def diarization(self, wavfile, bucket_name, project_name):
if not os.path.exists(project_name):
os.mkdir(project_name)
print("Uploading {} to google cloud storage bucket".format(wavfile))
set_value("label_wav_file_transcribe",
"Uploading file to cloud storage bucket...")
self.upload_blob(bucket_name, wavfile, "temp_audio.wav")
gcs_uri = "gs://{}/temp_audio.wav".format(bucket_name)
set_value("label_wav_file_transcribe", "Finished uploading.")
client = speech.SpeechClient()
audio = speech.RecognitionAudio(uri=gcs_uri)
info = mediainfo(wavfile)
sample_rate = info['sample_rate']
print("Transcribing {} with audio rate {}".format(
wavfile, sample_rate))
config = speech.RecognitionConfig(
encoding=speech.RecognitionConfig.AudioEncoding.LINEAR16,
sample_rate_hertz=int(sample_rate),
language_code="en-US",
enable_automatic_punctuation=True,
enable_word_time_offsets=True,
enable_speaker_diarization=True,
diarization_speaker_count=int(get_value("input_diarization_num")),
)
operation = client.long_running_recognize(config=config, audio=audio)
print(
"Waiting for operation to complete, this may take several minutes..."
)
set_value(
"label_wav_file_transcribe",
"Waiting for operation to complete, this may take several minutes..."
)
response = operation.result(timeout=28800)
result = response.results[-1]
words = result.alternatives[0].words
active_speaker = 1
transcript = []
current_cut = 0
previous_cut = 0
speaker_wavs = []
for x in range(int(get_value("input_diarization_num"))):
speaker_wavs.append(AudioSegment.empty())
transcript.append("")
w = AudioSegment.from_wav(wavfile)
for word in words:
if word.speaker_tag == active_speaker:
end_time = word.end_time
current_cut = end_time.total_seconds() * 1e3
#print(current_cut)
transcript[active_speaker - 1] += word.word + ' '
else:
#speaker has changed
transcript[active_speaker - 1] += word.word + ' '
w_cut = w[(previous_cut):current_cut]
previous_cut = current_cut
speaker_wavs[active_speaker -
1] = speaker_wavs[active_speaker - 1] + w_cut
active_speaker = word.speaker_tag
#finish last wav cut
w_cut = w[previous_cut:current_cut]
speaker_wavs[active_speaker -
1] = speaker_wavs[active_speaker - 1] + w_cut
for i, wave in enumerate(speaker_wavs):
speaker_wavs[i].export("{}/speaker_{}.wav".format(
project_name, i + 1),
format="wav")
for i, text in enumerate(transcript):
f = open("{}/speaker_{}.txt".format(project_name, i + 1), 'w')
f.write(transcript[i])
f.close()
set_value("label_wav_file_transcribe", "Done!")
print("Done with diarization!")
print('\a') #system beep
def transcribe_audio_to_tsv_with_diarization(input_audio_paths,
output_tsv_path,
sample_rate,
language_code,
speaker_count,
begin_sec=0.0):
"""Transcribe speech in input audio files and write results to .tsv file.
This method differs from transcribe_audio_to_tsv() in that it performs speaker
diarization and uses the word-level speaker indices to regroup the transcripts.
"""
client = speech.SpeechClient()
enable_speaker_diarization = speaker_count > 0
config = speech.RecognitionConfig(
encoding=speech.RecognitionConfig.AudioEncoding.LINEAR16,
sample_rate_hertz=sample_rate,
audio_channel_count=1,
enable_separate_recognition_per_channel=False,
language_code=language_code,
enable_speaker_diarization=enable_speaker_diarization,
diarization_speaker_count=speaker_count)
streaming_config = speech.StreamingRecognitionConfig(config=config,
interim_results=False)
requests = audio_data_generator(input_audio_paths, config)
responses = client.streaming_recognize(streaming_config, requests)
with open(output_tsv_path, "w" if not begin_sec else "a") as f:
if not begin_sec:
# Write the TSV header.
f.write(tsv_data.HEADER + "\n")
utterances = []
for response in responses:
if not response.results:
continue
results = [
result for result in response.results if result.is_final
]
max_confidence = -1
best_transcript = None
result_end_time = None
for result in results:
for alt in result.alternatives:
if alt.confidence > max_confidence:
max_confidence = alt.confidence
best_transcript = alt.transcript.strip()
diarized_words = [(word.word, word.speaker_tag,
word.start_time.total_seconds(),
word.end_time.total_seconds())
for word in alt.words]
result_end_time = result.result_end_time
if not best_transcript:
continue
end_time_sec = result_end_time.total_seconds()
utterances.append(best_transcript)
regrouped_utterances = regroup_utterances(utterances, diarized_words)
utterance_counter = 0
for (regrouped_utterance, speaker_index, start_time_sec,
end_time_sec) in regrouped_utterances:
utterance_counter += 1
line = "%.3f\t%.3f\t%s\t%s [U%d] [Speaker #%d]" % (
start_time_sec + begin_sec, end_time_sec + begin_sec,
tsv_data.SPEECH_TRANSCRIPT_TIER, regrouped_utterance,
utterance_counter, speaker_index)
print(line)
f.write(line + "\n")
def get_transcript(speech_file,content_type):
# google authentication
os.environ['GOOGLE_APPLICATION_CREDENTIALS'] = '/Users/andrewfung/Programming/Multiple Speaker Detection/multiple-speaker-detection-3ed65d50eff1.json'
# wget -nc https://realenglishconversations.com/...
# instantiate a speech client and declare an audio file
client = speech.SpeechClient()
with io.open(speech_file, "rb") as audio_file:
content = audio_file.read()
audio = speech.RecognitionAudio(content=content)
if 'wav' in content_type:
config = speech.RecognitionConfig(
encoding=speech.RecognitionConfig.AudioEncoding.LINEAR16,
sample_rate_hertz=16000,
language_code="en-US",
enable_speaker_diarization=True,
diarization_speaker_count=2,
)
elif 'mpeg' in content_type:
config = speech.RecognitionConfig(
encoding=speech.RecognitionConfig.AudioEncoding.MP3,
sample_rate_hertz=16000,
language_code="en-US",
enable_speaker_diarization=True,
diarization_speaker_count=2,
)
elif 'flac' in content_type:
config = speech.RecognitionConfig(
encoding=speech.RecognitionConfig.AudioEncoding.FLAC,
sample_rate_hertz=16000,
language_code="en-US",
enable_speaker_diarization=True,
diarization_speaker_count=2,
)
print("Waiting for operation to complete...")
