5Accuracy-EN (%): 80.4Accuracy-FR (%): 78.3
spoken-language-understanding-on-snips-smartspeakerFinstreder (Quartznet)#7Accuracy-EN (%): 77.6Accuracy-FR (%): 77.8
spoken-language-understanding-on-timers-and-suchFinstreder (Conformer)#1Accuracy (%): 95.4
spoken-language-understanding-on-timers-and-suchFinstreder (Quartznet)#5Accuracy (%): 90.0

Rank counts only results with a code link.