forked from Mirrors/freeswitch
speech recognition demo in python
git-svn-id: http://svn.freeswitch.org/svn/freeswitch/trunk@5659 d0543943-73ff-0310-b7d9-9358b9ac24b2
This commit is contained in:
Traun Leyden
parent
88bbc489d1
commit
b3454792b1
|
|
@ -0,0 +1 @@
|
|||
# empty
|
||||
|
|
@ -0,0 +1,198 @@
|
|||
from freeswitch import *
|
||||
from xml.dom import minidom
|
||||
|
||||
VOICE_ENGINE = "cepstral"
|
||||
VOICE = "William"
|
||||
|
||||
"""
|
||||
A few classes that make it easier to write speech applications
|
||||
using Python. It is roughly modelled after the equivalent that
|
||||
is written in JavaScript.
|
||||
|
||||
Status: should work, but not yet complete. some pending items
|
||||
are mentioned in comments
|
||||
"""
|
||||
|
||||
class Grammar:
|
||||
def __init__(self, name, path, obj_path,
|
||||
min_score=1, confirm_score=400, halt=False):
|
||||
"""
|
||||
@param name - name of grammar to reference it later
|
||||
@param path - path to xml grammar file
|
||||
@param obj_path - xml path to find interpretation from root
|
||||
in result xml, eg, 'interpretation'
|
||||
@param min_score - score threshold to accept result
|
||||
@param confirm_score - if score below this threshold, ask user
|
||||
if they are sure this is correct
|
||||
@param halt - not sure what was used for in js, currently unused
|
||||
"""
|
||||
self.name=name
|
||||
self.path=path
|
||||
self.obj_path=obj_path
|
||||
self.min_score=min_score
|
||||
self.confirm_score=confirm_score
|
||||
self.halt=halt
|
||||
|
||||
|
||||
class SpeechDetect:
|
||||
|
||||
def __init__(self, session, module_name, ip_addr):
|
||||
self.session=session
|
||||
self.module_name=module_name
|
||||
self.ip_addr=ip_addr
|
||||
self.grammars = {}
|
||||
|
||||
def addGrammar(self, grammar):
|
||||
self.grammars[grammar.name]=grammar
|
||||
|
||||
def setGrammar(self, name):
|
||||
self.grammar = self.grammars[name]
|
||||
|
||||
def detectSpeech(self):
|
||||
# TODO: we might not always want to call detect_speech
|
||||
# with this cmd, see js version for other options
|
||||
# also see detect_speech_function() in mod_dptools.c
|
||||
cmd = "%s %s %s %s" % (self.module_name,
|
||||
self.grammar.name,
|
||||
self.grammar.path,
|
||||
self.ip_addr)
|
||||
console_log("debug", "calling detect_speech with: %s\n" % cmd)
|
||||
self.session.execute("detect_speech", cmd)
|
||||
console_log("debug", "finished calling detect_speech\n")
|
||||
|
||||
class SpeechObtainer:
|
||||
|
||||
def __init__(self, speech_detect, required_phrases, wait_time, max_tries):
|
||||
"""
|
||||
@param speech_detect - the speech detect object, which holds a
|
||||
reference to underlying session and can
|
||||
be re-used by many SpeechObtainers
|
||||
@param required_phrases - the number of required phrases from the
|
||||
grammar. for example if its prompting for
|
||||
the toppings on a sandwhich and min toppings
|
||||
is 3, use 3. normally will be 1.
