diff --git a/egs/mini_librispeech/s5/local/grammar/extend_vocab_demo.sh b/egs/mini_librispeech/s5/local/grammar/extend_vocab_demo.sh
index 382f9f4f6c6..1ec4a0d575b 100755
--- a/egs/mini_librispeech/s5/local/grammar/extend_vocab_demo.sh
+++ b/egs/mini_librispeech/s5/local/grammar/extend_vocab_demo.sh
@@ -96,7 +96,7 @@ if [ $stage -le 4 ]; then
   if $run_g2p; then
     steps/dict/apply_g2p.sh $tree_dir/extvocab_nosp_lexicon/words $tree_dir/extvocab_nosp_g2p  $tree_dir/extvocab_nosp_lexicon
   else
-    cat <<EOF >$tree_dir/extvocab_nosp_lexicon//lexicon.lex
+    cat <<EOF >$tree_dir/extvocab_nosp_lexicon/lexicon.lex
 HARDWIGG	0.962436	HH AA1 R D W IH1 G
 SUDVESTR	0.162048	S AH1 D V EY1 S T R
 SUDVESTR	0.133349	S AH1 D V EH1 S T R
diff --git a/src/base/io-funcs.h b/src/base/io-funcs.h
index ca476033950..6c2b690f54c 100644
--- a/src/base/io-funcs.h
+++ b/src/base/io-funcs.h
@@ -31,7 +31,9 @@
 #include <cctype>
 #include <vector>
 #include <string>
+
 #include "base/kaldi-common.h"
+#include "base/io-funcs-inl.h"
 
 namespace kaldi {
 
@@ -235,7 +237,4 @@ inline void InitKaldiOutputStream(std::ostream &os, bool binary);
 inline bool InitKaldiInputStream(std::istream &is, bool *binary);
 
 }  // end namespace kaldi.
-
-#include "base/io-funcs-inl.h"
-
 #endif  // KALDI_BASE_IO_FUNCS_H_
diff --git a/src/bin/draw-tree.cc b/src/bin/draw-tree.cc
index ad1dd41a53f..c9be5586933 100644
--- a/src/bin/draw-tree.cc
+++ b/src/bin/draw-tree.cc
@@ -18,6 +18,7 @@
 // limitations under the License.
 
 #include "tree/tree-renderer.h"
+#include "tree/context-dep.h"
 
 void MakeEvent(std::string &qry, fst::SymbolTable *phone_syms,
                kaldi::EventType **query)
diff --git a/src/chainbin/chain-get-supervision.cc b/src/chainbin/chain-get-supervision.cc
index 6090d9f0058..1ac89d4630b 100644
--- a/src/chainbin/chain-get-supervision.cc
+++ b/src/chainbin/chain-get-supervision.cc
@@ -22,6 +22,7 @@
 #include "base/kaldi-common.h"
 #include "util/common-utils.h"
 #include "chain/chain-supervision.h"
+#include "tree/context-dep.h"
 
 namespace kaldi {
 namespace chain {
diff --git a/src/decoder/decodable-matrix.h b/src/decoder/decodable-matrix.h
index f32a007e6ca..475638a35af 100644
--- a/src/decoder/decodable-matrix.h
+++ b/src/decoder/decodable-matrix.h
@@ -26,6 +26,7 @@
 #include "base/kaldi-common.h"
 #include "hmm/transition-model.h"
 #include "itf/decodable-itf.h"
+#include "matrix/kaldi-matrix.h"
 
 namespace kaldi {
 
@@ -241,8 +242,6 @@ class DecodableMatrixScaled: public DecodableInterface {
   BaseFloat scale_;
   KALDI_DISALLOW_COPY_AND_ASSIGN(DecodableMatrixScaled);
 };
-
-
 }  // namespace kaldi
 
 #endif  // KALDI_DECODER_DECODABLE_MATRIX_H_
diff --git a/src/decoder/lattice-faster-online-decoder.h b/src/decoder/lattice-faster-online-decoder.h
index e56f24a2474..69bf8b6d98d 100644
--- a/src/decoder/lattice-faster-online-decoder.h
+++ b/src/decoder/lattice-faster-online-decoder.h
@@ -90,7 +90,7 @@ class LatticeFasterOnlineDecoderTpl:
 
