diff --git a/src/base/kaldi-math.cc b/src/base/kaldi-math.cc
index cf06875030b..991e46a590c 100644
--- a/src/base/kaldi-math.cc
+++ b/src/base/kaldi-math.cc
@@ -42,8 +42,8 @@ int32 RoundUpToNearestPowerOfTwo(int32 n) {
 static std::mutex _RandMutex;
 
 int Rand(struct RandomState* state) {
-#ifdef _MSC_VER
-  // On Windows, just call Rand()
+#if defined(_MSC_VER) || defined(__CYGWIN__)
+  // On Windows and Cygwin, just call Rand()
   return rand();
 #else
   if (state) {
diff --git a/src/base/kaldi-utils.cc b/src/base/kaldi-utils.cc
index 1ae1dc0b758..432da426bc3 100644
--- a/src/base/kaldi-utils.cc
+++ b/src/base/kaldi-utils.cc
@@ -45,6 +45,8 @@ std::string CharToString(const char &c) {
 void Sleep(float seconds) {
 #if defined(_MSC_VER) || defined(MINGW)
   ::Sleep(static_cast<int>(seconds * 1000.0));
+#elif defined(__CYGWIN__)
+  sleep(static_cast<int>(seconds));
 #else
   usleep(static_cast<int>(seconds * 1000000.0));
 #endif
diff --git a/src/base/kaldi-utils.h b/src/base/kaldi-utils.h
index bd2da25dce7..ad606b963f4 100644
--- a/src/base/kaldi-utils.h
+++ b/src/base/kaldi-utils.h
@@ -46,16 +46,6 @@
 #endif
 #endif
 
-#ifdef _MSC_VER
-#  define KALDI_MEMALIGN(align, size, pp_orig) \
-  (*(pp_orig) = _aligned_malloc(size, align))
-#  define KALDI_MEMALIGN_FREE(x) _aligned_free(x)
-#else
-#  define KALDI_MEMALIGN(align, size, pp_orig) \
-     (!posix_memalign(pp_orig, align, size) ? *(pp_orig) : NULL)
-#  define KALDI_MEMALIGN_FREE(x) free(x)
-#endif
-
 #ifdef __ICC
 #pragma warning(disable: 383)  // ICPC remark we don't want.
 #pragma warning(disable: 810)  // ICPC remark we don't want.
@@ -136,6 +126,9 @@ template<> class KaldiCompileTimeAssert<true> {
 
 #ifdef _MSC_VER
 #define KALDI_STRCASECMP _stricmp
+#elif defined(__CYGWIN__)
+#include <strings.h>
+#define KALDI_STRCASECMP strcasecmp
 #else
 #define KALDI_STRCASECMP strcasecmp
 #endif
diff --git a/src/cudamatrix/cu-matrix.cc b/src/cudamatrix/cu-matrix.cc
index 34290561cc5..20bc46e8154 100644
--- a/src/cudamatrix/cu-matrix.cc
+++ b/src/cudamatrix/cu-matrix.cc
@@ -101,7 +101,7 @@ void CuMatrix<Real>::Destroy() {
   } else
 #endif
   {
-    if (this->data_ != NULL) KALDI_MEMALIGN_FREE(this->data_);
+    if (this->data_ != NULL) free(this->data_);
   }
   this->data_ = NULL;
   this->num_rows_ = 0;
diff --git a/src/cudamatrix/cu-packed-matrix.cc b/src/cudamatrix/cu-packed-matrix.cc
index 64f8afe0616..ad2a5694e18 100644
--- a/src/cudamatrix/cu-packed-matrix.cc
+++ b/src/cudamatrix/cu-packed-matrix.cc
@@ -91,7 +91,7 @@ void CuPackedMatrix<Real>::Destroy() {
   } else
 #endif
   {
-    if (this->data_ != NULL) KALDI_MEMALIGN_FREE(this->data_);
+    if (this->data_ != NULL) free(this->data_);
   }
   this->data_ = NULL;
   this->num_rows_ = 0;
diff --git a/src/cudamatrix/cu-vector.cc b/src/cudamatrix/cu-vector.cc
index f85d20d37f1..08417c8fffb 100644
--- a/src/cudamatrix/cu-vector.cc
+++ b/src/cudamatrix/cu-vector.cc
@@ -1015,7 +1015,7 @@ void CuVector<Real>::Destroy() {
   } else
 #endif
   {
-    if (this->data_ != NULL) KALDI_MEMALIGN_FREE(this->data_);
+    if (this->data_ != NULL) free(this->data_);
   }
   this->data_ = NULL;
   this->dim_ = 0;
diff --git a/src/fstext/lattice-weight-test.cc b/src/fstext/lattice-weight-test.cc
index 8bcbeb4eba3..ae768e711f7 100644
--- a/src/fstext/lattice-weight-test.cc
+++ b/src/fstext/lattice-weight-test.cc
@@ -34,11 +34,12 @@ typedef CompactLatticeWeightCommonDivisorTpl<LatticeWeight, int32>
 
