Spaces:

nvidia

/

canary-1b

like 146

Running on T4

I've tried the model here on the space but the transcribed results in German were really not good (I've seen better results from Whisper Medium...). Additionally, in ALL of the 4-5 German samples I've uploaded (that were between 1-3 min long), I got hallucinated text passages (mostly whole sentences) that were either added to the end OR replaced the actual end of the audio. Anyone with similiar experience for German audio? Guess I should switch back to Universal-1 again xD