SeanNaren / deepspeech.torch

Speech Recognition using DeepSpeech2 network and the CTC activation function.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

custom dataset formatting issue

joseph-osborne opened this issue · comments

So I'm trying to replace what is in AN4 with my own data. I'm not sure if I'm formatting it properly because when I run MakeLMDB.lua I get the following error. I don't know it means "attempt to index field 'file' (a nil value)" Does that have to do with if I format the data properly. When I run FormatAN4.lua it works and I don't get any warnings or errors so I don't know what I am doing wrong. Thank you guys for any help you can give.

th MakeLMDB.lua -rootPath prepare_datasets/an4_dataset -lmdbPath prepare_datasets/an4_lmdb -wind
owSize 0.02 -stride 0.01 -sampleRate 16000
Torch 7.0 Copyright (C) 2001-2011 Idiap, NEC Labs, NYU
Torch 7.0 Copyright (C) 2001-2011 Idiap, NEC Labs, NYU
Torch 7.0 Copyright (C) 2001-2011 Idiap, NEC Labs, NYU
Torch 7.0 Copyright (C) 2001-2011 Idiap, NEC Labs, NYU
Torch 7.0 Copyright (C) 2001-2011 Idiap, NEC Labs, NYU
Torch 7.0 Copyright (C) 2001-2011 Idiap, NEC Labs, NYU
Torch 7.0 Copyright (C) 2001-2011 Idiap, NEC Labs, NYU
Torch 7.0 Copyright (C) 2001-2011 Idiap, NEC Labs, NYU
Retrieving sizes for sorting...
[============================== 10/10 =============================>] Tot: 21ms | Step: 2ms
Sorting...
Creating LMDB dataset to: prepare_datasets/an4_lmdb/train
[============================== 10/10 =============================>] Tot: 22ms | Step: 2ms
Retrieving sizes for sorting...
luajit: MakeLMDB.lua:61: attempt to index field 'file' (a nil value)
stack traceback:
MakeLMDB.lua:61: in function 'code'
MakeLMDB.lua:168: in function 'f'
(command line):4: in main chunk
[C]: at 0x00405d50
luajit: MakeLMDB.lua:61: attempt to index field 'file' (a nil value)
stack traceback:
MakeLMDB.lua:61: in function 'code'
MakeLMDB.lua:168: in function 'f'
(command line):4: in main chunk
[C]: at 0x00405d50
luajit: MakeLMDB.lua:61: attempt to index field 'file' (a nil value)
stack traceback:
MakeLMDB.lua:61: in function 'code'
MakeLMDB.lua:168: in function 'f'
(command line):4: in main chunk
[C]: at 0x00405d50
[============================== 5/5 ===============================>] Tot: 3ms | Step: 0ms
Sorting...
Creating LMDB dataset to: prepare_datasets/an4_lmdb/test
MakeLMDB.lua:125: attempt to index local 'vec' (a nil value)
<parallel#000> closing session

Modify FormatAN4.lua, your format is error. Make sure root directory
FormatAN4 use .sph not .wav, edit source-code FormatAN4. Create a folder etc(train/test) and wav (wav_train/wav_test).
Edit path to etc and wav