 
 LatticeWeight RandomLatticeWeight() {
-  if (kaldi::Rand() % 3 == 0) {
+  int tmp = kaldi::Rand() % 4;
+  if (tmp == 0) {
     return LatticeWeight::Zero();
-  } else if (kaldi::Rand() % 3 ==  0) {
+  } else if (tmp == 1) {
     return LatticeWeight( 1, 2);  // sometimes return special values..
-  } else if (kaldi::Rand() % 3 ==  0) {
+  } else if (tmp == 2) {
     return LatticeWeight( 2, 1);  // this tests more thoroughly certain properties...
   } else {
     return LatticeWeight( 100 * kaldi::RandGauss(), 100 * kaldi::RandGauss());
@@ -107,7 +108,7 @@ void LatticeWeightTest() {
     s1 << l1;
     std::istringstream s2(s1.str());
     s2 >> l2;
-    KALDI_ASSERT(ApproxEqual(l1, l2));
+    KALDI_ASSERT(ApproxEqual(l1, l2, 0.001));
     std::cout << s1.str() << '\n';
     {
       std::ostringstream s1b;
diff --git a/src/makefiles/cygwin.mk b/src/makefiles/cygwin.mk
index c58cd3a42da..d2b9c0e6474 100644
--- a/src/makefiles/cygwin.mk
+++ b/src/makefiles/cygwin.mk
@@ -10,12 +10,12 @@ ifndef OPENFSTLIBS
 $(error OPENFSTLIBS not defined.)
 endif
 
-CXXFLAGS = -std=c++11 -I.. -I$(OPENFSTINC) $(EXTRA_CXXFLAGS) \
+CXXFLAGS = -std=c++11 -U__STRICT_ANSI__ -I.. -I$(OPENFSTINC) $(EXTRA_CXXFLAGS) \
            -Wall -Wno-sign-compare -Wno-unused-local-typedefs \
            -Wno-deprecated-declarations -Winit-self \
            -DKALDI_DOUBLEPRECISION=$(DOUBLE_PRECISION) \
            -DHAVE_CLAPACK -I../../tools/CLAPACK/ \
-           -msse -msse2 \
+           -msse -msse2 -O -Wa,-mbig-obj \
            -g # -O0 -DKALDI_PARANOID
 
 ifeq ($(KALDI_FLAVOR), dynamic)
diff --git a/src/matrix/kaldi-matrix.cc b/src/matrix/kaldi-matrix.cc
index fcfe0616b64..d24a29556f7 100644
--- a/src/matrix/kaldi-matrix.cc
+++ b/src/matrix/kaldi-matrix.cc
@@ -50,9 +50,8 @@ void MatrixBase<Real>::Invert(Real *log_det, Real *det_sign,
   KaldiBlasInt result = -1;
   KaldiBlasInt l_work = std::max<KaldiBlasInt>(1, N);
   Real *p_work;
-  void *temp;
   if ((p_work = static_cast<Real*>(
-          KALDI_MEMALIGN(16, sizeof(Real)*l_work, &temp))) == NULL) {
+          aligned_alloc(16, sizeof(Real)*l_work))) == NULL) {
     delete[] pivot;
     throw std::bad_alloc();
   }
@@ -75,7 +74,7 @@ void MatrixBase<Real>::Invert(Real *log_det, Real *det_sign,
       if (det_sign) *det_sign = 0;
       delete[] pivot;
 #ifndef HAVE_ATLAS
-      KALDI_MEMALIGN_FREE(p_work);
+      free(p_work);
 #endif
       return;
     }
@@ -103,7 +102,7 @@ void MatrixBase<Real>::Invert(Real *log_det, Real *det_sign,
   if (inverse_needed) clapack_Xgetri2(&M, data_, &LDA, pivot, p_work, &l_work,
                               &result);
   delete[] pivot;
-  KALDI_MEMALIGN_FREE(p_work);
+  free(p_work);
 #else
   if (inverse_needed)
     clapack_Xgetri(num_rows_, data_, stride_, pivot, &result);
@@ -710,9 +709,8 @@ void MatrixBase<Real>::LapackGesvd(VectorBase<Real> *s, MatrixBase<Real> *U_in,
 
   l_work = static_cast<KaldiBlasInt>(work_query);
   Real *p_work;
-  void *temp;
   if ((p_work = static_cast<Real*>(
-          KALDI_MEMALIGN(16, sizeof(Real)*l_work, &temp))) == NULL)
+          aligned_alloc(16, sizeof(Real)*l_work))) == NULL)
     throw std::bad_alloc();
 
