search

dtjchen / spoken-command-processor

Neural network-based speech transcription model. Built with Keras (Python) and trained with TIMIT.
19 stars 6 forks source link

Train the "phoneme strings --> words" BRNN #2

Closed mamigot closed 8 years ago

mamigot commented 8 years ago

Architecture of the network + hyperparameters.

First BRNN had one-to-one inputs and outputs (one label per 20ms-MFCC vector)... but this has many inputs (phonemes) for each word.