def compute_fbank(target,rate=16000,frameWidth=25,frameShift=10,
melBins=23,windowType='povey',useSuffix=None,
config=None,name="fbank",outFile=None):
'''
Compute fbank feature.
Share Args:
Null
Parallel Args:
<target>: wave file,scp file,exkaldi ListTable object or WavSegment object. If it is wave file,we will use it's file name as utterance ID.
<rate>: sample rate.
<frameWidth>: windows width (ms).
<frameShift>: shift windows width (ms).
<melbins>: the numbers of mel filter banks.
<windowType>: windows type.
<useSuffix>: If the suffix of file is not .scp or .wav,use this to specify it.
<config>: extra optional configurations.
<name>: the name of output feature.
<outFile>: output file name.
Some usual options can be assigned directly. If you want use more,set <config> = your-configure.
You can use exkaldi.check_config('compute_fbank') function to get the reference of extra configurations.
Also you can run shell command "compute-fbank-feats" to look their usage.
Return:
exkaldi feature or index table object.
'''
# check the basis configure parameters to build base commands
stdParameters = check_multiple_resources(rate,frameWidth,frameShift,melBins,windowType,config)
baseCmds = []
for rate,frameWidth,frameShift,melBins,windowType,config,_ in zip(*stdParameters):
declare.is_positive_int("rate",rate)
declare.is_positive_int("frameWidth",frameWidth)
declare.is_positive_int("frameShift",frameShift)
declare.is_positive_int("melBins",melBins)
declare.greater_equal("frameWidth",frameWidth,"frameShift",frameShift)
declare.is_instances("windowType",windowType,["hamming","hanning","povey","rectangular","blackmann"])
kaldiTool = 'compute-fbank-feats --allow-downsample --allow-upsample '
kaldiTool += f'--sample-frequency={rate} '
kaldiTool += f'--frame-length={frameWidth} '
kaldiTool += f'--frame-shift={frameShift} '
kaldiTool += f'--num-mel-bins={melBins} '
kaldiTool += f'--window-type={windowType} '
if config is not None:
if check_config(name='compute_fbank',config=config):
for key,value in config.items():
if isinstance(value,bool):
if value is True:
kaldiTool += f"{key} "
else:
kaldiTool += f"{key}={value} "
baseCmds.append(kaldiTool)
# run the common function
return __compute_feature(target,baseCmds,useSuffix,name,outFile)
def compute_fbank(wavFile,
rate=16000,
frameWidth=25,
frameShift=10,
melBins=23,
windowType='povey',
useSuffix=None,
config=None,
name="fbank"):
'''
Compute fbank feature.
Args:
<wavFile>: wave file or scp file or exkaldi SriptTable object. If it is wave file, use it's file name as utterance ID.
<rate>: sample rate.
<frameWidth>: sample windows width.
<frameShift>: shift windows width.
<melbins>: the numbers of mel filter banks.
<windowType>: sample windows type.
<useSuffix>: If the suffix of file is not .scp, use this to specify it.
<config>: use this to assign more configures. If use it, all these configures above will be skipped.
<name>: the name of feature.
Some usual options can be assigned directly. If you want use more, set <config> = your-configure, but if you do this, these usual configures we provided will be ignored.
You can use .check_config('compute_fbank') function to get configure information that you can set.
Also you can run shell command "compute-fbank-feats" to look their meaning.
Return:
A exkaldi bytes feature object.