   /// Outputs an FST corresponding to the single best path through the lattice.
   /// This is quite efficient because it doesn't get the entire raw lattice and find
-  /// the best path through it; insterad, it uses the BestPathEnd and BestPathIterator
+  /// the best path through it; instead, it uses the BestPathEnd and BestPathIterator
   /// so it basically traces it back through the lattice.
   /// Returns true if result is nonempty (using the return status is deprecated,
   /// it will become void).  If "use_final_probs" is true AND we reached the
diff --git a/src/decoder/training-graph-compiler.h b/src/decoder/training-graph-compiler.h
index 77c5735687f..ee56c6dfb3d 100644
--- a/src/decoder/training-graph-compiler.h
+++ b/src/decoder/training-graph-compiler.h
@@ -24,6 +24,7 @@
 #include "hmm/transition-model.h"
 #include "fst/fstlib.h"
 #include "fstext/fstext-lib.h"
+#include "tree/context-dep.h"
 
 
 namespace kaldi {
diff --git a/src/doc/grammar.dox b/src/doc/grammar.dox
index 80000c0b067..d1c6f51f349 100644
--- a/src/doc/grammar.dox
+++ b/src/doc/grammar.dox
@@ -336,7 +336,7 @@ Z_S  243
   that consume CLG.fst always also consume the <code>ilabel_info</code>, which is a <code>vector<vector<int32> ></code>.
   For a particular ilabel, say 1536, <code>ilabel_info[1536] = { 5, 21 }</code> is a vector of integers representing
   a phone-in-context.  E.g. this would represent the phone 21 with a left-context of 5.
-  Disambiguation symbols also appear on the input of CLG.fst, and they are are represented in the <code>ilabel_info</code>
+  Disambiguation symbols also appear on the input of CLG.fst, and they are represented in the <code>ilabel_info</code>
   a 1-dimensional vector like <code>{ -104 }</code> containing the negative of the disambiguation symbol's
   integer id.
 
@@ -352,7 +352,7 @@ Z_S  243
   The special symbols in CLG.fst will be as follows.
 
   The following special symbols may appear in any CLG graph, top-level or not:
-   - When any graph invokes a sub-graph, there will ben arc with an ilabel
+   - When any graph invokes a sub-graph, there will be n arc with an ilabel
      (</code>\#nonterm:foo</code>, <em>left-context-phone</em>) representing the
      user-specified nonterminal and the actual left-context, which will be
      followed by arcs with ilabels of the form (</code>\#nonterm_reenter</code>,
diff --git a/src/gmmbin/gmm-init-biphone.cc b/src/gmmbin/gmm-init-biphone.cc
index e5cc182f94c..42a9d1a91a0 100644
--- a/src/gmmbin/gmm-init-biphone.cc
+++ b/src/gmmbin/gmm-init-biphone.cc
@@ -22,6 +22,7 @@
 #include "util/common-utils.h"
 #include "gmm/am-diag-gmm.h"
 #include "tree/event-map.h"
+#include "tree/context-dep.h"
 #include "hmm/hmm-topology.h"
 #include "hmm/transition-model.h"
 
diff --git a/src/gmmbin/gmm-init-mono.cc b/src/gmmbin/gmm-init-mono.cc
index 0aac769eb70..3c370c36515 100644
--- a/src/gmmbin/gmm-init-mono.cc
+++ b/src/gmmbin/gmm-init-mono.cc
@@ -23,6 +23,7 @@
 #include "gmm/am-diag-gmm.h"
 #include "hmm/hmm-topology.h"
 #include "hmm/transition-model.h"
+#include "tree/context-dep.h"
 
 namespace kaldi {
 // This function reads a file like:
diff --git a/src/hmm/hmm-test-utils.h b/src/hmm/hmm-test-utils.h
index 495ebf278ae..4faaa92fa66 100644
--- a/src/hmm/hmm-test-utils.h
+++ b/src/hmm/hmm-test-utils.h
@@ -24,6 +24,7 @@
 #include "hmm/hmm-topology.h"
 #include "hmm/transition-model.h"
 #include "lat/kaldi-lattice.h"
+#include "tree/context-dep.h"
 
 namespace kaldi {
 
diff --git a/src/hmm/hmm-topology.h b/src/hmm/hmm-topology.h
index edea02998c0..750d35bcfe4 100644
--- a/src/hmm/hmm-topology.h
+++ b/src/hmm/hmm-topology.h
@@ -21,7 +21,6 @@
 #define KALDI_HMM_HMM_TOPOLOGY_H_
 
 #include "base/kaldi-common.h"
-#include "tree/context-dep.h"
 #include "util/const-integer-set.h"
 
 
diff --git a/src/hmm/posterior.h b/src/hmm/posterior.h
index 0c255845dd5..e153c249740 100644
--- a/src/hmm/posterior.h
+++ b/src/hmm/posterior.h
@@ -24,7 +24,6 @@
 #define KALDI_HMM_POSTERIOR_H_
 
 #include "base/kaldi-common.h"
-#include "tree/context-dep.h"
 #include "util/const-integer-set.h"
 #include "util/kaldi-table.h"
 #include "hmm/transition-model.h"
diff --git a/src/hmm/transition-model.h b/src/hmm/transition-model.h
index f03b54e8b71..e453c24f9cb 100644
--- a/src/hmm/transition-model.h
+++ b/src/hmm/transition-model.h
@@ -22,11 +22,12 @@
 #define KALDI_HMM_TRANSITION_MODEL_H_
 
 #include "base/kaldi-common.h"
-#include "tree/context-dep.h"
 #include "util/const-integer-set.h"
 #include "fst/fst-decl.h" // forward declarations.
 #include "hmm/hmm-topology.h"
 #include "itf/options-itf.h"
+#include "itf/context-dep-itf.h"
+#include "matrix/kaldi-vector.h"
 
 namespace kaldi {
 
diff --git a/src/lat/minimize-lattice.h b/src/lat/minimize-lattice.h
index fcf6c0f36df..eb13fc1c851 100644
--- a/src/lat/minimize-lattice.h
+++ b/src/lat/minimize-lattice.h
@@ -28,7 +28,6 @@
 #include "base/kaldi-common.h"
 #include "util/common-utils.h"
 #include "fstext/fstext-lib.h"
-#include "hmm/transition-model.h"
 #include "lat/kaldi-lattice.h"
 
 namespace fst {
diff --git a/src/lat/push-lattice.h b/src/lat/push-lattice.h
index e782aadc0f3..080bb637604 100644
--- a/src/lat/push-lattice.h
+++ b/src/lat/push-lattice.h
@@ -28,7 +28,6 @@
 #include "base/kaldi-common.h"
 #include "util/common-utils.h"
 #include "fstext/fstext-lib.h"
-#include "hmm/transition-model.h"
 #include "lat/kaldi-lattice.h"
 
 namespace fst {
diff --git a/src/matrix/kaldi-blas.h b/src/matrix/kaldi-blas.h
index 5d25ab852bd..8a06540bba2 100644
--- a/src/matrix/kaldi-blas.h
+++ b/src/matrix/kaldi-blas.h
@@ -50,8 +50,8 @@
 
 #ifdef HAVE_ATLAS
   extern "C" {
-    #include <cblas.h>
-    #include <clapack.h>
+    #include "cblas.h"
+    #include "clapack.h"
   }
 #elif defined(HAVE_CLAPACK)
   #ifdef __APPLE__
@@ -74,7 +74,7 @@
       // from the tools/CLAPACK_include directory.
       #include <cblas.h>
       #include <f2c.h>
-      #include <clapack.h>  
+      #include <clapack.h>
 