response = client.recognize(config=config, audio=audio)
result = response.results[-1]
words_info = result.alternatives[0].words
words_list = []
# Printing out the output:
for word_info in words_info:
words_list.append(
{
'word': word_info.word,
'speaker_tag': word_info.speaker_tag,
'start_time': word_info.start_time,
'end_time': word_info.end_time,
}
)
# print(words_list)
# create a script based on the words_list
current_speaker = words_list[0]['speaker_tag']
current_line = []
script = []
for item in words_list:
if item['speaker_tag'] != current_speaker:
# speaker changed, end of line
script.append(
{
'speaker': current_speaker,
'line': current_line
}
)
current_line = []
current_speaker = item['speaker_tag']
else:
# same speaker, add to the current line
current_line.append(item['word'])
script.append(
{
'speaker': current_speaker,
'line': current_line
}
)
script = [(f"Speaker {line['speaker']}: " + " ".join(line['line']) + "\n") for line in script]
return script
def get_transcripts_json(gcsPath,
langCode,
phraseHints=[],
speakerCount=1,
enhancedModel=None):
"""Transcribes audio files.
Args:
gcsPath (String): path to file in cloud storage (i.e. "gs://audio/clip.mp4")
langCode (String): language code (i.e. "en-US", see https://cloud.google.com/speech-to-text/docs/languages)
phraseHints (String[]): list of words that are unusual but likely to appear in the audio file.
speakerCount (int, optional): Number of speakers in the audio. Only works on English. Defaults to None.
enhancedModel (String, optional): Option to use an enhanced speech model, i.e. "video"
Returns:
list | Operation.error
"""
# Helper function for simplifying Google speech client response
def _jsonify(result):
json = []
for section in result.results:
data = {
"transcript": section.alternatives[0].transcript,
"words": []
}
for word in section.alternatives[0].words:
data["words"].append({
"word":
word.word,
"start_time":
word.start_time.total_seconds(),
"end_time":
word.end_time.total_seconds(),
"speaker_tag":
word.speaker_tag
})
json.append(data)
return json
client = speech.SpeechClient()
audio = speech.RecognitionAudio(uri=gcsPath)
diarize = speakerCount if speakerCount > 1 else False
print(f"Diarizing: {diarize}")
diarizationConfig = speech.SpeakerDiarizationConfig(
enable_speaker_diarization=speakerCount
if speakerCount > 1 else False, )
# In English only, we can use the optimized video model
if langCode == "en":
enhancedModel = "video"
config = speech.RecognitionConfig(
language_code="en-US" if langCode == "en" else langCode,
enable_automatic_punctuation=True,
enable_word_time_offsets=True,
speech_contexts=[{
"phrases": phraseHints,
"boost": 15
}],
diarization_config=diarizationConfig,
profanity_filter=True,
use_enhanced=True if enhancedModel else False,
model="video" if enhancedModel else None)
res = client.long_running_recognize(config=config, audio=audio).result()
return _jsonify(res)
def transcribe_file(speech_file):
"""Transcribe the given audio file asynchronously."""
from google.cloud import speech_v1p1beta1
from google.cloud.speech_v1p1beta1 import enums
from google.cloud.speech_v1p1beta1 import types
client = speech_v1p1beta1.SpeechClient()
# [START migration_async_request]
with io.open(speech_file, 'rb') as audio_file:
content = audio_file.read()
#print('Using ', speech_file, ', with the below config:')
#print("")
#print("importing speech_v1p1beta1")
#print("language_code='en-US'")
#print("use_enhanced=True")
#print("enable_automatic_punctuation=False")
#print("enable_word_time_offsets=False")
#print("profanity_filter=True")
#print("sample_rate=48000hz")
#print("")
#print("Transcript is as follows")
audio = types.RecognitionAudio(content=content)
config = types.RecognitionConfig(
encoding=enums.RecognitionConfig.AudioEncoding.LINEAR16,
sample_rate_hertz=int(args.samplerate),
language_code='en-US',
# alternative_language_codes='yue-Hant-HK',
use_enhanced=True,
# A model must be specified to use enhanced model.
model='phone_call',
enable_automatic_punctuation=False,
enable_word_time_offsets=False,
profanity_filter=True,
enable_speaker_diarization=True,
speech_contexts=[
speech_v1p1beta1.types.SpeechContext(phrases=[
keyword1, keyword2, keyword3, keyword4, keyword5, keyword6,
keyword7, keyword8, keyword9, keyword10
], )
],
)
# [START migration_async_response]
operation = client.long_running_recognize(config, audio)
# [END migration_async_request]
os.chdir("..")
os.chdir("Trans_Output_" + foldernametime)
with open("output_transcription.txt", "a") as myfile:
myfile.write(datetime.now().strftime("%Y-%m-%d %H:%M:%S") + "\n")
#myfile.write(' - Starting a new transcription.......\n')
#print('Waiting for operation to complete...')
response = operation.result(timeout=90)
# Each result is for a consecutive portion of the audio. Iterate through
# them to get the transcripts for the entire audio file.
for result in response.results:
# The first alternative is the most likely one for this portion.
print(('Transcript: {}'.format(result.alternatives[0].transcript)))
print(('Confidence: {}'.format(result.alternatives[0].confidence)))
with open("output_transcription.txt", "a") as myfile:
myfile.write(('Transcript: {}'.format(
result.alternatives[0].transcript)) + "\n")
myfile.write(('Confidence: {}'.format(
result.alternatives[0].confidence)) + "\n")
with open("output_transcription.txt", "a") as myfile:
myfile.write('')
script_config = configparser.ConfigParser()
try:
script_config.read("config.ini")
except Exception as e:
print("Error reading config file. Exiting.")
print(e)
exit()
# Auth
credentials = service_account.Credentials.from_service_account_file(
PurePath(Path(__file__).resolve().parent).joinpath(
Path(str(script_config["OPTS"]["Credentials"]))))
# Instantiate GC Speech client
client = speech.SpeechClient(credentials=credentials)
if str(script_config["OPTS"]["Mode"]) == "local":
# Read-in audio from local file (60s limit, gs is recommended Mode)
with io.open(
PurePath(Path(__file__).resolve().parent).joinpath(
str(script_config["OPTS"]["Path"])), "rb") as audio_file:
content = audio_file.read()
audio = speech.RecognitionAudio(content=content)
else:
# Read-in audio from GS
print(str(script_config["OPTS"]["Path"]))
audio = speech.RecognitionAudio(uri=str(script_config["OPTS"]["Path"]))
# Config request
req_config = speech.RecognitionConfig(
def transcribe_streaming(self, stream_file, configuration):
"""Streams transcription of the given audio file."""