'''
assert isinstance(
frameWidth, int
) and frameWidth > 0, f"<frameWidth> should be a positive int value but got {type_name(frameWidth)}."
assert isinstance(
frameShift, int
) and frameShift > 0, f"<frameShift> should be a positive int value but got {type_name(frameShift)}."
assert frameWidth > frameShift, f"<frameWidth> and <frameShift> is unavaliable."
assert windowType in [
"hamming", "hanning", "povey", "rectangular", "blackmann"
], f'<windowType> should be "hamming","hanning","povey","rectangular","blackmann", but got: {windowType}.'
kaldiTool = 'compute-fbank-feats --allow-downsample --allow-upsample '
kaldiTool += f'--sample-frequency={rate} '
kaldiTool += f'--frame-length={frameWidth} '
kaldiTool += f'--frame-shift={frameShift} '
kaldiTool += f'--num-mel-bins={melBins} '
kaldiTool += f'--window-type={windowType} '
if config is not None:
if check_config(name='compute_fbank', config=config):
for key, value in config.items():
if isinstance(value, bool):
if value is True:
kaldiTool += f"{key} "
else:
kaldiTool += f" {key}={value}"
result = __compute_feature(wavFile, kaldiTool, useSuffix, name)
return result
def gmm_decode(feat, hmm, HCLGFile, wordSymbolTable, beam=10, latBeam=8, acwt=1,
minActive=200, maxActive=7000, maxMem=50000000, config=None, maxThreads=1):
'''
Decode by generating lattice from feature and GMM model.
Args:
<feat>: An exkaldi feature object.
<hmm>: An exkaldi HMM object or file path.
<HCLGFile>: HCLG file path.
<wordSymbolTable>: words.txt file path or exkaldi LexiconBank object or exkaldi ListTable object.
<beam>: beam size.
<latBeam>: lattice beam size.
<acwt>: acoustic model weight.
<minActivate>: .
<maxActive>: .
<maxMem>: .
<config>: decode configure file.
<maxThreads>: the number of mutiple threads.
Some usual options can be assigned directly. If you want use more, set <config> = your-configure, but if you do this, these usual configures we provided will be ignored.
You can use .check_config('gmm_decode') function to get configure information you could set.
Also run shell command "gmm-latgen-faster" to look their meaning.
Return:
An exkaldi Lattice object.
'''
ExkaldiInfo.vertify_kaldi_existed()
if type_name(feat) == "BytesFeature":
pass
elif type_name(feat) == "NumpyFeature":
feat = feat.to_bytes()
else:
raise UnsupportedType(f"Expected <feat> is an exkaldi feature object but got: {type_name(feat)}.")
assert isinstance(HCLGFile, str), "<HCLGFile> should be a file path."
if not os.path.isfile(HCLGFile):
raise WrongPath(f"No such file:{HCLGFile}")
if maxThreads > 1:
kaldiTool = f"gmm-latgen-faster-parallel --num-threads={maxThreads} "
else:
kaldiTool = "gmm-latgen-faster "
kaldiTool += f'--allow-partial=true '
kaldiTool += f'--min-active={minActive} '
kaldiTool += f'--max-active={maxActive} '
kaldiTool += f'--max_mem={maxMem} '
kaldiTool += f'--beam={beam} '
kaldiTool += f'--lattice-beam={latBeam} '
kaldiTool += f'--acoustic-scale={acwt} '
wordsTemp = tempfile.NamedTemporaryFile("w+", suffix="_words.txt", encoding="utf-8")
modelTemp = tempfile.NamedTemporaryFile("wb+", suffix=".mdl")
try:
if type_name(wordSymbolTable) == "LexiconBank":
wordSymbolTable.dump_dict("words", wordsTemp)
wordsFile = wordsTemp.name
elif type_name(wordSymbolTable) == "ListTable":
wordSymbolTable.save(wordsTemp)
wordsTemp.seek(0)
wordsFile = wordsTemp.name
elif isinstance(wordSymbolTable, str):
if not os.path.isfile(wordSymbolTable):
raise WrongPath(f"No such file:{wordSymbolTable}.")
else:
wordsFile = wordSymbolTable
else:
raise UnsupportedType(f"<wordSymbolTable> should be a file path or exkaldi LexiconBank object but got {type_name(wordSymbolTable)}.")
