Use Transformer in-memory with stdin/stdout (#102)

* added in-memory build_array for sox transformer

* adding pysoundfile to travis

* upping coverage and adding docstrings

* merging build_array into build, handling cases for both inputs and outputs to be either arrays or filepaths.

* add numpy to docs requirements

* increase test coverage

Co-authored-by: Rachel Bittner <rmb456@nyu.edu>

.gitignore

@@ -1,4 +1,5 @@
 tests/data/output.wav
+tests/data/output_alt.wav
 *.DS_Store
 
 # Byte-compiled / optimized / DLL files
@@ -63,3 +64,5 @@ target/
 
 #Ipython Notebook
 .ipynb_checkpoints
+
+**/noise.prof

.travis.yml

@@ -15,6 +15,7 @@ addons:
 # command to install dependencies
 install:
   - pip install coveralls
+  - pip install pysoundfile
   - pip install -e .
 
 # command to run tests

docs/requirements.txt

@@ -1 +1,2 @@
-numpydoc>=0.5
+numpydoc>=0.5
+numpy >= 1.9.0

setup.py

@@ -20,12 +20,14 @@
         keywords='audio effects SoX',
         license='BSD-3-Clause',
         install_requires=[
+            'numpy >= 1.9.0',
         ],
         extras_require={
             'tests': [
                 'pytest',
                 'pytest-cov',
                 'pytest-pep8',
+                'pysoundfile >= 0.9.0',
             ],
             'docs': [
                 'sphinx==1.2.3',  # autodoc was broken in 1.3.1

sox/core.py

@@ -3,6 +3,7 @@
 
 import subprocess
 from subprocess import CalledProcessError
+import numpy as np
 
 from . import NO_SOX
 
@@ -14,19 +15,31 @@
 ]
 
 
-def sox(args):
+def sox(args, src_array=None, decode_out_with_utf=True):
     '''Pass an argument list to SoX.
 
     Parameters
     ----------
     args : iterable
         Argument list for SoX. The first item can, but does not
         need to, be 'sox'.
+    src_array : np.ndarray, or None
+        If src_array is not None, then we make sure it's a numpy 
+        array and pass it into stdin.
+    decode_out_with_utf : bool, default=True
+        Whether or not sox is outputting a bytestring that should be
+        decoded with utf-8.
 
     Returns:
     --------
     status : bool
         True on success.
+    out : str, np.ndarray, or None
+        Returns a np.ndarray if src_array was an np.ndarray. 
+        Returns the stdout produced by sox if src_array is None.
+        Otherwise, returns None if there's an error.
+    err : str, or None
+        Returns stderr as a string.
 
     '''
     if args[0].lower() != "sox":
@@ -37,15 +50,33 @@ def sox(args):
     try:
         logger.info("Executing: %s", ' '.join(args))
 
-        process_handle = subprocess.Popen(
-            args, stdout=subprocess.PIPE, stderr=subprocess.PIPE
-        )
-
-        out, err = process_handle.communicate()
-        out = out.decode("utf-8")
-        err = err.decode("utf-8")
-
-        status = process_handle.returncode
+        if src_array is None:
+            process_handle = subprocess.Popen(
+                args, stdout=subprocess.PIPE, stderr=subprocess.PIPE
+            )
+
+            out, err = process_handle.communicate()
+            if decode_out_with_utf:
+                out = out.decode("utf-8")
+            err = err.decode("utf-8")
+
+            status = process_handle.returncode
+        elif isinstance(src_array, np.ndarray):
+            process_handle = subprocess.Popen(
+                args, 
+                stdin=subprocess.PIPE, 
+                stdout=subprocess.PIPE,
+                stderr=subprocess.PIPE
+            )
+            # We do order "F" for Fortran formatting of the numpy array, which is
+            # sox expects. When we reshape stdout later, we need to use the same
+            # order, otherwise tests fail.
+            out, err = process_handle.communicate(src_array.T.tobytes(order='F'))
+            err = err.decode("utf-8")
+            status = process_handle.returncode
+        else:
+            raise TypeError("src_array must be an np.ndarray!")
+
         return status, out, err
 
     except OSError as error_msg:

sox/file_info.py

@@ -24,7 +24,7 @@ def bitdepth(input_filepath):
         Number of bits per sample.
         Returns None if not applicable.
     '''
-    
+
     validate_input_file(input_filepath)
     output = soxi(input_filepath, 'b')
     if output == '0':
@@ -49,7 +49,7 @@ def bitrate(input_filepath):
         Bit rate, expressed in bytes per second.
         Returns None if not applicable.
     '''
-    
+
     validate_input_file(input_filepath)
     output = soxi(input_filepath, 'B')
     # The characters below stand for kilo, Mega, Giga, etc.
@@ -291,10 +291,12 @@ def validate_output_file(output_filepath):
         The output filepath.
 
     '''
+    if output_filepath == '-n':
+        return
 
     nowrite_conditions = [
-        bool(os.path.dirname(output_filepath)) or\
-            not os.access(os.getcwd(), os.W_OK),
+        bool(os.path.dirname(output_filepath)) or
+             not os.access(os.getcwd(), os.W_OK),
         not os.access(os.path.dirname(output_filepath), os.W_OK)]
 
     if all(nowrite_conditions):