import io
client = speech.SpeechClient()
output = ''
with io.open(stream_file, 'rb') as audio_file:
audio_content = audio_file.read()
config = {
"model":
configuration.get_model(),
"use_enhanced":
configuration.get_use_enhanced(),
"encoding":
configuration.get_encoding(),
"sample_rate_hertz":
configuration.get_sample_rate_hertz(),
"language_code":
configuration.get_language_code(),
"alternative_language_codes":
configuration.get_alternative_language_codes(),
"audio_channel_count":
configuration.get_audio_channel_count(),
"enable_separate_recognition_per_channel":
configuration.get_enable_separate_recognition_per_channel(),
"enable_speaker_diarization":
configuration.get_enableSpeakerDiarization(),
"diarization_speaker_count":
configuration.get_diarizationSpeakerCount(),
"enable_automatic_punctuation":
configuration.get_enableAutomaticPunctuation(),
"speech_contexts":
configuration.get_speech_context()
}
streaming_config = speech.types.StreamingRecognitionConfig(
config=config, interim_results=True)
# BUG IS HERE
#requests = speech.types.StreamingRecognizeRequest(
# audio_content=audio_content)
stream = [audio_content]
requests = (speech.types.StreamingRecognizeRequest(audio_content=chunk)
for chunk in stream)
responses = client.streaming_recognize(streaming_config, requests)
#import pdb; pdb.set_trace()
for response in responses:
# Once the transcription has settled, the first result will contain the
# is_final result. The other results will be for subsequent portions of
# the audio.
for result in response.results:
alternatives = result.alternatives
# The alternatives are ordered from most likely to least.
for alternative in alternatives:
output = ''.join(alternative.transcript)
return output
def analyze_audio(ogg_file_path, speaker_count=3):
"""Takes an audio file and outputs meeting statistics as a dictionary.
Args:
ogg_file_path (str): Path to the audio file in ogg-format.
speaker_count (int): Number of people participating in the meeting.
Returns:
Dict[str, Any]
"""
# Convert audio files to flac
if ogg_file_path.split(".")[-1] != "flac":
ogg_file_path = convert_ogg2flac(ogg_file_path)
speech_client = speech_v1p1beta1.SpeechClient(credentials=CREDENTIALS)
config = {
"enable_speaker_diarization": True,
"diarization_speaker_count": speaker_count,
"language_code": "en-US",
"encoding":
speech_v1p1beta1.enums.RecognitionConfig.AudioEncoding.FLAC,
"max_alternatives": 1,
"use_enhanced": True,
"sample_rate_hertz": 48000,
}
# Upload file to GCS Storage bucket
client = storage.Client(credentials=CREDENTIALS)
bucket = client.get_bucket(GCS_BUCKET_NAME)
blob = bucket.blob(ogg_file_path)
blob.upload_from_filename(ogg_file_path)
audio = {"uri": f"gs://{GCS_BUCKET_NAME}/{ogg_file_path}"}
operation = speech_client.long_running_recognize(config, audio)
response = operation.result()
# Empty response when speech to text failed
if not response.results:
json_out = {
"google_transcript": "",
"raw_transcript": "",
"transcript": [],
"speakers": [],
"topics": [],
"sentiment": {
"score": 0,
"magnitude": 0
},
}
return json_out
result = response.results[-1]
alternative = result.alternatives[0]
json_out = {
"google_transcript": alternative.transcript,
"raw_transcript": ' '.join([word.word for word in alternative.words])
}
# Get transcript distributed by speakers
transcript = []
sentence = []
last_speaker = alternative.words[0].speaker_tag
for word in alternative.words:
current_speaker = word.speaker_tag
if current_speaker == last_speaker:
sentence.append(word.word)
else:
transcript.append({
"speaker_id": last_speaker,
"line": ' '.join(sentence)
})
sentence = [word.word]
last_speaker = current_speaker
transcript.append({"speaker_id": last_speaker, "line": ' '.join(sentence)})
json_out["transcript"] = transcript
# Analyze speakers
speaker_tags = [word.speaker_tag for word in alternative.words]
unique_speakers = set(speaker_tags)
speaker_ratios = []
for speaker in unique_speakers:
speaker_ratios.append({
"speaker_id":
speaker,
"ratio":
round(speaker_tags.count(speaker) / len(speaker_tags), 2)
})
json_out["speakers"] = speaker_ratios
# Analyze sentiment and topics
sentiment, topics = analyze_text(json_out["raw_transcript"])
json_out["sentiment"] = sentiment
json_out["topics"] = topics
# Include speaker sentiment
speaker_sentiment = analyze_speaker_sentiment(json_out['transcript'])
for line in json_out['speakers']:
line.update({'sentiment_score': speaker_sentiment[line['speaker_id']]})
return json_out
def get_transcript_audio_file(audio_path, langs):
if 'en-US' in langs:
main_lang = 'en-US'
else:
main_lang = langs[0]
parent_dir = '/'.join(audio_path.split('/')[:-1])
temp_dir = parent_dir + '/temp'
fname = audio_path.split('/')[-1]
try:
os.mkdir(temp_dir)
except FileExistsError:
pass
full_audio = pydub.AudioSegment.from_wav(audio_path)
transcript = {}
for t in range(0, len(full_audio), SIZE):
try:
chunk = full_audio[t:t + SIZE]
except IndexError:
chunk = full_audio[t:]
chunk = chunk.set_sample_width(2)
chunk.export(temp_dir + '/' + str(int(t / SIZE)) + '_' + fname,
format='wav',
bitrate='16k')
chunk_info = mediainfo(temp_dir + '/' + str(int(t / SIZE)) + '_' +
fname)
config = {
"language_code": main_lang,
"sample_rate_hertz": int(chunk_info['sample_rate']),
"encoding": enums.RecognitionConfig.AudioEncoding.LINEAR16,
"profanity_filter": False,
"audio_channel_count": int(chunk_info['channels'])
}
if len(langs) > 1:
config["alternative_language_codes"] = langs[1:]
with io.open(temp_dir + '/' + str(int(t / SIZE)) + '_' + fname,
'rb') as f:
content = f.read()
audio = {"content": content}
try:
client = speech_v1p1beta1.SpeechClient.from_service_account_json(
'api_keys/google_cloud.json')
except:
client = speech_v1p1beta1.SpeechClient()
response = client.recognize(config, audio)
for result in response.results:
if int(t / 1000) not in transcript.keys():
transcript[int(t / 1000)] = []
alternative = result.alternatives[0]
transcript[int(t / 1000)].append(alternative.transcript)
os.remove(temp_dir + '/' + str(int(t / SIZE)) + '_' + fname)
os.rmdir(temp_dir)
return transcript
def main(new_connection_index):
"""start bidirectional streaming from microphone input to speech API"""
# diarization_config = {
# "enable_speaker_diarization": True,
# "min_speaker_count": 1,
# "max_speaker_count": 6,
# }
# alternative_language_codes = ['zh'],
client = speech_v1p1beta1.SpeechClient()