       // get rid of macros from f2c.h -- these are dangerous.
       #undef abs
@@ -110,7 +110,7 @@
   #undef bit_clear
   #undef bit_set
 #else
-  #error "You need to define (using the preprocessor) either HAVE_CLAPACK or HAVE_ATLAS or HAVE_MKL (but not more than one)"  
+  #error "You need to define (using the preprocessor) either HAVE_CLAPACK or HAVE_ATLAS or HAVE_MKL (but not more than one)"
 #endif
 
 #ifdef HAVE_OPENBLAS
diff --git a/src/online2/online-gmm-decodable.h b/src/online2/online-gmm-decodable.h
index c037ad0efe4..1a1d37ba2a2 100644
--- a/src/online2/online-gmm-decodable.h
+++ b/src/online2/online-gmm-decodable.h
@@ -24,8 +24,10 @@
 #define KALDI_ONLINE2_ONLINE_GMM_DECODABLE_H_
 
 #include "itf/online-feature-itf.h"
-#include "gmm/decodable-am-diag-gmm.h"
 #include "matrix/matrix-lib.h"
+#include "itf/decodable-itf.h"
+#include "gmm/am-diag-gmm.h"
+#include "hmm/transition-model.h"
 
 namespace kaldi {
 
@@ -37,20 +39,20 @@ class DecodableDiagGmmScaledOnline : public DecodableInterface {
                                const BaseFloat scale,
                                OnlineFeatureInterface *input_feats);
 
-  
+
   /// Returns the scaled log likelihood
   virtual BaseFloat LogLikelihood(int32 frame, int32 index);
-  
+
   virtual bool IsLastFrame(int32 frame) const;
 
-  virtual int32 NumFramesReady() const;  
-  
+  virtual int32 NumFramesReady() const;
+
   /// Indices are one-based!  This is for compatibility with OpenFst.
   virtual int32 NumIndices() const { return trans_model_.NumTransitionIds(); }
 
  private:
   void CacheFrame(int32 frame);
-  
+
   OnlineFeatureInterface *features_;
   const AmDiagGmm &ac_model_;
   BaseFloat ac_scale_;
diff --git a/src/online2/online-speex-wrapper.cc b/src/online2/online-speex-wrapper.cc
index 0af5bd90bd0..e41a812ca32 100644
--- a/src/online2/online-speex-wrapper.cc
+++ b/src/online2/online-speex-wrapper.cc
@@ -18,7 +18,7 @@
 // limitations under the License.
 
 #include <cstring>
-#include "online-speex-wrapper.h"
+#include "online2/online-speex-wrapper.h"
 
 namespace kaldi {
 
diff --git a/src/online2/onlinebin-util.cc b/src/online2/onlinebin-util.cc
index 74c594eeb79..f143ebbc5f7 100644
--- a/src/online2/onlinebin-util.cc
+++ b/src/online2/onlinebin-util.cc
@@ -20,7 +20,7 @@
 // See the Apache 2 License for the specific language governing permissions and
 // limitations under the License.
 
-#include "onlinebin-util.h"
+#include "online2/onlinebin-util.h"
 
 namespace kaldi {
 
diff --git a/src/tree/tree-renderer.cc b/src/tree/tree-renderer.cc
index cbc2ce05677..bbaa5cda162 100644
--- a/src/tree/tree-renderer.cc
+++ b/src/tree/tree-renderer.cc
@@ -19,6 +19,8 @@
 
 #include "tree/tree-renderer.h"
 
+#include "tree/context-dep.h"
+
 namespace kaldi {
 const int32 TreeRenderer::kEdgeWidth = 1;
 const int32 TreeRenderer::kEdgeWidthQuery = 3;
diff --git a/src/tree/tree-renderer.h b/src/tree/tree-renderer.h
index 5e0b0d89198..78f4b9aa403 100644
--- a/src/tree/tree-renderer.h
+++ b/src/tree/tree-renderer.h
@@ -23,7 +23,6 @@
 #include "base/kaldi-common.h"
 #include "tree/event-map.h"
 #include "util/common-utils.h"
-#include "hmm/transition-model.h"
 #include "fst/fstlib.h"
 
 namespace kaldi {