kaldiTool += f'--word-symbol-table={wordsFile} '
if config is not None:
if check_config(name='gmm_decode', config=config):
for key,value in config.items():
if isinstance(value, bool):
if value is True:
kaldiTool += f"{key} "
else:
kaldiTool += f" {key}={value}"
if type_name(hmm) in ["MonophoneHMM", "TriphoneHMM"]:
modelTemp.write(hmm.data)
modelTemp.seek(0)
hmmFile = modelTemp.name
elif isinstance(hmm, str):
if not os.path.isfile(hmm):
raise WrongPath(f"No such file:{hmm}.")
else:
hmmFile = hmm
else:
raise UnsupportedType(f"<hmm> should be exkaldi HMM object or file path but got {type_name(hmm)}.")
cmd = f'{kaldiTool} {hmmFile} {HCLGFile} ark:- ark:-'
out, err, cod = run_shell_command(cmd, stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.PIPE, inputs=feat.data)
if cod != 0 or out == b'':
print(err.decode())
raise KaldiProcessError('Failed to generate lattice.')
else:
newName = f"lat({feat.name})"
return Lattice(data=out, name=newName)
finally:
wordsTemp.close()
modelTemp.close()
def train_ngrams_srilm(lexicons, order, text, outFile, config=None):
'''
Train N-Grams language model with SriLM tookit.
If you don't specified the discount by the <config> option, We defaultly use "kndiscount".
Args:
<lexicons>: an exkaldi LexiconBank object.
<order>: the maximum order of N-Grams.
<text>: a text corpus file or an exkaldi transcription object.
<outFile>: output file name of arpa LM.
<config>: extra configurations, a Python dict object.
You can use .check_config("train_ngrams_srilm") function to get a reference of extra configurations.
Also you can run shell command "ngram-count" to look their usage.
'''
declare.is_lexicon_bank("lexicons", lexicons)
declare.is_positive_int("order", order)
declare.is_potential_transcription("text", text)
declare.is_valid_file_name("outFile", outFile)
# verify the max order
declare.less_equal("order", order, "max order", 9)
# prepare srilm tool
ExkaldiInfo.prepare_srilm()
with FileHandleManager() as fhm:
# check whether this is a reasonable text corpus that should be splited by space.
if isinstance(text, str):
cmd = f"shuf {text} -n 100"
out, err, cod = run_shell_command(cmd,
stdout="PIPE",
stderr="PIPE")
if (isinstance(cod, int) and cod != 0):
print(err.decode())
raise ShellProcessError(
f"Failed to sample from text file:{text}.")
elif out == b'':
raise WrongDataFormat(f"Void text file:{text}.")
else:
out = out.decode().strip().split("\n")
spaceCount = 0
for line in out:
spaceCount += line.count(" ")
if spaceCount < len(out) // 2:
raise WrongDataFormat(
"The text file doesn't seem to be separated by spaces or sentences are extremely short."
)
else:
sampleText = text.subset(nRandom=100)
spaceCount = 0
for key, value in sampleText.items():
assert isinstance(
value, str
), f"Transcription must be string but got: {type_name(value)}."
spaceCount += value.count(" ")
if spaceCount < len(sampleText) // 2:
raise WrongDataFormat(
"The text file doesn't seem to be separated by spaces or sentences are extremely short."