config = speech_v1p1beta1.RecognitionConfig(
encoding=speech_v1p1beta1.RecognitionConfig.AudioEncoding.LINEAR16,
sample_rate_hertz=SAMPLE_RATE,
language_code="en-US",
max_alternatives=1,
enable_word_time_offsets=True,
use_enhanced=True,
model="default",
)
# model="video",
# diarization_config=diarization_config,
streaming_config = speech_v1p1beta1.StreamingRecognitionConfig(
config=config, interim_results=True, single_utterance=False)
# send data of the start of the speech recognition
import requests
url = "http://localhost:3000/api/zoom/recog_start"
now = datetime.datetime.now()
recog_start = now.strftime('%Y-%m-%d %H:%M:%S')
offset_time_start = time.time()
data = {"recog_start": recog_start}
res = requests.post(url, json=data)
mic_manager = ResumableMicrophoneStream(SAMPLE_RATE, CHUNK_SIZE)
print(mic_manager.chunk_size)
sys.stdout.write(YELLOW)
sys.stdout.write('\nListening, say "Quit" or "Exit" to stop.\n\n')
sys.stdout.write("End (ms) Transcript Results/Status\n")
sys.stdout.write("=====================================================\n")
with mic_manager as stream:
while not stream.closed:
new_connection_index += 1
sys.stdout.write(YELLOW)
sys.stdout.write("\n" +
str(STREAMING_LIMIT * stream.restart_counter) +
": NEW REQUEST\n")
print("#### new_connection_index: {}".format(new_connection_index))
stream.audio_input = []
audio_generator = stream.generator()
requests = (speech_v1p1beta1.StreamingRecognizeRequest(
audio_content=content) for content in audio_generator)
responses = client.streaming_recognize(streaming_config, requests)
# Now, put the transcription responses to use.
listen_print_loop(responses, stream, new_connection_index)
if stream.result_end_time > 0:
stream.final_request_end_time = stream.is_final_end_time
stream.result_end_time = 0
stream.last_audio_input = []
stream.last_audio_input = stream.audio_input
stream.audio_input = []
stream.restart_counter = stream.restart_counter + 1
if not stream.last_transcript_was_final:
sys.stdout.write("\n")
stream.new_stream = True
def __init__(self, config, link_db):
self.client = speech.SpeechClient()
def speech_to_text(audio_path, SPEECHTOTEXT_SPEAKER_COUNT):
# Instantiates a client
client = speech.SpeechClient()
# Loads the audio into memory
with io.open(audio_path, 'rb') as audio_file:
content = audio_file.read()
audio = types.RecognitionAudio(content=content)
config = types.RecognitionConfig(
encoding=enums.RecognitionConfig.AudioEncoding.LINEAR16,
# encoding=enums.RecognitionConfig.AudioEncoding.ENCODING_UNSPECIFIED,
sample_rate_hertz=16000,
language_code='en-US',
enable_word_time_offsets=True,
enable_speaker_diarization=True,
diarization_speaker_count=SPEECHTOTEXT_SPEAKER_COUNT,
audio_channel_count=2,
# model='video',
)
# async longrunnning audio file to text
operation = client.long_running_recognize(config, audio)
# Detects speech in the audio file
print("Waiting for operation to complete...")
# response = client.recognize(config, audio)
response = operation.result(timeout=90)
'''
Each result is for a consecutive portion of the audio. Iterate through
them to get the transcripts for the entire audio file.
'''
words_list = []
# for result in response.results:
result = response.results[-1]
alternative = result.alternatives[0]
print(u'Transcript: {}'.format(alternative.transcript))
print('Confidence: {}'.format(alternative.confidence))
for word_info in alternative.words:
word = word_info.word
start_time = word_info.start_time
start_secs = start_time.seconds + start_time.nanos * 1e-9
end_time = word_info.end_time
end_secs = end_time.seconds + end_time.nanos * 1e-9
print('Word: {}, start_time: {}, end_time: {}, speaker_tag: {}'.format(
word,
start_time.seconds + start_time.nanos * 1e-9,
end_time.seconds + end_time.nanos * 1e-9,
word_info.speaker_tag,
))
words_list.append({
'value': word_info.word,
'start_secs': start_secs,
'end_secs': end_secs,
'speaker_tag': word_info.speaker_tag,
})
return words_list
class Transcriber(TranscriberBaseClass):
name = NAME
SUPPORTED_FORMATS = ["flac"]
cost_per_15_seconds = [0.004, 0.006, 0.009]
no_config_error_message = (
"Please sign up for the Google Speech-to-Text API "
"and put the path to your credentials in an "
'environment variable "GOOGLE_APPLICATION_CREDENTIALS"')
transcript_type = TRANSCRIPT_TYPE
# https://cloud.google.com/speech-to-text/docs/languages
# Array.from(document.querySelector('.devsite-table-wrapper').querySelectorAll('table tr')).slice(1).map(row => row.children[1].innerText)
_language_list = [
"af-ZA",
"am-ET",
"hy-AM",
"az-AZ",
"id-ID",
"ms-MY",
"bn-BD",
"bn-IN",
"ca-ES",
"cs-CZ",
"da-DK",
"de-DE",
"en-AU",
"en-CA",
"en-GH",
"en-GB",
"en-IN",
"en-IE",
"en-KE",
"en-NZ",
"en-NG",
"en-PH",
"en-SG",
"en-ZA",
"en-TZ",
"en-US",
"es-AR",
"es-BO",
"es-CL",
"es-CO",
"es-CR",
"es-EC",
"es-SV",
"es-ES",
"es-US",
"es-GT",
"es-HN",
"es-MX",
"es-NI",
"es-PA",
"es-PY",
"es-PE",
"es-PR",
"es-DO",
"es-UY",
"es-VE",
"eu-ES",
"fil-PH",
"fr-CA",
"fr-FR",
"gl-ES",
"ka-GE",
"gu-IN",
"hr-HR",
"zu-ZA",
"is-IS",
"it-IT",
"jv-ID",
"kn-IN",
"km-KH",
"lo-LA",
"lv-LV",
"lt-LT",
"hu-HU",
"ml-IN",
"mr-IN",
"nl-NL",
"ne-NP",
"nb-NO",
"pl-PL",
"pt-BR",
"pt-PT",
"ro-RO",
"si-LK",
"sk-SK",
"sl-SI",
"su-ID",
"sw-TZ",
"sw-KE",
"fi-FI",
"sv-SE",
"ta-IN",
"ta-SG",
"ta-LK",
"ta-MY",
"te-IN",
"vi-VN",
"tr-TR",
"ur-PK",
"ur-IN",
"el-GR",
"bg-BG",
"ru-RU",
"sr-RS",
"uk-UA",
"he-IL",
"ar-IL",
"ar-JO",
"ar-AE",
"ar-BH",
"ar-DZ",
"ar-SA",
"ar-IQ",
"ar-KW",
"ar-MA",
"ar-TN",
"ar-OM",
"ar-PS",
"ar-QA",
"ar-LB",
"ar-EG",
"fa-IR",
"hi-IN",
"th-TH",
"ko-KR",
"zh-TW",
"yue-Hant-HK",
"ja-JP",
"zh-HK",
"zh",
]
if _check_for_config():
speech_client = speech.SpeechClient()
storage_client = storage.Client()
transcript_bucket = storage_client.get_bucket(BUCKET_NAME_TRANSCRIPT)
def __init__(self, filepath):
super().__init__(filepath)
@classmethod
def _setup(cls):
super()._setup()
if not shutil.which("gsutil"):
raise exceptions.DependencyRequired(
"Please install gcloud using the steps here:"
"https://cloud.google.com/storage/docs/gsutil_install")
cls._make_bucket_if_doesnt_exist(BUCKET_NAME_TRANSCRIPT)
@classmethod
def _make_bucket_if_doesnt_exist(cls, bucket_name):
try:
cls.storage_client.create_bucket(bucket_name)
except gc_exceptions.Conflict:
