search

NervanaSystems / deepspeech

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

no handler for detected file type `flac' #27

Closed striki70 closed 7 years ago

striki70 commented 7 years ago

I am trying to make this work. Unfortunately, with evaluate.py and train.py I'm receiving the following errors:

2017-05-08 16:51:52,109 - neon.util.argparser - WARNING - No schedule given for model serialization, using default 1 2017-05-08 16:51:52,505 - neon.backends - WARNING - deterministic_update and deterministic args are deprecated in favor of specifying random seed formats: no handler for detected file type flac' formats: no handler for detected file typeflac' formats: no handler for detected file type flac' formats: no handler for detected file typeflac'formats: no handler for detected file type flac'formats: no handler for detected file typeflac'

formats: no handler for detected file type flac' Unable to readUnable to readUnable to readUnable to readUnable to readUnable to readdecode_thread_pool exception: decode_thread_pool exception: number of frames is negativenumber of frames is negative decode_thread_pool exception: decode_thread_pool exception: number of frames is negative number of frames is negative decode_thread_pool exception: number of frames is negative decode_thread_pool exception: number of frames is negative formats: no handler for detected file typeflac' Unable to readdecode_thread_pool exception: number of frames is negative Unable to readdecode_thread_pool exception: number of frames is negative Segmentation fault (core dumped)

Any hint how to deal with flac files?

tyler-nervana commented 7 years ago

Hmm, we definitely expect flac decoding to work properly (Librispeech uses flac encoding, after all). Could you make sure you have all of the required dependencies for aeon and try reinstalling it (https://aeon.nervanasys.com/index.html/getting_started.html)?

striki70 commented 7 years ago

I installed required dependencies, and still, again the same issue. Is there a way that I use wav files as input for training and evaluation?

tyler-nervana commented 7 years ago

Yes, you can definitely just convert the files to wav for training and evaluation. The sox command line utility should make pretty quick work of that. Then just update your manifest file to reference the wav files and you should be good to go.