)
textTemp = fhm.create("a+", suffix=".txt", encoding="utf-8")
text.save(textTemp, discardUttID=True)
text = textTemp.name
unkSymbol = lexicons("oov")
wordlistTemp = fhm.create("w+", encoding='utf-8', suffix=".txt")
words = lexicons("words")
words = "\n".join(words.keys())
wordlistTemp.write(words)
wordlistTemp.seek(0)
extraConfig = " "
specifyDiscount = False
if config is not None:
if check_config(name='train_ngrams_srilm', config=config):
for key, value in config.items():
if isinstance(value, bool):
if value is True:
extraConfig += f"{key} "
if key.endswith("discount"):
specifyDiscount = True
else:
extraConfig += f" {key} {value}"
cmd = f'ngram-count -text {text} -order {order} -limit-vocab -vocab {wordlistTemp.name} -unk -map-unk "{unkSymbol}" '
if specifyDiscount is False:
cmd += "-kndiscount "
cmd += "-interpolate "
if not outFile.rstrip().endswith(".arpa"):
outFile += ".arpa"
make_dependent_dirs(outFile, pathIsFile=True)
cmd += f" -lm {outFile}"
out, err, cod = run_shell_command(cmd, stderr="PIPE")
if (isinstance(cod, int) and cod != 0) or (
not os.path.isfile(outFile)) or os.path.getsize(outFile) == 0:
print(err.decode())
if os.path.isfile(outFile):
os.remove(outFile)
raise KaldiProcessError(
f'Failed to generate N-Grams language model.')
return outFile
def train_ngrams_kenlm(lexicons, order, text, outFile, config=None):
'''
Train N-Grams language model with SriLM tookit.
Args:
<lexicons>: an exkaldi LexiconBank object.
<order>: the maximum order of N-Grams.
<text>: a text corpus file or an exkaldi transcription object.
<outFile>: output file name of arpa LM.
<config>: extra configurations, a Python dict object.
You can use .check_config("train_ngrams_kenlm") function to get a reference of extra configurations.
Also you can run shell command "lmplz" to look their usage.
'''
declare.is_lexicon_bank("lexicons", lexicons)
declare.is_positive_int("order", order)
declare.is_potential_transcription("text", text)
declare.is_valid_file_name("outFile", outFile)
declare.less_equal("order", order, "max order", 9)
with FileHandleManager() as fhm:
# check whether this is a reasonable text corpus that should be splited by space.
if isinstance(text, str):
cmd = f"shuf {text} -n 100"
out, err, cod = run_shell_command(cmd,
stdout="PIPE",
stderr="PIPE")
if (isinstance(cod, int) and cod != 0):
print(err.decode())
raise ShellProcessError(
f"Failed to sample from text file:{text}.")
elif out == b'':
raise WrongDataFormat(f"Void text file:{text}.")
else:
out = out.decode().strip().split("\n")
spaceCount = 0
for line in out:
spaceCount += line.count(" ")
if spaceCount < len(out) // 2:
raise WrongDataFormat(
"The text file doesn't seem to be separated by spaces or sentences are extremely short."
)
else:
sampleText = text.subset(nRandom=100)
spaceCount = 0
for key, value in sampleText.items():
assert isinstance(
value, str
), f"Transcription must be string but got: {type_name(value)}."
spaceCount += value.count(" ")
if spaceCount < len(sampleText) // 2:
raise WrongDataFormat(
"The text file doesn't seem to be separated by spaces or sentences are extremely short."
)
textTemp = fhm.create("a+", suffix=".txt", encoding="utf-8")
text.save(textTemp, discardUttID=True)
text = textTemp.name
extraConfig = " "
if config is not None:
if check_config(name='train_ngrams_kenlm', config=config):
if "--temp_prefix" in config.keys() and "-T" in config.keys():
raise WrongOperation(
f'"--temp_prefix" and "-T" is the same configuration so only one of them is expected.'
)
if "--memory" in config.keys() and "-S" in config.keys():
raise WrongOperation(
f'"--memory" and "-S" is the same configuration so only one of them is expected.'