# this might fail if a bucket by the name exists *anywhere* on GCS?
return
else:
print("made Google Cloud Storage Bucket for transcripts")
def convert_file_format_if_needed(self):
if self.file_format not in self.SUPPORTED_FORMATS:
if not shutil.which("ffmpeg"):
raise exceptions.DependencyRequired("please install ffmpeg")
self.filepath = helpers.convert_file(self.filepath, "flac")
@property
def file_format(self):
return pathlib.Path(self.filepath).suffix[1:].lower()
@staticmethod
def check_for_config() -> bool:
return _check_for_config()
def upload_file_if_too_big(self):
"""10MB limit as of Mar 7, 2019"""
pass
def transcribe(self, **kwargs) -> str:
self.convert_file_format_if_needed()
self.upload_file_if_too_big()
self._request_transcription(**kwargs)
def _check_if_transcript_exists(self, transcript_name=None):
return storage.Blob(bucket=self.transcript_bucket,
name=transcript_name
or self.basename).exists(self.storage_client)
def _request_transcription(
self,
language_code="en-US",
enable_automatic_punctuation=True,
enable_speaker_diarization=True,
num_speakers=2,
model="phone_call",
use_enhanced=True,
) -> str:
"""Returns the job_name"""
if self._check_if_transcript_exists():
raise exceptions.AlreadyExistsError(
f"{self.basename} already exists on {NAME}")
num_audio_channels = helpers.get_num_audio_channels(self.filepath)
sample_rate = helpers.get_sample_rate(self.filepath)
with io.open(self.filepath, "rb") as audio_file:
content = audio_file.read()
audio = speech.types.RecognitionAudio(content=content)
if language_code != "en-US":
model = None
config = speech.types.RecognitionConfig(
encoding=speech.enums.RecognitionConfig.AudioEncoding.FLAC,
sample_rate_hertz=sample_rate,
audio_channel_count=num_audio_channels,
enable_separate_recognition_per_channel=True,
enable_word_confidence=True,
enable_word_time_offsets=True,
language_code=language_code,
enable_automatic_punctuation=enable_automatic_punctuation,
enable_speaker_diarization=enable_speaker_diarization,
diarization_speaker_count=num_speakers,
model=model,
use_enhanced=use_enhanced,
)
self.operation = self.speech_client.long_running_recognize(
config, audio)
print("transcribing...")
while not self.operation.done():
sleep(1)
print(".")
result_list = []
for result in self.operation.result().results:
result_list.append(str(result))
print("saving transcript")
transcript_path = "/tmp/transcript.txt"
with open(transcript_path, "w") as fout:
fout.write("\n".join(result_list))
print("uploading transcript")
self.upload_file(BUCKET_NAME_TRANSCRIPT, transcript_path)
os.remove(transcript_path)
return self.basename
@classmethod
def retrieve_transcript(cls,
transcription_job_name: str) -> TRANSCRIPT_TYPE:
"""Get transcript from BUCKET_NAME_TRANSCRIPT"""
if not cls._check_if_transcript_exists(
cls, transcript_name=transcription_job_name):
raise exceptions.DoesntExistError("no such transcript!")
blob = cls.transcript_bucket.blob(transcription_job_name)
f = tempfile.NamedTemporaryFile(delete=False)
f.close()
blob.download_to_filename(f.name)
with open(f.name) as fin:
transcript_text = fin.read()
os.remove(f.name)
return transcript_text
def upload_file(self, bucket_name, path):
blob = self.transcript_bucket.blob(self.basename)
blob.upload_from_filename(path)
@classmethod
def get_transcription_jobs(cls,
job_name_query=None,
status=None) -> List[dict]:
if status and status.lower() != "completed":
return []
jobs = []
for t in cls.transcript_bucket.list_blobs():
if job_name_query is not None and t.name != job_name_query:
continue
jobs.append({"name": t.name, "status": "COMPLETED"})
return jobs
def build_dataset(self):
print("running")
output_wav_path = "{}/wavs/".format(self.project_name)
if not os.path.exists(self.project_name):
os.mkdir(self.project_name)
if not os.path.exists(output_wav_path):
os.mkdir(output_wav_path)
if self.split_method == 0:
#Google API mode
if not get_value("input_project_name") or not get_value(
"label_wav_file_path"):
print("Error, please choose text and/or audio files.")
return
set_value("label_build_status",
"Detecting silences. This may take several minutes...")
audio_name = self.wav_file_path
w = AudioSegment.from_wav(audio_name)
s_len = 1000
silence_cuts = silence.split_on_silence(w,
min_silence_len=s_len,
silence_thresh=-45,
keep_silence=True)
cuts = []
final_cuts = []
def split_wav(wav, l):
if (wav.duration_seconds * 1000) < (self.cut_length * 1000):
output = []
output.append(wav)
return output
too_long = False
while True:
l -= 50
if l == 0:
print(
"Error, could not find small enough silence period for split, giving up"
)
output = []
output.append(wav)
return output
splits = silence.split_on_silence(wav,
min_silence_len=l,
silence_thresh=-45,
keep_silence=True)
print("Trying resplit...")
