Ceci est une ancienne révision du document !

Cette page est en cours de rédaction

Installation

à partir des dépôts

Les paquet présent dans raring semblent ne pas fonctionner

Installer les paquets python-pocketsphinx libpocketsphinx1 gstreamer0.10-pocketsphinx python-pyaudio 

apt-get install python-pocketsphinx libpocketsphinx1 gstreamer0.10-pocketsphinx python-pyaudio

à partir des sources

D'abord il faut installer les dépendances 

sudo apt-get build-dep pocketsphinx
sudo apt-get install bison python-pyaudio

sphinxbase

On récupère ensuite les sources de sphinxbase-0.8 

wget -O sphinxbase-0.8.tar.gz http://sourceforge.net/projects/cmusphinx/files/sphinxbase/0.8/sphinxbase-0.8.tar.gz/download

On les décompresse 

tar -xvzf sphinxbase-0.8.tar.gz

On compile

Si vous avez l'erreur : 
    import pocketsphinx as ps
  File "sphinxbase.pxd", line 150, in init pocketsphinx (pocketsphinx.c:7935)
ValueError: PyCapsule_GetPointer called with invalid PyCapsule object

il faut régénérer le fichier python/sphinxbase.c ref 

sudo apt-get install cython
mv python/sphinxbase.c ~

puis relancer la compilation a partir de ./configure

Par défaut au moment de la configuration de sphinxbase, si les librairies de compilation de pulseaudio sont installées elles seront utilisées. Si comme moi vous devez utiliser Alsa, il faut supprimer la librairie pulseaudio : 
sudo apt-get remove libpulse-dev

./configure --prefix=/usr/local
make
sudo make install

pocketsphinx

wget -O pocketsphinx-0.8.tar.gz http://sourceforge.net/projects/cmusphinx/files/pocketsphinx/0.8/pocketsphinx-0.8.tar.gz/download
tar -xvzf pocketsphinx-0.8.tar.gz
cd pocketsphinx-0.8
./configure --prefix=/usr/local
make
sudo make install

modèles français

Téléchargement: 

wget -O lium_french_f0.tar.gz http://sourceforge.net/projects/cmusphinx/files/Acoustic%20and%20Language%20Models/French%20F0%20Broadcast%20News%20Acoustic%20Model/lium_french_f0.tar.gz/download
tar -xvzf lium_french_f0.tar.gz
cd lium_french_f0/
sudo mkdir /usr/share/pocketsphinx/model/FR/
sudo mv * /usr/share/pocketsphinx/model/FR/

wget -O french3g62K.lm.dmp.bz2 http://sourceforge.net/projects/cmusphinx/files/Acoustic%20and%20Language%20Models/French%20Language%20Model/french3g62K.lm.dmp.bz2/download
bzip2 -d french3g62K.lm.dmp.bz2
sudo mv french3g62K.lm.dmp /usr/share/pocketsphinx/model/FR/

wget -O frenchWords62K.dic http://sourceforge.net/projects/cmusphinx/files/Acoustic%20and%20Language%20Models/French%20Language%20Model/frenchWords62K.dic/download
sudo mv frenchWords62K.dic /usr/share/pocketsphinx/model/FR/


wget -O lium_french_f2.tar.gz http://sourceforge.net/projects/cmusphinx/files/Acoustic%20and%20Language%20Models/French%20F2%20Telephone%20Acoustic%20Model/lium_french_f2.tar.gz/download

Bon alors, est ce que ça marche ?

pocketsphinx_continuous

pocketsphinx_continuous -dict /usr/share/pocketsphinx/model/FR/frenchWords62K.dic -hmm /usr/share/pocketsphinx/model/FR/ -lm /usr/share/pocketsphinx/model/FR/french3g62K.lm.dmp

scripts python

lister tous les périphériques audio

ref 

#!/usr/bin/python
"""
PyAudio Example:
 
Query and print PortAudio HostAPIs, Devices, and their
support rates.
"""
 
import pyaudio
 
standard_sample_rates = [8000.0, 9600.0, 11025.0, 12000.0,
                         16000.0, 22050.0, 24000.0, 32000.0,
                         44100.0, 48000.0, 88200.0, 96000.0,
                         192000.0]
 
p = pyaudio.PyAudio()
max_apis = p.get_host_api_count()
max_devs = p.get_device_count()
 
print("\nPortAudio System Info:\n======================")
print("Version: %d" % pyaudio.get_portaudio_version())
print("Version Text: %s" % pyaudio.get_portaudio_version_text())
print("Number of Host APIs: %d" % max_apis)
print("Number of Devices  : %d" % max_devs)
 
print("\nHost APIs:\n==========")
 
for i in range(max_apis):
    apiinfo = p.get_host_api_info_by_index(i)
    for k in list(apiinfo.items()):
        print("%s: %s" % k)
    print("--------------------------")
 
print("\nDevices:\n========")
 
for i in range(max_devs):
    devinfo = p.get_device_info_by_index(i)
 
    # print out device parameters
    for k in list(devinfo.items()):
        name, value = k
 
        # if host API, then get friendly name
 
        if name == 'hostApi':
            value = str(value) + \
                    " (%s)" % p.get_host_api_info_by_index(k[1])['name']
        print("\t%s: %s" % (name, value))
 
    # print out supported format rates
 
    input_supported_rates = []
    output_supported_rates = []
    full_duplex_rates = []
 
    for f in standard_sample_rates:
 
        if devinfo['maxInputChannels'] > 0:
            try:
                if p.is_format_supported(
                    f,
                    input_device = devinfo['index'],
                    input_channels = devinfo['maxInputChannels'],
                    input_format = pyaudio.paInt16):
                    input_supported_rates.append(f)
            except ValueError:
                pass
 
        if devinfo['maxOutputChannels'] > 0:
            try:
                if p.is_format_supported(
                    f,
                    output_device = devinfo['index'],
                    output_channels = devinfo['maxOutputChannels'],
                    output_format = pyaudio.paInt16):
                    output_supported_rates.append(f)
            except ValueError:
                pass
 
        if (devinfo['maxInputChannels'] > 0) and \
           (devinfo['maxOutputChannels'] > 0):
            try:
                if p.is_format_supported(
                    f,
                    input_device = devinfo['index'],
                    input_channels = devinfo['maxInputChannels'],
                    input_format = pyaudio.paInt16,
                    output_device = devinfo['index'],
                    output_channels = devinfo['maxOutputChannels'],
                    output_format = pyaudio.paInt16):
                    full_duplex_rates.append(f)
            except ValueError:
                pass
 
    if len(input_supported_rates):
        print("\tInput rates: %s" % input_supported_rates)
    if len(output_supported_rates):
        print("\tOutput rates: %s" % output_supported_rates)
    if len(full_duplex_rates):
        print("\tFull duplex: %s" % full_duplex_rates)
 
    print("\t--------------------------------")
 
print("\nDefault Devices:\n================")
try:
    def_index = p.get_default_input_device_info()['index']
    print("Default Input Device : %s" % def_index)
    devinfo = p.get_device_info_by_index(def_index)
    for k in list(devinfo.items()):
        name, value = k
        if name == 'hostApi':
            value = str(value) + \
                    " (%s)" % p.get_host_api_info_by_index(k[1])['name']
        print("\t%s: %s" % (name, value))
    print("\t--------------------------------")
except IOError as e:
    print("No Input devices: %s" % e[0])
 
try:
    def_index = p.get_default_output_device_info()['index']
    print("Default Output Device: %s" % def_index)
    devinfo = p.get_device_info_by_index(def_index)
    for k in list(devinfo.items()):
        name, value = k
        if name == 'hostApi':
            value = str(value) + \
                    " (%s)" % p.get_host_api_info_by_index(k[1])['name']
        print("\t%s: %s" % (name, value))
    print("\t--------------------------------")
except IOError as e:
    print("No Output devices: %s" % e[0])
 
p.terminate()

essayer de faire une reconnaissance vocale

#!/usr/bin/python
 
import sys,os
import pyaudio
import wave
 
hmdir = "/usr/share/pocketsphinx/model/FR/"
lmd = "/usr/share/pocketsphinx/model/FR/french3g62K.lm.dmp"
dictd = "/usr/share/pocketsphinx/model/FR/frenchWords62K.dic"
 
def decodeSpeech(hmmd,lmdir,dictp,wavfile):
 
    import pocketsphinx as ps
    import sphinxbase
 
    speechRec = ps.Decoder(hmm = hmmd, lm = lmdir, dict = dictp)
    wavFile = file(wavfile,'rb')
    wavFile.seek(44)
    speechRec.decode_raw(wavFile)
    result = speechRec.get_hyp()
 
    return result[0]
 
#CHUNK = 1024
CHUNK = 512
#FORMAT = pyaudio.paInt16
FORMAT = pyaudio.paALSA
CHANNELS = 1
RATE = 16000
#RATE = 44100
RECORD_SECONDS = 10
 
for x in range(10):
    fn = "o"+str(x)+".wav"
    p = pyaudio.PyAudio()
    stream = p.open(format=FORMAT, channels=CHANNELS, rate=RATE, input=True, frames_per_buffer=CHUNK)
    print("* recording")
    frames = []
    print str(RATE / CHUNK * RECORD_SECONDS) + " size\n"
    for i in range(0, int(RATE / CHUNK * RECORD_SECONDS)):
        data = stream.read(CHUNK)
        frames.append(data)
    print("* done recording")
    stream.stop_stream()
    stream.close()
    wf = wave.open(fn, 'wb')
    wf.setnchannels(CHANNELS)
    wf.setsampwidth(p.get_sample_size(FORMAT))
    p.terminate()
    wf.setframerate(RATE)
    wf.writeframes(b''.join(frames))
    wf.close()
    wavfile = fn
    recognised = decodeSpeech(hmdir,lmd,dictd,wavfile)
    print recognised
    cm = 'espeak "'+recognised+'"'
    os.system(cm)

références

Site officiel : http://cmusphinx.sourceforge.net/wiki/download/

Python audio : http://people.csail.mit.edu/hubert/pyaudio/docs/

Exemple en python: http://pythonism.wordpress.com/2013/06/06/pocketsphinx-voice-recognition-with-python/

http://stackoverflow.com/questions/8567366/geting-ioerror-errno-input-overflowed-9981-when-setting-pyaudio-stream-input

  • pocketsphinx.1389475014.txt.gz
  • Dernière modification: Le 11/01/2014, 22:16
  • par mmaura