search

NervanaSystems / deepspeech

DeepSpeech neon implementation
Apache License 2.0
222 stars 69 forks source link

phoneme classification #18

Closed pankaj2701 closed 7 years ago

pankaj2701 commented 7 years ago

I am trying to use deepspeech for phoneme classification. I am not able to figure out how to specify the phoneme set. For character recognition the alphabet is a set of unique characters but not in case of phonemes. I tried to specify the symbols for phonemes as a list but that gives an error

aeon.dataloader.LoaderRuntimeError: error in loader: type must be string, but is array

Any suggestion?

rkimballn1 commented 7 years ago

That error message is from the parsing of config data. Can I see how you are configuring the dataloader?

pankaj2701 commented 7 years ago

I have done the following change in train.py. I have specified the alphabet as alphabet = ["SIL","AA","AE","AH","AO","AW","AY","B","CH","D","DH","EH", "ER","EY","F","G","HH","IH","IY","JH","K","L","M","N","NG", "OW","OY","P","R","S","SH","T","TH","UH","UW","V","W","Z","ZH"]

From the error message I guess the alphabet should be specified as a string of unique characters. But then how to specify phoneme set as a string, as it contains repetition of characters.

tyler-nervana commented 7 years ago

It looks like you have found a workaround for this issue (#32). That should work fine as a stopgap until unicode support is added to aeon, the dataloader we use in this repo.