   // perform svd
@@ -729,7 +727,7 @@ void MatrixBase<Real>::LapackGesvd(VectorBase<Real> *s, MatrixBase<Real> *U_in,
   if (result != 0) {
     KALDI_WARN << "CLAPACK sgesvd_ : some weird convergence not satisfied";
   }
-  KALDI_MEMALIGN_FREE(p_work);
+  free(p_work);
 }
 
 #endif
@@ -794,7 +792,6 @@ inline void Matrix<Real>::Init(const MatrixIndexT rows,
   MatrixIndexT skip, stride;
   size_t size;
   void *data;  // aligned memory block
-  void *temp;  // memory block to be really freed
 
   // compute the size of skip and real cols
   skip = ((16 / sizeof(Real)) - cols % (16 / sizeof(Real)))
@@ -804,7 +801,7 @@ inline void Matrix<Real>::Init(const MatrixIndexT rows,
       * sizeof(Real);
 
   // allocate the memory and set the right dimensions and parameters
-  if (NULL != (data = KALDI_MEMALIGN(16, size, &temp))) {
+  if (NULL != (data = aligned_alloc(16, size))) {
     MatrixBase<Real>::data_        = static_cast<Real *> (data);
     MatrixBase<Real>::num_rows_      = rows;
     MatrixBase<Real>::num_cols_      = cols;
@@ -1127,7 +1124,7 @@ template<typename Real>
 void Matrix<Real>::Destroy() {
   // we need to free the data block if it was defined
   if (NULL != MatrixBase<Real>::data_)
-    KALDI_MEMALIGN_FREE( MatrixBase<Real>::data_);
+    free( MatrixBase<Real>::data_);
   MatrixBase<Real>::data_ = NULL;
   MatrixBase<Real>::num_rows_ = MatrixBase<Real>::num_cols_
       = MatrixBase<Real>::stride_ = 0;
diff --git a/src/matrix/kaldi-vector.cc b/src/matrix/kaldi-vector.cc
index 8cedc9c0487..2bdae2f8efc 100644
--- a/src/matrix/kaldi-vector.cc
+++ b/src/matrix/kaldi-vector.cc
@@ -173,11 +173,10 @@ inline void Vector<Real>::Init(const MatrixIndexT dim) {
   }
   MatrixIndexT size;
   void *data;
-  void *free_data;
 
   size = dim * sizeof(Real);
 
-  if ((data = KALDI_MEMALIGN(16, size, &free_data)) != NULL) {
+  if ((data = aligned_alloc(16, size)) != NULL) {
     this->data_ = static_cast<Real*> (data);
     this->dim_ = dim;
   } else {
@@ -279,7 +278,7 @@ template<typename Real>
 void Vector<Real>::Destroy() {
   /// we need to free the data block if it was defined
   if (this->data_ != NULL)
-    KALDI_MEMALIGN_FREE(this->data_);
+    free(this->data_);
   this->data_ = NULL;
   this->dim_ = 0;
 }
diff --git a/src/matrix/packed-matrix.cc b/src/matrix/packed-matrix.cc
index 80bf5891998..9963eaf16fc 100644
--- a/src/matrix/packed-matrix.cc
+++ b/src/matrix/packed-matrix.cc
@@ -67,9 +67,8 @@ inline void PackedMatrix<Real>::Init(MatrixIndexT r) {
   }
 
   void *data;  // aligned memory block
-  void *temp;
 
-  if ((data = KALDI_MEMALIGN(16, size * sizeof(Real), &temp)) != NULL) {
+  if ((data = aligned_alloc(16, size * sizeof(Real))) != NULL) {
     this->data_ = static_cast<Real *> (data);
     this->num_rows_ = r;
   } else {
@@ -226,7 +225,7 @@ Real PackedMatrix<Real>::Trace() const {
 template<typename Real>
 void PackedMatrix<Real>::Destroy() {
   // we need to free the data block if it was defined
-  if (data_ != NULL) KALDI_MEMALIGN_FREE(data_);
+  if (data_ != NULL) free(data_);
   data_ = NULL;
   num_rows_ = 0;
 }
diff --git a/src/matrix/sp-matrix.cc b/src/matrix/sp-matrix.cc
index 224ef39fb6e..5ecb79f0bf5 100644
--- a/src/matrix/sp-matrix.cc
+++ b/src/matrix/sp-matrix.cc
@@ -222,9 +222,8 @@ void SpMatrix<Real>::Invert(Real *logdet, Real *det_sign, bool need_inverse) {
   KaldiBlasInt   rows = static_cast<int>(this->num_rows_);
   KaldiBlasInt*  p_ipiv = new KaldiBlasInt[rows];
   Real *p_work;  // workspace for the lapack function
-  void *temp;
   if ((p_work = static_cast<Real*>(
-          KALDI_MEMALIGN(16, sizeof(Real) * rows, &temp))) == NULL) {
+          aligned_alloc(16, sizeof(Real) * rows))) == NULL) {
     delete[] p_ipiv;
     throw std::bad_alloc();
   }
@@ -277,7 +276,7 @@ void SpMatrix<Real>::Invert(Real *logdet, Real *det_sign, bool need_inverse) {
   }
   if (!need_inverse) {
     delete [] p_ipiv;
-    KALDI_MEMALIGN_FREE(p_work);
+    free(p_work);
     return;  // Don't need what is computed next.
   }
   // NOTE: Even though "U" is for upper, lapack assumes column-wise storage
@@ -292,7 +291,7 @@ void SpMatrix<Real>::Invert(Real *logdet, Real *det_sign, bool need_inverse) {
   }
 
