def get(cls):
if cache.get("AccessToken") is not None:
return cache.get("AccessToken")
res = get_access_token() or ""
cache.set("AccessToken", res, timeout=60 * 60)
return res
def get_child_ids(cls, rid):
child_ids = cache.get(cls.PREFIX_CHILDREN.format(rid))
if not child_ids:
from api.lib.perm.acl.role import RoleRelationCRUD
child_ids = RoleRelationCRUD.get_child_ids(rid)
cache.set(cls.PREFIX_CHILDREN.format(rid), child_ids, timeout=0)
return child_ids
def get_by_name(cls, app_id, name):
role = cache.get(cls.PREFIX_NAME.format(app_id, name))
if role is None:
role = Role.get_by(app_id=app_id, name=name, first=True, to_dict=False)
if role is not None:
cache.set(cls.PREFIX_NAME.format(app_id, name), role)
return role
def get_parent_ids(cls, rid):
parent_ids = cache.get(cls.PREFIX_PARENT.format(rid))
if not parent_ids:
from api.lib.perm.acl.role import RoleRelationCRUD
parent_ids = RoleRelationCRUD.get_parent_ids(rid)
cache.set(cls.PREFIX_PARENT.format(rid), parent_ids, timeout=0)
return parent_ids
def get(cls, rid):
role = cache.get(cls.PREFIX_ID.format(rid))
if role is None:
role = Role.get_by_id(rid)
if role is not None:
cache.set(cls.PREFIX_ID.format(rid), role)
return role
def get(cls, key):
perm = cache.get(cls.PREFIX_ID.format(key))
perm = perm or cache.get(cls.PREFIX_NAME.format(key))
if perm is None:
perm = Permission.get_by_id(key)
perm = perm or Permission.get_by(name=key, first=True, to_dict=False)
if perm is not None:
cache.set(cls.PREFIX_ID.format(key), perm)
return perm
def get_resources(cls, rid):
"""
:param rid:
:return: {id2perms: {resource_id: [perm,]}, group2perms: {group_id: [perm, ]}}
"""
resources = cache.get(cls.PREFIX_RESOURCES.format(rid))
if not resources:
from api.lib.perm.acl.role import RoleCRUD
resources = RoleCRUD.get_resources(rid)
cache.set(cls.PREFIX_RESOURCES.format(rid), resources, timeout=0)
return resources or {}
def context_query(query_text, blob_uri, op_result, search_output_mode=SearchOutputMode.EXACT_MATCH):
match_result_cache = cache.get('match_result_cache') or dict()
match_cache_key = MatchCacheKey(blob_uri, query_text, True)
if match_cache_key in match_result_cache: return match_result_cache[match_cache_key]
match_results = []
for result in op_result.results:
# Check if the response is valid, which happens if and only if the result alternatives
# is at least of length 1 AND the transcript is non-empty.
if len(result.alternatives) == 0 or not result.alternatives[0].transcript: continue
tokens = [word.strip().lower() for word in result.alternatives[0].transcript.split(' ') if word != '']
puncutation_translator = str.maketrans('', '', string.punctuation)
sanitized_tokens = [token.translate(puncutation_translator) for token in tokens]
predictions = get_context_search_model().predict((result.alternatives[0].transcript, query_text))
for prediction in predictions:
answer = prediction[0].strip()
answer_tokens = answer.lower().translate(puncutation_translator).split(' ')
if not answer: continue
sublist = find_sub_list(sanitized_tokens, answer_tokens)
if len(sublist) == 0: continue
sublist_bounds = sublist[0]
if search_output_mode == SearchOutputMode.EXACT_MATCH:
start_word_index = sublist_bounds[0]
end_word_index = sublist_bounds[1]
transcript = answer
elif search_output_mode == SearchOutputMode.SENTENCE:
pass
start_word = result.alternatives[0].words[start_word_index]
end_word = result.alternatives[0].words[end_word_index]
match_results.append({
'matched_query': query_text,
'start_time': get_word_time_seconds(start_word.start_time),
'end_time': get_word_time_seconds(end_word.end_time),
'confidence': prediction[2],
'transcript': transcript
})
match_result_cache[match_cache_key] = match_results
cache.set('match_result_cache', match_result_cache, timeout=0)
return match_results
def set(cls, user):
'''
设置key
'''
cache.set(cls.PREFIX_ID.format(user.id), user)
cache.set(cls.PREFIX_NAME.format(user.username), user)
cache.set(cls.PREFIX_NICK.format(user.nickname), user)
def set(cls, type_id, attr_id, attr):
cache.set(cls.PREFIX_ID.format(type_id, attr_id), attr)
def string_query(query_text, blob_uri, op_result, search_output_mode=SearchOutputMode.EXACT_MATCH):
match_result_cache = cache.get('match_result_cache') or dict()
match_cache_key = MatchCacheKey(blob_uri, query_text, False)
if match_cache_key in match_result_cache: return match_result_cache[match_cache_key]
match_results = []
for result in op_result.results:
# Check if the response is valid, which happens if and only if the result alternatives
# is at least of length 1 AND the transcript is non-empty.
if len(result.alternatives) == 0 or not result.alternatives[0].transcript: continue
sentences = [sentence.strip().lower() for sentence in nltk.tokenize.sent_tokenize(result.alternatives[0].transcript)]
tokens = [word.strip().lower() for word in result.alternatives[0].transcript.split(' ') if word != '']
for sentence in sentences:
sublist_bounds = find_sub_list(tokens, sentence.split(' '))[0]
sentence_word_infos = result.alternatives[0].words[sublist_bounds[0]:sublist_bounds[1]+1]
matches = re.finditer(query_text, sentence)
match_cache = set()
for match in matches:
if search_output_mode == SearchOutputMode.EXACT_MATCH:
start_boundary = i = match.start()
# Check if the match starts inside a word
if i != 0 and sentence[i - 1] != ' ':
start_boundary = sentence.rfind(' ', 0, i) + 1
end_boundary = j = match.end() - 1
# Check if the match ends inside a word
if j != len(sentence) - 1 and sentence[j + 1] != ' ':
end_boundary = sentence.find(' ', j)
if end_boundary == -1:
end_boundary = len(sentence) - 1
else:
end_boundary -= 1
# Add boundary
boundary = (start_boundary, end_boundary)
if boundary in match_cache: continue
match_cache.add(boundary)
start_word_index = sentence.count(' ', 0, start_boundary)
end_word_index = sentence.count(' ', 0, end_boundary)
elif search_output_mode == SearchOutputMode.SENTENCE:
start_word_index = 0
end_word_index = len(sentence_word_infos) - 1
confidence = result.alternatives[0].confidence
start_word = sentence_word_infos[start_word_index]
end_word = sentence_word_infos[end_word_index]
transcript = sentence[match.start():match.end()]
match_results.append({
'matched_query': transcript,
'start_time': get_word_time_seconds(start_word.start_time),
'end_time': get_word_time_seconds(end_word.end_time),
'confidence': confidence,
'transcript': transcript
})
match_result_cache[match_cache_key] = match_results
cache.set('match_result_cache', match_result_cache, timeout=0)
return match_results
def set(cls, app):
cache.set(cls.PREFIX_ID.format(app.id), app)
cache.set(cls.PREFIX_NAME.format(app.name), app)
def get(cls, id):
role = cache.get(cls.PREFIX_NAME.format(id))
if role is None:
role = Role.get_by_id(id)
if role is not None:
cache.set(cls.PREFIX_ID.format(rid), role)
def set(cls, id, permission):
cache.set(cls.PREFIX_ID.format(id), permission)
def set_count_error(cls, key, value):
cache.set(cls.PREFIX_ERROR.format(key), value)
def query():
route_start_time = start_time = time.time()
print('='*20)
data = get_validator_data()
query_text = data['query'].strip().lower()
search_output_mode = data['search_output_mode']
auth_filepath = Path(current_app.instance_path) / Path(current_app.config['GOOGLE_CLOUD_AUTH_FILENAME'])
speech_client = speech.SpeechClient.from_service_account_json(auth_filepath)
storage_client = storage.Client.from_service_account_json(auth_filepath)
bucket = storage_client.get_bucket(current_app.config['GOOGLE_CLOUD_STORAGE_BUCKET_NAME'])
with open(get_tmp_filepath(), 'w+b') as input_file:
data['file_input'] = data['file_input'][data['file_input'].find(',')+1:]
file_input = data['file_input'] + '=' * ((4 - len(data['file_input']) % 4) % 4)
input_file.write(base64.b64decode(file_input))
input_hash = hash_util.get_md5_str(file_input)
print('Decoding and writing input file took {:.3f} seconds'.format(time.time() - start_time))
start_time = time.time()
input_hash_to_blob_uri = cache.get('input_hash_to_blob_uri') or dict()
if input_hash in input_hash_to_blob_uri:
blob_uri = input_hash_to_blob_uri[input_hash]
blob = bucket.blob(blob_uri.replace('gs://{}/'.format(current_app.config['GOOGLE_CLOUD_STORAGE_BUCKET_NAME']), ''))
else:
try:
audio, sample_rate = librosa.load(input_file.name)
except Exception as exception:
raise exceptions.AudioFileLoadError('file_input', exception)
