search

NervanaSystems / deepspeech

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

same error seems to happen while trying to train and evaulating with pre trained librispeech_16epoch.prm file #49

Open Murasaki-Z opened 7 years ago

Murasaki-Z commented 7 years ago

formats: can't open input file ': No such file or directory Unable to readdecode_thread_pool exception: number of frames is negative formats: can't open input file': No such file or directory Unable to readdecode_thread_pool exception: number of frames is negative formats: can't open input file ': No such file or directory Unable to readdecode_thread_pool exception: number of frames is negative formats: can't open input file': No such file or directory Unable to readdecode_thread_pool exception: number of frames is negative [1] 13594 segmentation fault python train.py --manifest --manifest -e 2000 -z 32 -s

i installed all dependencies and checked, even then it shows the same. any help would be appreciable

tyler-nervana commented 7 years ago

Could you provide the command you are running exactly? Which dataset are you using? The error seems to indicate something is wrong with the audio files you are using.

Anandpitti commented 7 years ago

i installed the dependencies needed first. as i dint have time to train my own model, i downloaded the pretained model from Model Zoo as mentioned in the readme section(https://s3-us-west-1.amazonaws.com/nervana-modelzoo/Deep_Speech/Librispeech/librispeech_16_epochs.prm) then i downloaded the librespeech test data and created the manifest. the manifest seems to be alright with the <path to flac>,<path to transcript> format. then i tried to evaluate the model with this command, python evaluate.py --manifest val:/path/to/manifest.csv --model_file /path/to/librispeech_16_epoch.prm

then the above error showed. i tried to comment part of codes so as to find which statements cause the error, and found somewhere in this code snippet.

file_name: evaluate.py

eval_set = DataLoader(backend=be, config=eval_cfg_dict)
eval_set = data_transform(eval_set)

# Load the model
model = Model(args.model_file)

should i try changing the be (backend) to mkl or cpu.? Thanks in advance.

tyler-nervana commented 7 years ago

The issue is definitely in the aeon dataloader. Are you sure that the <path to flac> is valid and resolves to an audio file? We use sox to load the audio files inside of aeon, so if the audio file exists, sox must be encountering some issue loading it. The problem should have nothing to do with neon or the backend.