|
||||
@param wait_time - the time, in millisconds, to wait for
|
||||
input during each loop iteration
|
||||
@param max_tries - this number multiplied by wait time gives the
|
||||
'total wait time' before we give up and return
|
||||
partial or no result
|
||||
"""
|
||||
self.speech_detect=speech_detect
|
||||
self.required_phrases=required_phrases
|
||||
self.wait_time=wait_time
|
||||
self.max_tries=max_tries
|
||||
|
||||
self.detected_phrases = []
|
||||
|
||||
def setGrammar(self, grammar):
|
||||
"""
|
||||
@param grammar - instance of grammar class
|
||||
"""
|
||||
self.grammar=grammar
|
||||
self.speech_detect.addGrammar(grammar)
|
||||
self.speech_detect.setGrammar(self.grammar.name)
|
||||
|
||||
def detectSpeech(self):
|
||||
self.speech_detect.detectSpeech()
|
||||
|
||||
def run(self):
|
||||
"""
|
||||
start speech detection with the current grammar,
|
||||
and listen for results from asr engine. once a result
|
||||
has been returned, return it to caller
|
||||
"""
|
||||
|
||||
def dtmf_handler(input, itype, funcargs):
|
||||
console_log("INFO","\n\nDTMF itype: %s\n" % itype)
|
||||
if itype == 1: # TODO!! use names for comparison instead of number
|
||||
return self.handle_event(input, funcargs)
|
||||
elif itype== 0:
|
||||
console_log("INFO","\n\nDTMF input: %s\n" % input)
|
||||
else:
|
||||
console_log("INFO","\n\nUnknown input type: %s\n" % itype)
|
||||
return None
|
||||
|
||||
|
||||
num_tries = 0
|
||||
|
||||
session = self.speech_detect.session
|
||||
|
||||
console_log("debug", "setting dtmf callback\n")
|
||||
session.setDTMFCallback(dtmf_handler, "")
|
||||
console_log("debug", "calling getDigits\n")
|
||||
|
||||
console_log("debug", "starting run() while loop\n")
|
||||
while (session.ready() and
|
||||
num_tries < self.max_tries and
|
||||
len(self.detected_phrases) < self.required_phrases):
|
||||
console_log("debug", "top of run() while loop\n")
|
||||
session.collectDigits(self.wait_time)
|
||||
num_tries += 1
|
||||
|
||||
console_log("debug", "while loop finished\n")
|
||||
return self.detected_phrases
|
||||
|
||||
def handle_event(self, event, funcargs):
|
||||
"""
|
||||
when the dtmf handler receives an event, it calls back
|
||||
this method. event is a dictionary with subdictionaries ..
|
||||
|
||||
Example 1
|
||||
=========
|
||||
|
||||
{'body': None, 'headers': {'Speech-Type': 'begin-speaking'}}
|
||||
|
||||
Example 2
|
||||
=========
|
||||
{'body': '<result xmlns='http://www.ietf.org/xml/ns/mrcpv2'
|
||||
xmlns:ex='http://www.example.com/example' score='100'
|
||||
grammar='session:request1@form-level.store'><interpretation>
|
||||
<input mode='speech'>waffles</input></interpretation></result>',
|
||||
'headers': {'Speech-Type': 'detected-speech'}}
|
||||
|
||||
This dictionary is constructed in run_dtmf_callback() in
|
||||
freeswitch_python.cpp
|
||||
|
||||
"""
|
||||
|
||||
# what kind of event?
|
||||
headers = event['headers']
|
||||
speech_type = headers['Speech-Type']
|
||||
if speech_type == "begin-speaking":
|
||||
# not sure what to do with this, try returning "stop"
|
||||
# so that it might stop playing a sound file once
|
||||
# speech has been detected
|
||||
return "stop"
|
||||
elif speech_type == "detected-speech":
|
||||
# extract the detected phrase. from result
|
||||
# BUG: this assumes only ONE interpretation in the xml
|
||||
# result. rest will get igored
|
||||
# NOTE: have to wrap everything with str() (at least
|
||||
# calls to console_log because otherwise it chokes on
|
||||
# unicode strings.
|
||||
# TODO: check the score
|
||||
body = event['body']
|
||||
dom = minidom.parseString(body)
|
||||
phrase = dom.getElementsByTagName(self.grammar.obj_path)[0]
|
||||
phrase_text = self.getText(phrase)
|
||||
if phrase_text:
|
||||
self.detected_phrases.append(str(phrase_text))
|
||||
# do we want to return stop? what should we return?
|
||||
return "stop"
|
||||
else:
|
||||
raise Exception("Unknown speech event: %s" % speech_type)
|
||||
|
||||
|
||||
def getText(self, elt):
|
||||
|
||||
""" given an element, get its text. if there is more than
|
||||
one text node child, just append all the text together.