print('Libroa audio file loading took {:.3f} seconds'.format(time.time() - start_time))
start_time = time.time()
tmp_filepath = get_tmp_filepath()
# Preprocess the audio data by converting it to a mono WAVE
audio_mono = librosa.to_mono(audio)
soundfile.write(tmp_filepath, audio_mono, sample_rate, subtype='PCM_16', format='wav')
print('Audio processing and exporting took {:.3f} seconds'.format(time.time() - start_time))
start_time = time.time()
# Upload the file to GCS if it doesn't already exists
bucket_audio_root = current_app.config['GOOGLE_CLOUD_STORAGE_BUCKET_AUDIO_ROOT']
blob_root = bucket_audio_root
if blob_root and not bucket_audio_root.endswith('/'):
blob_root = bucket_audio_root + '/'
with open(tmp_filepath, 'rb') as audio_file:
blob_filename = '{}{}'.format(blob_root, hash_util.get_crc32_str(audio_file))
blob = bucket.blob(blob_filename)
if not blob.exists():
blob.upload_from_file(audio_file, rewind=True)
blob_uri = 'gs://{}/{}'.format(current_app.config['GOOGLE_CLOUD_STORAGE_BUCKET_NAME'], blob_filename)
input_hash_to_blob_uri[input_hash] = blob_uri
cache.set('input_hash_to_blob_uri', input_hash_to_blob_uri, timeout=0)
# Remove audio file now that we are done with it
tmp_filepath.unlink()
print('Uploading to GCS took {:.3f} seconds'.format(time.time() - start_time))
start_time = time.time()
Path(input_file.name).unlink()
transcription_cache = cache.get('transcription_cache') or dict()
if blob_uri in transcription_cache:
op_result = transcription_cache[blob_uri]
print('Loaded {} from cache'.format(blob_uri))
else:
# This configuration is predetermined...All audio files are converted
# a WAVE format with 16-bit PCM encoding. Sample rate is determined using librosa.
config = types.RecognitionConfig(
encoding=enums.RecognitionConfig.AudioEncoding.LINEAR16,
sample_rate_hertz=sample_rate,
enable_speaker_diarization=True,
enable_automatic_punctuation=True,
language_code='en-US')
operation = speech_client.long_running_recognize(config, types.RecognitionAudio(uri=blob_uri))
op_result = operation.result()
transcription_cache[blob_uri] = op_result
cache.set('transcription_cache', transcription_cache, timeout=0)
print('Transcription took {:.3f} seconds'.format(time.time() - start_time))
start_time = time.time()
if not data['is_context_search']:
match_results = string_query(query_text, blob_uri, op_result, search_output_mode)
else:
match_results = context_query(query_text, blob_uri, op_result, search_output_mode)
print('Search took {:.3f} seconds'.format(time.time() - start_time))
end_time = time.time()
access_link = blob.generate_signed_url(timedelta(hours=1))
return jsonify(status_code=201, message='Query was successful!', matches=match_results,
access_link=access_link, elapsed_time=end_time - route_start_time, success=True)
def set(cls, ct):
cache.set("CIType::Name::{0}".format(ct.name), ct)
cache.set("CIType::ID::{0}".format(ct.id), ct)
cache.set("CIType::Alias::{0}".format(ct.alias), ct)
def set(cls, key, values):
ci_type = CITypeCache.get(key)
if ci_type is not None:
cache.set("CITypeAttribute::ID::{0}".format(ci_type.id), values)
cache.set("CITypeAttribute::Name::{0}".format(ci_type.name),
values)
def set(cls, key, values):
ci_type = CITypeCache.get(key)
if ci_type is not None:
cache.set(cls.PREFIX_ID.format(ci_type.id), values)
cache.set(cls.PREFIX_NAME.format(ci_type.name), values)
def set(cls, attr):
cache.set('Field::ID::{0}'.format(attr.id), attr)
cache.set('Field::Name::{0}'.format(attr.name), attr)
cache.set('Field::Alias::{0}'.format(attr.alias), attr)
def set(cls, attr):
cache.set(cls.PREFIX_ID.format(attr.id), attr)
cache.set(cls.PREFIX_NAME.format(attr.name), attr)
cache.set(cls.PREFIX_ALIAS.format(attr.alias), attr)
def set(cls, ct):
cache.set("RelationType::Name::{0}".format(ct.name), ct)
cache.set("RelationType::ID::{0}".format(ct.id), ct)
def set(cls, ct):
cache.set(cls.PREFIX_NAME.format(ct.name), ct)
cache.set(cls.PREFIX_ID.format(ct.id), ct)
cache.set(cls.PREFIX_ALIAS.format(ct.alias), ct)
