some notes about TIMIT (#603)

examples/CTC-TIMIT/README.md

@@ -5,10 +5,21 @@
 + python binding for `lmdb`
   + `pip install --user lmdb`
 + `bob.ap` package for MFCC extraction
-	+ install blitz and openblas as dependencies of bob.ap
-	+ `pip install --user bob.ap`
+  + install [blitz](https://github.com/blitzpp/blitz) and openblas as dependencies of bob.ap
+  + `pip install --user bob.extension bob.blitz bob.core bob.sp bob.ap`
 
 ### Prepare Data:
+We assume the following file structure:
+```
+TRAIN/
+  DR1/
+    FCJF0/
+      *.WAV     # NIST WAV file
+      *.TXT
+      *.PHN
+  ...
+```
+
 Convert NIST wav format to RIFF wav format:
 ```
 cd /PATH/TO/TIMIT

examples/CTC-TIMIT/create-lmdb.py

@@ -8,12 +8,12 @@ import string
 import numpy as np
 import argparse
 
+import bob.ap
 from tensorpack.dataflow import dftools, DataFlow, LMDBDataPoint
 from tensorpack.utils.argtools import memoized
 from tensorpack.utils.stats import OnlineMoments
 from tensorpack.utils import serialize, fs, logger
 from tensorpack.utils.utils import get_tqdm
-import bob.ap
 
 CHARSET = set(string.ascii_lowercase + ' ')
 PHONEME_LIST = [
@@ -85,7 +85,7 @@ class RawTIMIT(DataFlow):
         self.filelists = [k for k in fs.recursive_walk(self.dirname)
                           if k.endswith('.wav')]
         logger.info("Found {} wav files ...".format(len(self.filelists)))
-        assert len(self.filelists), self.filelists
+        assert len(self.filelists), "Found no '.wav' files!"
         assert label in ['phoneme', 'letter'], label
         self.label = label
 

examples/CTC-TIMIT/train-timit.py

@@ -79,7 +79,7 @@ class Model(ModelDesc):
 
 def get_data(path, isTrain, stat_file):
     ds = LMDBDataPoint(path, shuffle=isTrain)
-    mean, std = serialize.loads(open(stat_file).read())
+    mean, std = serialize.loads(open(stat_file, 'rb').read())
     ds = MapDataComponent(ds, lambda x: (x - mean) / std)
     ds = TIMITBatch(ds, BATCH)
     if isTrain: