import os
import subprocess
import sys
import tempfile
from contextlib import contextmanager
from util.paths import get_binary
BINARY_PATH = os.path.join("ffmpeg",
"ffmpeg.exe" if os.name == "nt" else "ffmpeg")
FFMPEG = get_binary(BINARY_PATH)
def resample(infile, outfile):
if not os.path.isfile(infile):
raise IOError("Not a file: %s" % infile)
'''
Use FFMPEG to convert a media file to a wav file sampled at 8K
'''
return subprocess.call([
FFMPEG, '-loglevel', 'panic', '-y', '-i', infile, '-ac', '1', '-ar',
'8000', '-acodec', 'pcm_s16le', outfile
],
cwd=os.path.dirname(sys.executable))
@contextmanager
def resampled(infile):
with tempfile.NamedTemporaryFile(suffix='.wav') as fp:
if resample(infile, fp.name) != 0:
'''Glue code for communicating with standard_kaldi C++ process'''
import json
import logging
import os
import subprocess
import tempfile
import wave
from util.paths import get_binary
from gentle.rpc import RPCProtocol
from gentle.resources import Resources
EXECUTABLE_PATH = get_binary("ext/standard_kaldi")
class Kaldi(object):
'''Kaldi spawns a standard_kaldi subprocess and provides a
Python wrapper for communicating with it.'''
def __init__(self, nnet_dir, hclg_path, proto_langdir):
self.proto_langdir = proto_langdir
devnull = open(os.devnull, 'w')
cmd = [EXECUTABLE_PATH, nnet_dir, hclg_path, proto_langdir]
self._subprocess = subprocess.Popen(cmd,
stdin=subprocess.PIPE,
stdout=subprocess.PIPE,
stderr=devnull)
self._transitions = None
self._words = None
self._stopped = False
import os
import subprocess
import tempfile
import wave
from shutil import copyfile
from contextlib import contextmanager
from util.paths import get_binary
FFMPEG = get_binary("ffmpeg")
def resample(infile, outfile):
if not os.path.isfile(infile):
raise IOError("Not a file: %s" % infile)
'''
Use FFMPEG to convert a media file to a wav file sampled at 8K
'''
# Check if the file already meets our specifications
try:
wave_file = wave.open(infile, 'rb')
if wave_file.getframerate() == 8000:
copyfile(infile, outfile)
return 0
except wave.Error as err:
pass
return subprocess.call([
FFMPEG, '-loglevel', 'panic', '-y', '-i', infile, '-ac', '1', '-ar',
'8000', '-acodec', 'pcm_s16le', outfile
])
import subprocess
from util.paths import get_binary
import os
from resources import Config
EXECUTABLE_PATH = get_binary("ext/k3")
class Kaldi:
def __init__(self, nnet_dir=None, hclg_path=None, proto_langdir=None):
devnull = open(os.devnull, 'w')
cmd = [EXECUTABLE_PATH]
self.config = Config()
if nnet_dir is not None:
cmd.append(nnet_dir)
cmd.append(hclg_path)
cmd.append(self.config['silencephones'])
self._p = subprocess.Popen(cmd,
stdin=subprocess.PIPE,
stdout=subprocess.PIPE,
stderr=devnull)
self.finished = False
def _cmd(self, c):
self._p.stdin.write("%s\n" % (c))
self._p.stdin.flush()
import subprocess
from util.paths import get_binary
import os
BINARY_PATH = "ext\\gentleK3.exe" if os.name == "nt" else "ext/gentleK3"
EXECUTABLE_PATH = get_binary(BINARY_PATH)
class Kaldi:
def __init__(self, nnet_dir=None, hclg_path=None, proto_langdir=None):
devnull = open(os.devnull, 'wb')
cmd = [EXECUTABLE_PATH]
if nnet_dir is not None:
cmd.append(nnet_dir)
cmd.append(hclg_path)
self._p = subprocess.Popen(cmd,
stdin=subprocess.PIPE, stdout=subprocess.PIPE,
stderr=devnull)
self.finished = False
def _cmd(self, c):
self._p.stdin.write("%s\n" % (c))
self._p.stdin.flush()
def push_chunk(self, buf):
# Wait until we're ready
self._cmd("push-chunk")
cnt = len(buf)/2
import logging
import math
import os
import shutil
import subprocess
import sys
import tempfile
from util.paths import get_binary
from metasentence import MetaSentence
from resources import Resources
MKGRAPH_PATH = get_binary("ext/mkgraph")
def make_bigram_lm_fst(word_sequences, **kwargs):
'''
Use the given token sequence to make a bigram language model
in OpenFST plain text format.
When the "conservative" flag is set, an [oov] is interleaved
between successive words.
When the "disfluency" flag is set, a small set of disfluencies is
interleaved between successive words
`Word sequence` is a list of lists, each valid as a start
'''
if len(word_sequences) == 0 or type(word_sequences[0]) != list:
word_sequences = [word_sequences]
import logging
import math
import os
import shutil
import subprocess
import sys
import tempfile
from util.paths import get_binary
from metasentence import MetaSentence
from resources import Resources
BINARY_PATH = "ext\\gentleM3.exe" if os.name == "nt" else "ext/gentleM3"
MKGRAPH_PATH = get_binary(BINARY_PATH)
# [oov] no longer in words.txt
OOV_TERM = '<unk>'
def make_bigram_lm_fst(word_sequences, **kwargs):
'''
Use the given token sequence to make a bigram language model
in OpenFST plain text format.
When the "conservative" flag is set, an [oov] is interleaved
between successive words.
When the "disfluency" flag is set, a small set of disfluencies is
interleaved between successive words
`Word sequence` is a list of lists, each valid as a start
import logging
import math
import os
import shutil
import subprocess
import sys
import tempfile
from util.paths import get_binary
from metasentence import MetaSentence
from resources import Resources
MKGRAPH_PATH = get_binary("ext/m3")
# [oov] no longer in words.txt
OOV_TERM = '<unk>'
def make_bigram_lm_fst(word_sequences, **kwargs):
'''
Use the given token sequence to make a bigram language model
in OpenFST plain text format.
When the "conservative" flag is set, an [oov] is interleaved
between successive words.
When the "disfluency" flag is set, a small set of disfluencies is
interleaved between successive words
`Word sequence` is a list of lists, each valid as a start
'''
