AMI

This is an ASR recipe for the AMI corpus. AMI provides recordings from the speaker's headset and lapel microphones, and also 2 array microphones containing 8 channels each. We pool data in the following 4 ways and train a single model on the pooled data:

(i) individual headset microphone (IHM) (ii) IHM with simulated reverb (iii) Single distant microphone (SDM) (iv) GSS-enhanced array microphones

Speed perturbation and MUSAN noise augmentation are additionally performed on the pooled data.

Performance Record

pruned_transducer_stateless7

The following are decoded using modified_beam_search:

Evaluation set	dev WER	test WER
IHM	18.92	17.40
SDM	31.25	32.21
MDM (GSS-enhanced)	21.67	22.43

See the recipe for details.