def perplexity(self, transcription):
'''
Compute perplexity of a transcription.
Args:
<transcription>: file path or exkaldi Transcription object.
Return:
an named tuple: PPL(probs,sentences,words,ppl,ppl1).
'''
declare.is_potential_transcription("transcription", transcription)
if isinstance(transcription, str):
transcription = load_transcription(transcription)
prob = self.score(transcription)
sentences = len(prob)
words = transcription.sentence_length().sum()
sumProb = prob.sum()
ppl = 10**(-sumProb / (sentences + words))
ppl1 = 10**(-sumProb / (words))
return namedtuple("PPL",
["prob", "sentences", "words", "ppl", "ppl1"])(round(
sumProb,
2), sentences, words, round(ppl,
2), round(ppl1, 2))
def score(self, transcription, bos=True, eos=True):
'''
Score a transcription.
Args:
<transcription>: file path or exkaldi Transcription object.
<bos>: If True, add <s> to the head.
<eos>: If True, add </s> to the tail.
Return:
an exkaldi Metric object.
'''
declare.is_potential_transcription("transcription", transcription)
if isinstance(transcription, str):
transcription = load_transcription(transcription)
scores = Metric(name=f"LMscore({transcription.name})")
for uttID, txt in transcription.items():
scores[uttID] = self.score_sentence(txt, bos, eos)
return scores
def wer(ref, hyp, ignore=None, mode='all'):
'''
Compute WER (word error rate) between <ref> and <hyp>.
Args:
<ref>,<hyp>: exkaldi transcription object or file path.
<ignore>: ignore a symbol.
<mode>: "all" or "present".
Return:
a namedtuple of score information.
'''
declare.is_potential_transcription("ref", ref)
declare.is_potential_transcription("hyp", hyp)
declare.is_instances("mode", mode, ['all', 'present'])
declare.kaldi_existed()
if ignore is not None:
declare.is_valid_string("ignore", ignore)
with FileHandleManager() as fhm:
if ignore is None:
if type_name(hyp) == "Transcription":
hypTemp = fhm.create("w+", suffix=".txt", encoding="utf-8")
hyp.save(hypTemp)
hyp = hypTemp.name
if type_name(ref) == "Transcription":
refTemp = fhm.create("w+", suffix=".txt", encoding="utf-8")
ref.save(refTemp)
ref = refTemp.name
cmd = f'compute-wer --text --mode={mode} ark:{ref} ark,p:{hyp}'
scoreOut, scoreErr, _ = run_shell_command(cmd,
stdout="PIPE",
stderr="PIPE")
else:
# remove the ingored symbol in hyp
if type_name(hyp) == "Transcription":
hyp = hyp.save()
else:
with open(hyp, "r", encoding="utf-8") as fr:
hyp = fr.read()
hypTemp = fhm.create("w+", suffix=".txt", encoding="utf-8")
cmd = f'sed "s/{ignore} //g" > {hypTemp.name}'
hypOut, err, _ = run_shell_command(cmd,
stdin="PIPE",
stdout="PIPE",
stderr="PIPE",
inputs=hyp)
if len(hypOut) == 0:
raise WrongDataFormat("<hyp> has wrong data formation.",
err.decode())
# remove the ingored symbol in ref
if type_name(ref) == "Transcription":
ref = ref.save()
else:
with open(ref, "r", encoding="utf-8") as fr:
ref = fr.read()
refTemp = fhm.create("w+", suffix=".txt", encoding="utf-8")
cmd = f'sed "s/{ignore} //g" > {refTemp.name}'
refOut, err, cod = run_shell_command(cmd,
stdin="PIPE",
stdout="PIPE",
stderr="PIPE",
inputs=ref)
if cod != 0 or len(refOut) == 0:
raise WrongDataFormat("<ref> has wrong data formation.",
err.decode())
# score
cmd = f'compute-wer --text --mode={mode} ark:{refTemp.name} ark,p:{hypTemp.name}'
scoreOut, scoreErr, _ = run_shell_command(cmd,
stdout="PIPE",
stderr="PIPE")
if len(scoreOut) == 0:
raise KaldiProcessError("Failed to compute WER.", scoreErr.decode())
else:
out = scoreOut.decode().split("\n")
pattern1 = '%WER (.*) \[ (.*) \/ (.*),(.*) ins,(.*) del,(.*) sub \]'
pattern2 = "%SER (.*) \[ (.*) \/ (.*) \]"
pattern3 = "Scored (.*) sentences,(.*) not present in hyp."
s1 = re.findall(pattern1, out[0])[0]
s2 = re.findall(pattern2, out[1])[0]
s3 = re.findall(pattern3, out[2])[0]
return namedtuple("Score", [
"WER", "words", "insErr", "delErr", "subErr", "SER", "sentences",
"wrongSentences", "missedSentences"
])(
float(s1[0]), #WER
int(s1[2]), #words
int(s1[3]), #ins
int(s1[4]), #del
int(s1[5]), #sub
float(s2[0]), #SER
int(s2[1]), #sentences
int(s2[2]), #wrong sentences
int(s3[1]) #missed sentences
)
def edit_distance(ref, hyp, ignore=None, mode='present'):
'''
Compute edit-distance score.
Args:
<ref>,<hyp>: exkaldi Transcription objects.
<ignore>: Ignoring specific symbols.
<mode>: When both are Transcription objects,if mode is 'present',skip the missed utterances.
Return:
a namedtuple object including score information.
'''
declare.is_potential_transcription("ref", ref)
declare.is_potential_transcription("hyp", hyp)
declare.is_instances("mode", mode, ['all', 'present'])
if ignore is not None:
declare.is_valid_string("ignore", ignore)
if isinstance(ref, str):
ref = load_transcription(ref)
if isinstance(hyp, str):
hyp = load_transcription(hyp)
allED = 0
words = 0
SER = 0
sentences = 0
wrongSentences = 0
missedSentences = 0
ref = ref.sort()
hyp = hyp.sort()
for utt, hypTrans in hyp.items():
try:
refTrans = ref[utt]
except KeyError as e:
if mode == "all":
raise Exception(
"Missing transcription in reference,set <mode> as 'all' to skip it."
)
else:
missedSentences += 1
else:
sentences += 1
refTrans = refTrans.split()
hypTrans = hypTrans.split()
ed, wds = pure_edit_distance(refTrans, hypTrans, ignore=ignore)
allED += ed
words += wds
if ed > 0:
wrongSentences += 1
if sentences == 0:
raise Exception(
"Missing all transcription in reference. We don't think it's a reasonable result. Check the file please."
)
return namedtuple("Score", [
"editDistance", "words", "SER", "sentences", "wrongSentences",
"missedSentences"
])(allED, words, wrongSentences / sentences, sentences, wrongSentences,
missedSentences)
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