diff --git a/wechat/audio.py b/wechat/audio.py
index 12cc038..0a229a2 100644
--- a/wechat/audio.py
+++ b/wechat/audio.py
@@ -2,6 +2,7 @@
 # -*- coding: UTF-8 -*-
 
 import os
+import tempfile
 import logging
 logger = logging.getLogger(__name__)
 
@@ -10,9 +11,6 @@ from .common.procutil import subproc_succ
 
 SILK_DECODER = os.path.join(os.path.dirname(__file__),
                             '../third-party/silk/decoder')
-if not os.path.exists(SILK_DECODER):
-    logger.error("Silk decoder is not compiled. Please see README.md.")
-    raise RuntimeError()
 
 def parse_wechat_audio_file(file_name):
     try:
@@ -25,44 +23,46 @@ def do_parse_wechat_audio_file(file_name):
     """ return a mp3 stored in base64 unicode string, and the duration"""
     if not file_name: return "", 0
 
-    mp3_file = os.path.join('/tmp',
-                            os.path.basename(file_name)[:-4] + '.mp3')
-    with open(file_name, 'rb') as f:
-        header = f.read(10)
-    if b'AMR' in header:
-        cmd = f"sox -e signed -c 1 {file_name} {mp3_file}"
-        subproc_succ(cmd)
-        cmd = f"soxi -D {mp3_file}"
-        duration = float(subproc_succ(cmd))
+    with tempfile.TemporaryDirectory(prefix="wechatdump_audio") as temp:
+        mp3_file = os.path.join(temp,
+                                os.path.basename(file_name)[:-4] + '.mp3')
+        with open(file_name, 'rb') as f:
+            header = f.read(10)
+        if b'AMR' in header:
+            cmd = f"sox -e signed -c 1 {file_name} {mp3_file}"
+            subproc_succ(cmd)
+            cmd = f"soxi -D {mp3_file}"
+            duration = float(subproc_succ(cmd))
 
-        # The below is python2 only. It should be equivalent to using sox from command line
-        # import pysox
-        # infile = pysox.CSoxStream(file_name)
-        # outfile = pysox.CSoxStream(mp3_file, 'w', infile.get_signal())
-        # chain = pysox.CEffectsChain(infile, outfile)
-        # chain.flow_effects()
-        # outfile.close()
-        # signal = infile.get_signal().get_signalinfo()
-        # duration = signal['length'] * 1.0 / signal['rate']
-    elif b'SILK' in header:
-        raw_file = os.path.join('/tmp',
-                                os.path.basename(file_name)[:-4] + '.raw')
-        cmd = '{0} {1} {2}'.format(SILK_DECODER, file_name, raw_file)
-        out = subproc_succ(cmd)
-        for line in out.split(b'\n'):
-            if b'File length' in line:
-                duration = float(line[13:-3].strip())
-                break
+            # The below is python2 only. It should be equivalent to using sox from command line
+            # import pysox
+            # infile = pysox.CSoxStream(file_name)
+            # outfile = pysox.CSoxStream(mp3_file, 'w', infile.get_signal())
+            # chain = pysox.CEffectsChain(infile, outfile)
+            # chain.flow_effects()
+            # outfile.close()
+            # signal = infile.get_signal().get_signalinfo()
+            # duration = signal['length'] * 1.0 / signal['rate']
+        elif b'SILK' in header:
+            if not os.path.exists(SILK_DECODER):
+                raise RuntimeError("Silk decoder is not compiled. Please see README.md.")
+
+            raw_file = os.path.join(temp,
+                                    os.path.basename(file_name)[:-4] + '.raw')
+            cmd = '{0} {1} {2}'.format(SILK_DECODER, file_name, raw_file)
+            out = subproc_succ(cmd)
+            for line in out.split(b'\n'):
+                if b'File length' in line:
+                    duration = float(line[13:-3].strip())
+                    break
+            else:
+                raise RuntimeError("Error decoding silk audio file!" + out.decode('utf-8'))
+
+            # TODO don't know how to do this with python
+            subproc_succ('sox -r 24000 -e signed -b 16 -c 1 {} {}'.format(raw_file, mp3_file))
         else:
-            raise RuntimeError("Error decoding silk audio file!")
-
-        # TODO don't know how to do this with python
-        subproc_succ('sox -r 24000 -e signed -b 16 -c 1 {} {}'.format(raw_file, mp3_file))
-        os.unlink(raw_file)
-    else:
-        raise NotImplementedError("Audio file format cannot be recognized.")
-    mp3_string = get_file_b64(mp3_file)
-    os.unlink(mp3_file)
+            raise NotImplementedError("Audio file format cannot be recognized.")
+        mp3_string = get_file_b64(mp3_file)
     return mp3_string, duration
 
 if __name__ == '__main__':