Modelpublic

desh2608/icefall-asr-alimeeting-pruned-transducer-stateless7

sourceHugging Faceapache-2.0updated 4y agoView on Hugging Face

0likes

Model Card

AliMeeting

This is an ASR recipe for the AliMeeting corpus. AliMeeting provides recordings from the speaker's headset microphones and an 8-channel microphone array. 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

prunedtransducerstateless7

The following are decoded using modified_beam_search:

Evaluation set	eval CER	test CER
IHM	9.58	11.53
SDM	23.37	25.85
MDM (GSS-enhanced)	11.82	14.22

See the recipe for details.