Merge pull request #170 from Holzhaus/avoid-hardcoded-paths

Make HMM dir and FST model configurable.

2 files changed, 32 insertions, 9 deletions

diff --git a/client/g2p.py b/client/g2p.py
index 3225c73..87e8b0d 100644
--- a/client/g2p.py
+++ b/client/g2p.py
@@ -3,10 +3,22 @@ import os
 import tempfile
 import subprocess
 import re
+import yaml
 
 PHONE_MATCH = re.compile(r'<s> (.*) </s>')
-PHONETISAURUS_PATH = os.environ['JASPER_HOME'] + "/phonetisaurus"
 
+FST_MODEL = None
+
+# Try to get fst_model from config
+profile_path = os.path.join(os.path.dirname(__file__),'profile.yml')
+if os.path.exists(profile_path):
+    with open(profile_path, 'r') as f:
+        profile = yaml.safe_load(f)
+        if 'pocketsphinx' in profile and 'fst_model' in profile['pocketsphinx']:
+            FST_MODEL = profile['pocketsphinx']['fst_model']
+
+if not FST_MODEL:
+    FST_MODEL = os.environ['JASPER_HOME'] + "/phonetisaurus/g014b2b.fst"
 
 def parseLine(line):
     return PHONE_MATCH.search(line).group(1)
@@ -17,8 +29,7 @@ def parseOutput(output):
 
 
 def translateWord(word):
-    out = subprocess.check_output(['phonetisaurus-g2p', '--model=%s' %
-                                  PHONETISAURUS_PATH + "/g014b2b.fst", '--input=%s' % word])
+    out = subprocess.check_output(['phonetisaurus-g2p', '--model=%s' % FST_MODEL, '--input=%s' % word])
     return parseLine(out)
 
 
@@ -36,8 +47,7 @@ def translateWords(words):
 
 
 def translateFile(input_filename, output_filename=None):
-    out = subprocess.check_output(['phonetisaurus-g2p', '--model=%s' %
-        PHONETISAURUS_PATH + "/g014b2b.fst", '--input=%s' % input_filename, '--words', '--isfile'])
+    out = subprocess.check_output(['phonetisaurus-g2p', '--model=%s' % FST_MODEL, '--input=%s' % input_filename, '--words', '--isfile'])
     out = parseOutput(out)
 
     if output_filename:
diff --git a/client/stt.py b/client/stt.py
index e14ab26..d012245 100644
--- a/client/stt.py
+++ b/client/stt.py
@@ -1,8 +1,10 @@
 #!/usr/bin/env python2
 # -*- coding: utf-8-*-
+import os
 import traceback
 import json
 import requests
+import yaml
 
 """
 The default Speech-to-Text implementation which relies on PocketSphinx.
@@ -31,13 +33,24 @@ class PocketSphinxSTT(object):
         except:
             import pocketsphinx as ps
 
-        hmdir = "/usr/local/share/pocketsphinx/model/hmm/en_US/hub4wsj_sc_8k"
+        hmm_dir = None
+
+        # Try to get hmm_dir from config
+        profile_path = os.path.join(os.path.dirname(__file__),'profile.yml')
+        if os.path.exists(profile_path):
+            with open(profile_path, 'r') as f:
+                profile = yaml.safe_load(f)
+                if 'pocketsphinx' in profile and 'hmm_dir' in profile['pocketsphinx']:
+                    hmm_dir = profile['pocketsphinx']['hmm_dir']
+
+        if not hmm_dir:
+            hmm_dir = "/usr/local/share/pocketsphinx/model/hmm/en_US/hub4wsj_sc_8k"
 
         if lmd_music and dictd_music:
-            self.speechRec_music = ps.Decoder(hmm=hmdir, lm=lmd_music, dict=dictd_music)
+            self.speechRec_music = ps.Decoder(hmm=hmm_dir, lm=lmd_music, dict=dictd_music)
         self.speechRec_persona = ps.Decoder(
-            hmm=hmdir, lm=lmd_persona, dict=dictd_persona)
-        self.speechRec = ps.Decoder(hmm=hmdir, lm=lmd, dict=dictd)
+            hmm=hmm_dir, lm=lmd_persona, dict=dictd_persona)
+        self.speechRec = ps.Decoder(hmm=hmm_dir, lm=lmd, dict=dictd)
 
     def transcribe(self, audio_file_path, PERSONA_ONLY=False, MUSIC=False):
         """