for s in splits:
if (s.duration_seconds * 1000) > (self.cut_length *
1000):
too_long = True
if too_long == True:
too_long = False
else:
return splits
# Keep splitting until all cuts are under max len
for i, c in enumerate(silence_cuts):
print(f"Checking phrase {i}...")
c_splits = split_wav(c, 1000)
for s in c_splits:
cuts.append(s)
# c_split_len = 1
# s_len_temp = s_len - 100
# for c in silence_cuts:
# if (c.duration_seconds * 1000) > (self.cut_length * 1000):
# # cut again, too long
# #print("cutting again...")
# while c_split_len == 1:
# #print(s_len_temp)
# c_split = split_wav(c, s_len_temp)
# c_split_len = len(c_split)
# s_len_temp -= 100 #reduce split time for hopefully more cuts
# c_split_len = 1
# s_len_temp = s_len - 100
# for i in c_split:
# cuts.append(i)
# else:
# cuts.append(c)
# rebuild small cuts into larger, but below split len
temp_cuts = AudioSegment.empty()
prev_cuts = AudioSegment.empty()
for i, c in enumerate(cuts):
prev_cuts = temp_cuts
temp_cuts = temp_cuts + c
if i == (len(cuts) - 1):
#on final entry
if (temp_cuts.duration_seconds * 1000) > (self.cut_length *
1000):
final_cuts.append(prev_cuts)
final_cuts.append(c)
else:
final_cuts.append(temp_cuts)
else:
if ((temp_cuts.duration_seconds * 1000) +
(cuts[i + 1].duration_seconds * 1000)) > (
self.cut_length * 1000):
# combine failed, too long, add what has already been concatenated
final_cuts.append(temp_cuts)
temp_cuts = AudioSegment.empty()
if not os.path.exists("{}/wavs".format(self.project_name)):
os.mkdir("{}/wavs".format(self.project_name))
for i, w in enumerate(final_cuts):
w.export("{}/wavs/{}.wav".format(
self.project_name,
i + int(get_value("input_starting_index"))),
format="wav")
# Process each cut into google API and add result to csv
with open("{}/output.csv".format(self.project_name), 'w') as f:
bucket_name = get_value("input_storage_bucket")
newline = ''
for i, c in enumerate(final_cuts):
x = i + int(get_value("input_starting_index"))
print(f"Transcribing entry {x}")
self.upload_blob(
bucket_name,
"{}/wavs/{}.wav".format(self.project_name,
x), "temp_audio.wav")
gcs_uri = "gs://{}/temp_audio.wav".format(bucket_name)
client = speech.SpeechClient()
audio = speech.RecognitionAudio(uri=gcs_uri)
info = mediainfo("{}/wavs/{}.wav".format(
self.project_name, x))
sample_rate = info['sample_rate']
if get_value("input_use_videomodel") == 1:
print("Using enchanced google model...")
config = speech.RecognitionConfig(
encoding=speech.RecognitionConfig.AudioEncoding.
LINEAR16,
sample_rate_hertz=int(sample_rate),
language_code="en-US",
enable_automatic_punctuation=True,
enable_word_time_offsets=False,
enable_speaker_diarization=False,
# enhanced model for better performance?
use_enhanced=True,
model="video", #"phone_call or video"
)
else:
config = speech.RecognitionConfig(
encoding=speech.RecognitionConfig.AudioEncoding.
LINEAR16,
sample_rate_hertz=int(sample_rate),
language_code="en-US",
enable_automatic_punctuation=True,
enable_word_time_offsets=False,
enable_speaker_diarization=False,
)
operation = client.long_running_recognize(config=config,
audio=audio)
response = operation.result(timeout=28800)
for result in response.results:
text = result.alternatives[0].transcript
# replace some symbols and google API word choice
text = text.replace("%", " percent")
text = text.replace("cuz", "cause")
text = text.replace("-", " ")
text = text.replace("&", "and")
print(text)
set_value("label_build_status", text)
f.write("{}wavs/{}.wav|{}".format(newline, x, text))
newline = '\n'
print('\a') #system beep
set_value("label_build_status", "Done!")
print("Done running builder!")
else:
# Aeneas mode
if not get_value("input_project_name") or not get_value(
"label_speaker_text_path") or not get_value(
"label_wav_file_path"):
print("Error, please choose text and/or audio files.")
return
if not os.path.exists("aeneas_out"):
os.mkdir("aeneas_out")
else:
shutil.rmtree("aeneas_out")
os.mkdir("aeneas_out")
if not os.path.exists("aeneas_prepped"):
os.mkdir("aeneas_prepped")
else:
shutil.rmtree("aeneas_prepped")
os.mkdir("aeneas_prepped")
audio_name = self.wav_file_path
with open(self.speaker_text_path, 'r', encoding="utf8") as f:
text = f.read()
text = text.replace(';', '.')
text = text.replace(':', '.')
text = text.replace('-', ' ')
text = text.replace('”', '')
text = text.replace('“', '')
text = text.replace('"', '.')
text = text.replace('—', ' ')
text = text.replace('’', '\'')
text = text.replace(' –', '.')
