def test_audio_4(self):
if self.data_dir is None:
unittest.TestCase.skipTest(self, "DLPY_DATA_DIR is not set in the environment variables")
audio_table = AudioTable.load_audio_files(self.conn, self.data_dir+'listingFile.txt')
if audio_table is not None:
self.assertTrue(AudioTable.extract_audio_features(self.conn, audio_table) is not None)
else:
self.assertTrue(False)
def transcribe(self,
audio_path,
max_path_size=100,
alpha=1.0,
beta=0.0,
gpu=None):
"""
Transcribe the audio file into text.
Notice that for this API, we are assuming that the speech-to-test models published by SAS Viya 3.4 will be used.
Please download the acoustic and language model files from here:
https://support.sas.com/documentation/prod-p/vdmml/zip/speech_19w21.zip
Parameters
----------
audio_path : string
Specifies the location of the audio file (client-side, absolute/relative).
max_path_size : int, optional
Specifies the maximum number of paths kept as candidates of the final results during the decoding process.
Default = 100
alpha : double, optional
Specifies the weight of the language model, relative to the acoustic model.
Default = 1.0
beta : double, optional
Specifies the weight of the sentence length, relative to the acoustic model.
Default = 0.0
gpu : class : `dlpy.model.Gpu`, optional
When specified, the action uses Graphics Processing Unit hardware.
The simplest way to use GPU processing is to specify "gpu=1". In this case, the default values of
other GPU parameters are used.
Setting gpu=1 enables all available GPU devices for use. Setting gpu=0 disables GPU processing.
Returns
-------
string
Transcribed text from audio file located at 'audio_path'.
"""
# check if acoustic model is loaded
if self.acoustic_model is None:
raise DLPyError(
"acoustic model not found. "
"Please load the acoustic model with \"load_acoustic_model\" before calling \"transcribe\"."
)
# check if language model is loaded
if self.language_model_caslib is None:
raise DLPyError(
"language model not found. "
"Please load the language model with \"load_language_model\" before calling \"transcribe\"."
)
# step 1: preparation and segmentation
listing_path_after_caslib, listing_path_local, segment_path_after_caslib_list, segment_path_local_list = \
segment_audio(audio_path, self.local_path, self.data_path_after_caslib, 10, 16000, 2)
segment_path_list = [
self.data_caslib_path + segment_path_after_caslib
for segment_path_after_caslib in segment_path_after_caslib_list
]
# step 2: load audio
try:
audio_table = AudioTable.load_audio_files(
self.conn,
path=listing_path_after_caslib,
caslib=self.data_caslib)
except DLPyError as err:
if "cannot load audio files, something is wrong!" in str(err):
clean_audio(listing_path_local, segment_path_local_list)
raise DLPyError(
"Error: Cannot load the audio files. "
"Please verify that \"data_path\" and \"local_path\" are pointing to the same position."
)
raise err
# step 3: extract features
feature_table = AudioTable.extract_audio_features(self.conn,
table=audio_table,
n_output_frames=3500,
copyvars=["_path_"])
# step 4: score features
self.acoustic_model.score(table=feature_table,
model="asr",
init_weights="asr_weights",
copy_vars=["_path_"],
gpu=gpu,
casout=dict(name="score_table",
replace=True))
score_table = self.conn.CASTable(name="score_table")
# step 5: decode scores
rt = self.conn.retrieve("langModel.lmDecode",
_messagelevel='error',
table=score_table,
casout=dict(name="result_table", replace=True),
langModelTable=dict(
name=self.language_model_name,
caslib=self.language_model_caslib),
blankLabel=" ",
spaceLabel="&",
maxPathSize=max_path_size,
alpha=alpha,
beta=beta,
copyvars=["_path_"])
if rt.severity > 1:
for msg in rt.messages:
print(msg)
raise DLPyError("Failed to decode the scores.")
result_table = self.conn.CASTable(name="result_table")
# step 6: concatenate results
result_dict = dict(
zip(list(result_table["_path_"]),
list(result_table["_audio_content_"])))
result_list = [
result_dict[segment_path] for segment_path in segment_path_list
]
result_list = [result.strip() for result in result_list]
result_list = [result for result in result_list if len(result) > 0]
result = " ".join(result_list)
# step 7: cleaning
clean_audio(listing_path_local, segment_path_local_list)
return result
def test_audio_5(self):
self.assertTrue(
AudioTable.extract_audio_features(self.s, 'string') is None)