|
||||
"""
|
||||
|
||||
result = ""
|
||||
children = elt.childNodes
|
||||
for child in children:
|
||||
if child.nodeType == child.TEXT_NODE:
|
||||
result += str(child.nodeValue)
|
||||
return result
|
||||
|
||||
|
|
@ -0,0 +1,87 @@
|
|||
from freeswitch import *
|
||||
from py_modules.speechtools import Grammar, SpeechDetect
|
||||
from py_modules.speechtools import SpeechObtainer
|
||||
|
||||
import time, os
|
||||
|
||||
VOICE_ENGINE = "cepstral"
|
||||
VOICE = "William"
|
||||
GRAMMAR_ROOT = "/usr/src/freeswitch_trunk/scripts"
|
||||
|
||||
"""
|
||||
Example speech recognition application in python.
|
||||
|
||||
How to make this work:
|
||||
|
||||
* Get mod_openmrcp working along with an MRCP asr server
|
||||
* Add /usr/src/freeswitch/scripts or equivalent to your PYTHONPATH
|
||||
* Restart freeswitch
|
||||
* Create $GRAMMAR_ROOT/mainmenu.xml from contents in mainmenu() comments
|
||||
|
||||
"""
|
||||
|
||||
class RecipeWizard:
|
||||
|
||||
def __init__(self, session):
|
||||
self.session=session
|
||||
self.session.set_tts_parms(VOICE_ENGINE, VOICE)
|
||||
self.main()
|
||||
|
||||
def main(self):
|
||||
|
||||
console_log("debug", "recipe wizard main()\n")
|
||||
self.speechdetect = SpeechDetect(self.session, "openmrcp", "127.0.0.1");
|
||||
self.speechobtainer = SpeechObtainer(speech_detect=self.speechdetect,
|
||||
required_phrases=1,
|
||||
wait_time=5000,
|
||||
max_tries=3)
|
||||
gfile = os.path.join(GRAMMAR_ROOT, "mainmenu.xml")
|
||||
self.grammar = Grammar("mainmenu", gfile,"input",80,90)
|
||||
self.speechobtainer.setGrammar(self.grammar);
|
||||
console_log("debug", "calling speechobtainer.run()\n")
|
||||
self.speechobtainer.detectSpeech()
|
||||
self.session.speak("Hello. Welcome to the recipe wizard. Drinks or food?")
|
||||
result = self.speechobtainer.run()
|
||||
console_log("debug", "speechobtainer.run() result: %s\n" % result)
|
||||
if result:
|
||||
self.session.speak("Received result. Result is: %s" % result[0])
|
||||
else:
|
||||
self.session.speak("Sorry, I did not hear you")
|
||||
|
||||
console_log("debug", "speechobtainer.run() finished\n")
|
||||
|
||||
def mainmenu():
|
||||
"""
|
||||
<!DOCTYPE grammar PUBLIC "-//W3C//DTD GRAMMAR 1.0//EN"
|
||||
"http://www.w3.org/TR/speech-grammar/grammar.dtd">
|
||||
|
||||
<grammar xmlns="http://www.w3.org/2001/06/grammar" xml:lang="en"
|
||||
xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
|
||||
xsi:schemaLocation="http://www.w3.org/2001/06/grammar
|
||||
http://www.w3.org/TR/speech-grammar/grammar.xsd"
|
||||
version="1.0" mode="voice" root="root">
|
||||
|
||||
|
||||
<rule id="root" scope="public">
|
||||
|
||||
<rule id="main">
|
||||
<one-of>
|
||||
<item weight="10">drinks</item>
|
||||
<item weight="2">food</item>
|
||||
</one-of>
|
||||
</rule>
|
||||
|
||||
</rule>
|
||||
|
||||
</grammar>
|
||||
|
||||
"""
|
||||
pass
|
||||
|
||||
def handler(uuid):
|
||||
session = PySession(uuid)
|
||||
session.answer()
|
||||
rw = RecipeWizard(session)
|
||||
session.hangup("1")
|
||||
|
||||
|
||||
Loading…
Reference in New Issue