From 9405dce4d9f59f29b04eff83cb8909cc772249d5 Mon Sep 17 00:00:00 2001
From: "Matthew D. Scholefield" <matthew3311999@gmail.com>
Date: Thu, 30 Jun 2016 14:22:53 -0500
Subject: [PATCH] Decreased "wake up" false positives - Fixes #189 (#252)
* Increased kws threshold for the wakeup_recognizer
* Fixed pep8 in listener
* Fixed unit tests
---
mycroft/client/speech/listener.py | 16 ++++++++++++----
mycroft/client/speech/local_recognizer.py | 12 ++++++------
test/client/audio_consumer_test.py | 3 ++-
test/client/local_recognizer_test.py | 11 ++++++-----
4 files changed, 26 insertions(+), 16 deletions(-)
diff --git a/mycroft/client/speech/listener.py b/mycroft/client/speech/listener.py
index 77b190901c13..40181ceb07a5 100644
--- a/mycroft/client/speech/listener.py
+++ b/mycroft/client/speech/listener.py
@@ -199,21 +199,29 @@ def __init__(self):
class RecognizerLoop(pyee.EventEmitter):
def __init__(self, channels=int(speech_config.get('channels')),
- sample_rate=int(speech_config.get('sample_rate')),
+ rate=int(speech_config.get('sample_rate')),
device_index=None,
lang=core_config.get('lang')):
pyee.EventEmitter.__init__(self)
- self.microphone = MutableMicrophone(sample_rate=sample_rate,
+ self.microphone = MutableMicrophone(sample_rate=rate,
device_index=device_index)
# FIXME - channels are not been used
self.microphone.CHANNELS = channels
- self.mycroft_recognizer = LocalRecognizer(sample_rate, lang)
+ self.mycroft_recognizer = self.create_mycroft_recognizer(rate, lang)
# TODO - localization
- self.wakeup_recognizer = LocalRecognizer(sample_rate, lang, "wake up")
+ self.wakeup_recognizer = self.create_wakeup_recognizer(rate, lang)
self.remote_recognizer = ResponsiveRecognizer(self.mycroft_recognizer)
self.state = RecognizerLoopState()
+ @staticmethod
+ def create_mycroft_recognizer(rate, lang):
+ return LocalRecognizer("hey mycroft", "1e-90", rate, lang)
+
+ @staticmethod
+ def create_wakeup_recognizer(rate, lang):
+ return LocalRecognizer("wake up", "1e-10", rate, lang)
+
def start_async(self):
self.state.running = True
queue = Queue()
diff --git a/mycroft/client/speech/local_recognizer.py b/mycroft/client/speech/local_recognizer.py
index b3a61011d22d..973a255400fa 100644
--- a/mycroft/client/speech/local_recognizer.py
+++ b/mycroft/client/speech/local_recognizer.py
@@ -27,25 +27,25 @@
class LocalRecognizer(object):
- def __init__(self, sample_rate=16000, lang="en-us",
- key_phrase="hey mycroft"):
+ def __init__(self, key_phrase, threshold, sample_rate=16000, lang="en-us"):
self.lang = lang
self.key_phrase = key_phrase
self.sample_rate = sample_rate
- self.configure()
+ self.threshold = threshold
+ self.decoder = Decoder(self.create_config())
- def configure(self):
+ def create_config(self):
config = Decoder.default_config()
config.set_string('-hmm', os.path.join(BASEDIR, 'model', self.lang,
'hmm'))
config.set_string('-dict', os.path.join(BASEDIR, 'model', self.lang,
'mycroft-en-us.dict'))
config.set_string('-keyphrase', self.key_phrase)
- config.set_float('-kws_threshold', float('1e-90'))
+ config.set_float('-kws_threshold', float(self.threshold))
config.set_float('-samprate', self.sample_rate)
config.set_int('-nfft', 2048)
config.set_string('-logfn', '/dev/null')
- self.decoder = Decoder(config)
+ return config
def transcribe(self, byte_data, metrics=None):
start = time.time()
diff --git a/test/client/audio_consumer_test.py b/test/client/audio_consumer_test.py
index c811965060dd..4e5890860673 100644
--- a/test/client/audio_consumer_test.py
+++ b/test/client/audio_consumer_test.py
@@ -24,6 +24,7 @@
from speech_recognition import WavFile, AudioData
from mycroft.client.speech.listener import AudioConsumer, RecognizerLoop
+from mycroft.client.speech.local_recognizer import LocalRecognizer
from mycroft.client.speech.recognizer_wrapper import (
RemoteRecognizerWrapperFactory
)
@@ -59,7 +60,7 @@ def setUp(self):
self.loop.state,
self.queue,
self.loop,
- self.loop.wakeup_recognizer,
+ LocalRecognizer(self.loop.wakeup_recognizer.key_phrase, "1e-45"),
self.loop.mycroft_recognizer,
RemoteRecognizerWrapperFactory.wrap_recognizer(
self.recognizer, 'google'))
diff --git a/test/client/local_recognizer_test.py b/test/client/local_recognizer_test.py
index a9cb151355ae..d91a4434ce5f 100644
--- a/test/client/local_recognizer_test.py
+++ b/test/client/local_recognizer_test.py
@@ -3,7 +3,7 @@
import os
from speech_recognition import WavFile
-from mycroft.client.speech.local_recognizer import LocalRecognizer
+from mycroft.client.speech.listener import RecognizerLoop
__author__ = 'seanfitz'
@@ -12,20 +12,21 @@
class LocalRecognizerTest(unittest.TestCase):
def setUp(self):
- self.recognizer = LocalRecognizer()
+ self.recognizer = RecognizerLoop.create_mycroft_recognizer(16000,
+ "en-us")
def testRecognizerWrapper(self):
source = WavFile(os.path.join(DATA_DIR, "hey_mycroft.wav"))
with source as audio:
hyp = self.recognizer.transcribe(audio.stream.read())
- assert "mycroft" in hyp.hypstr.lower()
+ assert self.recognizer.key_phrase in hyp.hypstr.lower()
source = WavFile(os.path.join(DATA_DIR, "mycroft.wav"))
with source as audio:
hyp = self.recognizer.transcribe(audio.stream.read())
- assert "mycroft" in hyp.hypstr.lower()
+ assert self.recognizer.key_phrase in hyp.hypstr.lower()
def testRecognitionInLongerUtterance(self):
source = WavFile(os.path.join(DATA_DIR, "weather_mycroft.wav"))
with source as audio:
hyp = self.recognizer.transcribe(audio.stream.read())
- assert "mycroft" in hyp.hypstr.lower()
+ assert self.recognizer.key_phrase in hyp.hypstr.lower()