   delete [] p_ipiv;
-  KALDI_MEMALIGN_FREE(p_work);
+  free(p_work);
 }
 #else
 // in the ATLAS case, these are not implemented using a library and we back off to something else.
diff --git a/src/nnetbin/cuda-gpu-available.cc b/src/nnetbin/cuda-gpu-available.cc
index 89fd26be86f..390468d3046 100644
--- a/src/nnetbin/cuda-gpu-available.cc
+++ b/src/nnetbin/cuda-gpu-available.cc
@@ -41,7 +41,7 @@ void TestGpuComputation() {
 
 int main(int argc, char *argv[]) try {
   char hostname[100] = "UNKNOWN-HOSTNAME";
-#ifndef _MSC_VER
+#if !defined(_MSC_VER) && !defined(__CYGWIN__)
   if (gethostname(hostname, 100)) {
     KALDI_WARN << "Cannot get hostname, " << strerror(errno);
   }
diff --git a/src/util/kaldi-io.cc b/src/util/kaldi-io.cc
index d577c1e9a89..5ede0836243 100644
--- a/src/util/kaldi-io.cc
+++ b/src/util/kaldi-io.cc
@@ -26,6 +26,8 @@
 #include "util/kaldi-holder.h"
 #include "util/kaldi-pipebuf.h"
 #include "util/kaldi-table.h"  // for Classify{W,R}specifier
+#include <stdio.h>
+#include <stdlib.h>
 
 #ifdef KALDI_CYGWIN_COMPAT
 #include "util/kaldi-cygwin-io-inl.h"
@@ -34,7 +36,7 @@
 #define MapOsPath(x) x
 #endif  // KALDI_CYGWIN_COMPAT
 
-#ifdef _MSC_VER
+#ifdef _MSC_VER 
 static FILE *popen(const char* command, const char* mode) {
 #ifdef KALDI_CYGWIN_COMPAT
   return kaldi::CygwinCompatPopen(command, mode);
@@ -280,7 +282,7 @@ class PipeOutputImpl: public OutputImplBase {
     KALDI_ASSERT(wxfilename.length() != 0 && wxfilename[0] == '|');  // should
     // start with '|'
     std::string cmd_name(wxfilename, 1);
-#ifdef _MSC_VER
+#if defined(_MSC_VER) || defined(__CYGWIN__)
     f_ = popen(cmd_name.c_str(), (binary ? "wb" : "w"));
 #else
     f_ = popen(cmd_name.c_str(), "w");
@@ -457,7 +459,7 @@ class PipeInputImpl: public InputImplBase {
     KALDI_ASSERT(rxfilename.length() != 0 &&
            rxfilename[rxfilename.length()-1] == '|');  // should end with '|'
     std::string cmd_name(rxfilename, 0, rxfilename.length()-1);
-#ifdef _MSC_VER
+#if defined(_MSC_VER) || defined(__CYGWIN__)
     f_ = popen(cmd_name.c_str(), (binary ? "rb" : "r"));
 #else
     f_ = popen(cmd_name.c_str(), "r");
diff --git a/src/util/parse-options.h b/src/util/parse-options.h
index 12ed62bb55d..3d76b692c7d 100644
--- a/src/util/parse-options.h
+++ b/src/util/parse-options.h
@@ -38,7 +38,7 @@ class ParseOptions : public OptionsItf {
   explicit ParseOptions(const char *usage) :
     print_args_(true), help_(false), usage_(usage), argc_(0), argv_(NULL),
     prefix_(""), other_parser_(NULL) {
-#ifndef _MSC_VER  // This is just a convenient place to set the stderr to line
+#if !defined(_MSC_VER) && !defined(__CYGWIN__) // This is just a convenient place to set the stderr to line
     setlinebuf(stderr);  // buffering mode, since it's called at program start.
 #endif  // This helps ensure different programs' output is not mixed up.
     RegisterStandard("config", &config_, "Configuration file to read (this "