)
for key, value in config.items():
if isinstance(value, bool):
if value is True:
extraConfig += f"{key} "
else:
extraConfig += f"{key} {value} "
if not outFile.rstrip().endswith(".arpa"):
outFile += ".arpa"
make_dependent_dirs(outFile, pathIsFile=True)
wordlistTemp = fhm.create("w+", encoding='utf-8', suffix=".txt")
words = lexicons("words")
words_count = math.ceil(len(words) / 10) * 10
words = "\n".join(words.keys())
wordlistTemp.write(words)
wordlistTemp.seek(0)
KenLMTool = os.path.join(sys.prefix, "exkaldisrc", "tools", "lmplz")
cmd = f"{KenLMTool}{extraConfig}-o {order} --vocab_estimate {words_count} --text {text} --arpa {outFile} --limit_vocab_file {wordlistTemp.name}"
out, err, cod = run_shell_command(cmd, stderr="PIPE")
if (isinstance(cod, int) and cod != 0) or (
not os.path.isfile(outFile)) or (os.path.getsize(outFile)
== 0):
print(err.decode())
if os.path.isfile(outFile):
os.remove(outFile)
raise KenlmProcessError("Failed to generate arpa file.")
return outFile
def train_ngrams_srilm(lexicons, order, textFile, outFile, config=None):
'''
Train n-grams language model with Srilm tookit.
Args:
<lexicons>: words.txt file path or Exkaldi LexiconBank object.
<order>: the maxinum order of n-grams.
<textFile>: text corpus file.
<outFile>: ARPA out file name.
<config>: configures, a Python dict object.
You can use .check_config("train_ngrams_srilm") function to get configure information that you can set.
Also you can run shell command "lmplz" to look their meaning.
'''
assert isinstance(
order, int
) and order > 0 and order < 10, "Expected <n> is a positive int value and it must be smaller than 10."
assert isinstance(textFile,
str), "Expected <textFile> is name-like string."
assert isinstance(outFile, str), "Expected <outFile> is name-like string."
assert type_name(
lexicons
) == "LexiconBank", f"Expected <lexicons> is exkaldi LexiconBank object but got {type_name(lexicons)}."
ExkaldiInfo.prepare_srilm()
if not os.path.isfile(textFile):
raise WrongPath(f"No such file:{textFile}")
else:
## Should check the numbers of lines
cmd = f"shuf {textFile} -n 100"
out, err, cod = run_shell_command(cmd,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE)
if (isinstance(cod, int) and cod != 0):
print(err.decode())
raise ShellProcessError("Failed to sample from text file.")
elif out == b'':
raise WrongDataFormat("Void text file.")
else:
out = out.decode().strip().split("\n")
spaceCount = 0
for line in out:
spaceCount += line.count(" ")
if spaceCount < len(out) // 2:
raise WrongDataFormat(
"The text file doesn't seem to be separated by spaces or extremely short."
)
wordlist = tempfile.NamedTemporaryFile("w+",
encoding='utf-8',
suffix=".txt")
unkSymbol = lexicons("oov")
try:
lexiconp = lexicons("lexiconp")
words = [x[0] for x in lexiconp.keys()]
wordlist.write("\n".join(words))
wordlist.seek(0)
#cmd2 = f"ngram-count -text {textFile} -order {order}"
extraConfig = " "
specifyDiscount = False
if config is not None:
if check_config(name='train_ngrams_srilm', config=config):
for key, value in config.items():
if isinstance(value, bool):
if value is True:
extraConfig += f"{key} "
if key.endswith("discount"):
specifyDiscount = True
else:
extraConfig += f" {key} {value}"
cmd = f"ngram-count -text {textFile} -order {order} -limit-vocab -vocab {wordlist.name} -unk -map-unk {unkSymbol} "
if specifyDiscount is False:
cmd += "-kndiscount "
cmd += "-interpolate "
if not outFile.rstrip().endswith(".arpa"):
outFile += ".arpa"
make_dependent_dirs(outFile, pathIsFile=True)
cmd += f" -lm {outFile}"
out, err, cod = run_shell_command(cmd, stderr=subprocess.PIPE)
if (isinstance(cod, int) and cod != 0) or (
not os.path.isfile(outFile)) or os.path.getsize(outFile) == 0:
print(err.decode())
if os.path.isfile(outFile):
os.remove(outFile)
raise KaldiProcessError(
f'Failed to generate ngrams language model.')
else:
return os.path.abspath(outFile)
finally:
wordlist.close()
def train_ngrams_kenlm(lexicons, order, textFile, outFile, config=None):
'''
Train n-grams language model with KenLm tookit.