text = text.strip('\n')
if self.contains_punc:
#remove any duplicate whitespace between words
text = " ".join(text.split())
phrase_splits = re.split(
r'(?<=[\.\!\?])\s*',
text) #split on white space between sentences
phrase_splits = list(filter(
None, phrase_splits)) #remove empty splits
else:
#no punctuation from speech to text, so we must divid text by word count
phrase_splits = []
temp_line = []
text_split = text.split()
word_count_limit = 16
while len(text_split) > 0:
while len(temp_line) < word_count_limit and len(
text_split) > 0:
temp_line.append(text_split.pop(0))
phrase_splits.append(" ".join(temp_line))
temp_line = []
with open('aeneas_prepped/split_text', 'w') as f:
newline = ''
for s in phrase_splits:
if s:
stripped = s.strip() #remove whitespace
f.write(newline + stripped)
newline = '\n'
#os.system('python -m aeneas.tools.execute_task ' + audio_name + ' aeneas_prepped/split_text "task_adjust_boundary_percent_value=50|task_adjust_boundary_algorithm=percent|task_language=en|is_text_type=plain|os_task_file_format=csv" ' + 'aeneas_out/' + audio_name_no_ext + '.csv')
os.system(
'python -m aeneas.tools.execute_task ' + audio_name +
' aeneas_prepped/split_text "task_adjust_boundary_percent_value=50|task_adjust_boundary_algorithm=percent|task_language=en|is_text_type=plain|os_task_file_format=csv" '
+ 'aeneas_out/' + self.project_name + '.csv')
output_exists = False
if os.path.exists("{}/output.csv".format(self.project_name)):
#if file exists then prepare for append
output_exists = True
new_csv_file = open("{}/output.csv".format(self.project_name),
'a')
if output_exists:
new_csv_file.write("\n")
with open('aeneas_out/' + self.project_name + '.csv',
'r') as csv_file:
index_count = int(self.index_start)
csv_reader = csv.reader(csv_file, delimiter=',')
csv_reader = list(csv_reader) #convert to list
row_count = len(csv_reader)
newline = ""
for row in csv_reader:
beginning_cut = float(row[1])
end_cut = float(row[2])
text_out = row[3]
text_out = text_out.strip()
print("{} {} {} ".format(beginning_cut, end_cut,
text_out))
c_length = end_cut - beginning_cut
#if cut is longer than cut length then split it even more
cut_length = float(self.cut_length)
if c_length > cut_length:
more_cuts = open("aeneas_prepped/temp.csv", 'w')
#save the current cut wav file to run on aeneas again
w = AudioSegment.from_wav(audio_name)
wav_cut = w[(beginning_cut * 1000):(end_cut *
1000)]
wav_cut.export("aeneas_prepped/tempcut.wav",
format="wav")
split_list = []
num_cuts = math.ceil(c_length / cut_length)
text_list = text_out.split()
text_list_len = len(text_list)
split_len = math.ceil(text_list_len / num_cuts)
print(
"too long, making extra {} cuts. with length {}"
.format(num_cuts, split_len))
for i in range(1, num_cuts + 1):
words = []
for j in range(0, split_len):
if not text_list:
break
words.append(text_list.pop(0))
split_list.append(" ".join(words))
print(split_list)
print()
newline_splits = ''
for phrase in split_list:
more_cuts.write(newline_splits + phrase)
newline_splits = '\n'
more_cuts.close()
os.system(
'python -m aeneas.tools.execute_task ' +
"aeneas_prepped/tempcut.wav" +
' aeneas_prepped/temp.csv "task_adjust_boundary_percent_value=50|task_adjust_boundary_algorithm=percent|task_language=en|is_text_type=plain|os_task_file_format=csv" '
+ 'aeneas_out/temp_out.csv')
csv_file_temp = open('aeneas_out/temp_out.csv',
'r')
csv_reader_temp = csv.reader(csv_file_temp,
delimiter=',')
csv_reader_temp = list(
csv_reader_temp) #convert to list
row_count = len(csv_reader_temp)
w = AudioSegment.from_wav(
"aeneas_prepped/tempcut.wav")
for row in csv_reader_temp:
beginning_cut = float(row[1])
end_cut = float(row[2])
text_out = row[3]
text_out = text_out.strip()
wav_cut = w[(beginning_cut * 1000):(end_cut *
1000)]
new_wav_filename = "wavs/" + str(
index_count) + ".wav"
new_csv_file.write("{}{}|{}".format(
newline, new_wav_filename, text_out))
wav_cut.export("{}/{}".format(
self.project_name, new_wav_filename),
format="wav")
index_count += 1
newline = '\n'
csv_file_temp.close()
else:
w = AudioSegment.from_wav(audio_name)
wav_cut = w[(beginning_cut * 1000):(end_cut *
1000)]
new_wav_filename = "wavs/" + str(
index_count) + ".wav"
new_csv_file.write("{}{}|{}".format(
newline, new_wav_filename, text_out))
wav_cut.export("{}/{}".format(
self.project_name, new_wav_filename),
format="wav")
index_count += 1
newline = '\n'
new_csv_file.close()
set_value("label_build_status", "Building dataset done!")
#Remove temporary directories
shutil.rmtree("aeneas_prepped")
shutil.rmtree("aeneas_out")
print('\a') #system beep
print("Done with Aeneas!")
def gen_transcript(filename: str, script_path: str, to_lang: str):
"""generates a transcript"""
client = speech.SpeechClient()
#upload to gcp
uri_ = upload_to_gcp(filename)
audio = speech.types.RecognitionAudio(uri=uri_)
characters, sentences = script_sanitzer.santize(script_path,
['*,*', '[,]', '(,)'])
phrases_ = [x[0] if len(x[0]) < 100 else x[0][:100] for x in sentences]
config = speech.types.RecognitionConfig(encoding='FLAC',
language_code='en-US',
model='video',
sample_rate_hertz=16000,
enable_word_time_offsets=True)
operation = client.long_running_recognize(config, audio)
print('Waiting for operation to complete...')
result_data = operation.result(timeout=1000)
merged_transcript = ""
merged_words = []
for result in result_data.results:
alternative = result.alternatives[0]
merged_transcript += alternative.transcript
for word_info in alternative.words:
word = word_info.word
start_time = word_info.start_time
end_time = word_info.end_time
word_tup = (word, start_time.seconds + start_time.nanos * 1e-9,
end_time.seconds + end_time.nanos * 1e-9)
# print('Word: {}, start_time: {}, end_time: {}'.format(
# word,
# start_time.seconds + start_time.nanos * 1e-9,
# end_time.seconds + end_time.nanos * 1e-9))
merged_words.append(word_tup)
#santize the script
print(merged_words)
empty_queue = []
transcript_ptr = 0
start = -1.1
end = -1.1
prev_start = start
prev_end = end
for sentence in sentences:
actualSize = findSize(sentence[0])
print(transcript_ptr)
prev_start = start
prev_end = end
start = -1.0
end = -1.0
found = False
for word in sentence[0].split(" "):
if word.isspace():
continue
if (found):
break
for word2 in merged_words[transcript_ptr:transcript_ptr +
actualSize]:
#find start
if check_words_equal(word, word2[0]):
start = word2[1]
found = True
break
found = False
for word in sentence[0].split(" ")[::-1]:
print("WORD: " + str(word))
if word.isspace():
continue
if (found):
break
for word2 in range(transcript_ptr + actualSize, transcript_ptr - 1,
-1):
if (word2 >= len(merged_words)):
continue
print(actualSize)
print(sentence[0].split(" "))
#find start
print(word2)
print(len(merged_words))
print("WORD 2: " + str(merged_words[word2][0]))
if check_words_equal(word, merged_words[word2][0]):
end = merged_words[word2][2]
transcript_ptr = word2 + 1
found = True
break
#Could not find the correct start or end times for first and last words
#Time to estimate!
if start < 0 or end < 0:
'''
We know that, if all previous sentences were calculated correctly,
The start and end time of this sentence must be after the previous
end time of the last sentence (somewhere near the first word after the last sentence)
or 0 if its the first sentence. Once we have the start we will calculate the average
talking speed (wpm) of the characters. Using this speed we can define a low ball
estimate for how long the sentence that couldnt be defined will take, allowing us
to define the end time. If this is the first sentence we will attempt to use the
average persons wpm (150 wpm).