Args:
<lexicons>: words.txt file path or Exkaldi LexiconBank object.
<order>: the maxinum order of n-grams.
<textFile>: text corpus file.
<outFile>: ARPA out file name.
<config>: configures, a Python dict object.
You can use .check_config("train_ngrams_kenlm") function to get configure information that you can set.
Also you can run shell command "lmplz" to look their meaning.
'''
assert isinstance(
order, int
) and 0 < order <= 6, "We support maximum 6-grams LM in current version."
if not os.path.isfile(textFile):
raise WrongPath("No such file:{}".format(textFile))
else:
## Should check the numbers of lines
cmd = f"shuf {textFile} -n 100"
out, err, cod = run_shell_command(cmd,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE)
if (isinstance(cod, int) and cod != 0):
print(err.decode())
raise ShellProcessError("Failed to sample from text file.")
elif out == b'':
raise WrongDataFormat("Void text file.")
else:
out = out.decode().strip().split("\n")
spaceCount = 0
for line in out:
spaceCount += line.count(" ")
if spaceCount < len(out) // 2:
raise WrongDataFormat(
"The text file doesn't seem to be separated by spaces or extremely short."
)
extraConfig = " "
if config != None:
assert isinstance(
config, dict
), f"<config> should be dict object but got: {type_name(config)}."
if check_config(name='train_ngrams_kenlm', config=config):
if "--temp_prefix" in config.keys() and "-T" in config.keys():
raise WrongOperation(
f'"--temp_prefix" and "-T" is the same configure so only one of them is expected.'
)
if "--memory" in config.keys() and "-S" in config.keys():
raise WrongOperation(
f'"--memory" and "-S" is the same configure so only one of them is expected.'
)
for key, value in config.items():
if isinstance(value, bool):
if value is True:
extraConfig += f"{key} "
else:
extraConfig += f"{key} {value} "
assert isinstance(outFile, str), f"<outFile> should be a string."
if not outFile.rstrip().endswith(".arpa"):
outFile += ".arpa"
make_dependent_dirs(outFile, pathIsFile=True)
words = tempfile.NamedTemporaryFile("w+", suffix=".txt", encoding="utf-8")
try:
if type_name(lexicons) == "LexiconBank":
ws = lexicons("words")
words_count = math.ceil(len(ws) / 10) * 10
ws = "\n".join(ws.keys())
elif isinstance(lexicons, str):
if not os.path.isfile(lexicons):
raise WrongPath(f"No such file:{lexicons}.")
with open(lexicons, "r", encoding="utf-8") as fr:
lines = fr.readlines()
ws = []
for line in lines:
line = line.strip().split(maxsplit=1)
if len(line) < 1:
continue
else:
ws.append(line[0])
words_count = math.ceil(len(ws) / 10) * 10
ws = "\n".join(ws)
else:
raise UnsupportedType(
"<lexicons> should be LexiconBank object or file path.")
words.write(ws)
words.seek(0)
KenLMTool = os.path.join(sys.prefix, "exkaldisrc", "tools", "lmplz")
cmd = f"{KenLMTool}{extraConfig}-o {order} --vocab_estimate {words_count} --text {textFile} --arpa {outFile} --limit_vocab_file {words.name}"
out, err, cod = run_shell_command(cmd, stderr=subprocess.PIPE)
if (isinstance(cod, int) and cod != 0) or (
not os.path.isfile(outFile)) or (os.path.getsize(outFile)
== 0):
print(err.decode())
raise KenlmProcessError("Failed to generate arpa file.")
else:
return os.path.abspath(outFile)
finally:
words.close()