'''
#No previous sentences
if len(empty_queue) == 0:
start = merged_words[0][1]
end = actualSize * (14 / 6)
transcript_ptr = actualSize - int(actualSize * 1 / 4)
else:
start = merged_words[transcript_ptr][1]
avg_wpm = findAverageWPM(empty_queue)
end = actualSize * avg_wpm
transcript_ptr += actualSize - int(actualSize * 1 / 4)
else:
#create nodes
node_to_add = Node(
sentence[1], translate.translate_phrase(sentence[0], to_lang),
start, end)
empty_queue.append(node_to_add)
print(empty_queue)
return empty_queue
def __init__(self):
self._sound_to_text = speech.SpeechClient()
self._text_to_sound = texttospeech.TextToSpeechClient()
self._audio_config = texttospeech.types.AudioConfig(
audio_encoding=texttospeech.enums.AudioEncoding.MP3)
self._translator = translate.Client()
def __init__(self):
threading.Thread.__init__(self)
# 语音自适应
boost = 4 # 识别指定词语的概率 推荐 [0,20]
speech_contexts = [{
"phrases": "ホロライブ",
"boost": boost
}, {
"phrases": "しらかみ",
"boost": boost
}, {
"phrases": "ふぶき",
"boost": boost
}, {
"phrases": "うさだ",
"boost": boost
}, {
"phrases": "ぺこら",
"boost": boost
}, {
"phrases": "ぺこ",
"boost": boost
}, {
"phrases": "よし",
"boost": boost
}, {
"phrases": "よしょ",
"boost": boost
}, {
"phrases": "えしょう",
"boost": boost
}, {
"phrases": "ARK",
"boost": boost
}, {
"phrases": "やめろ",
"boost": boost
}, {
"phrases": "マリン",
"boost": boost
}, {
"phrases": "まつり",
"boost": boost
}, {
"phrases": "せんちょう",
"boost": boost
}]
speaker_diarization_config = speech.types.SpeakerDiarizationConfig( # 区分讲话人配置
enable_speaker_diarization=True,
min_speaker_count=1,
max_speaker_count=2)
self.config = speech.types.RecognitionConfig(
encoding=speech.enums.RecognitionConfig.AudioEncoding.LINEAR16,
sample_rate_hertz=44100,
language_code='ja-JP',
max_alternatives=1,
enable_automatic_punctuation=True, # 启用标点符号
diarization_config=speaker_diarization_config, # 区分讲话人
speech_contexts=speech_contexts # 语音自适应
)
self.streaming_config = speech.types.StreamingRecognitionConfig(
config=self.config, interim_results=True)
self.translate_client = translate.Client()
self.speech_client = speech.SpeechClient()
def __init__(self, sample_rate):
self.client = speech_v1p1beta1.SpeechClient()
self.sample_rate = sample_rate
def get_hypothesis(self, uri, configuration):
import time
"""Asynchronously transcribes the audio uri specified by the gcs_uri."""
client = speech.SpeechClient()
config = {
"model":
configuration.get_model(),
"use_enhanced":
configuration.get_use_enhanced(),
"encoding":
configuration.get_encoding(),
"sample_rate_hertz":
configuration.get_sample_rate_hertz(),
"language_code":
configuration.get_language_code(),
"alternative_language_codes":
configuration.get_alternative_language_codes(),
"audio_channel_count":
configuration.get_audio_channel_count(),
"enable_separate_recognition_per_channel":
configuration.get_enable_separate_recognition_per_channel(),
"enable_speaker_diarization":
configuration.get_enableSpeakerDiarization(),
"diarization_speaker_count":
configuration.get_diarizationSpeakerCount(),
"enable_automatic_punctuation":
configuration.get_enableAutomaticPunctuation(),
"speech_contexts":
configuration.get_speech_context()
}
audio = {"uri": uri}
operation = object
try:
operation = client.long_running_recognize(config=config,
audio=audio)
except google.api_core.exceptions.InvalidArgument as e:
raise e
count = 0
sleep_time = 5
while not operation.done() and count != 30000:
print(
f"{operation.metadata.progress_percent}% complete - updates every {sleep_time} seconds"
)
if count == 29999:
raise TimeoutError("Time out processing audio")
count += 1
time.sleep(sleep_time)
print(
f"{operation.metadata.progress_percent}% complete - updates every {sleep_time} seconds"
)
response = operation.result(timeout=1200)
transcript = str()
for result in response.results:
# First alternative is the most probable result
transcript += " " + result.alternatives[0].transcript
if not transcript:
logger.debug('No transcript returned')
utilities = Utilities()
t = utilities.strip_puc(text=transcript)
return t.lower()
def google_transcribe(audio_file_name):
file_name = filepath + audio_file_name
second_lang = "hi-IN"
# The name of the audio file to transcribe
frame_rate, channels = frame_rate_channel(file_name)
if channels > 1:
stereo_to_mono(file_name)
bucket_name = bucketname
source_file_name = filepath + audio_file_name
destination_blob_name = audio_file_name
upload_blob(bucket_name, source_file_name, destination_blob_name)
gcs_uri = 'gs://' + bucketname + '/' + audio_file_name
transcript = ''
credential_path = s.get("credential_path")
os.environ['GOOGLE_APPLICATION_CREDENTIALS'] = credential_path
client = speech.SpeechClient()
audio = speech.RecognitionAudio(uri=gcs_uri)
config = speech.RecognitionConfig(
encoding=speech.RecognitionConfig.AudioEncoding.LINEAR16,
sample_rate_hertz=frame_rate,
language_code='en-US',
alternative_language_codes=[second_lang],
enable_speaker_diarization=True,
diarization_speaker_count=2)
operation = client.long_running_recognize(request={
"config": config,
"audio": audio
})
response = operation.result(timeout=10000)
result = response.results[-1]
words_info = result.alternatives[0].words
tag = 1
speaker = ""
for word_info in words_info:
if word_info.speaker_tag == tag:
speaker = speaker + " " + word_info.word
else:
transcript += "speaker {}: {}".format(tag, speaker) + '\n'
tag = word_info.speaker_tag
speaker = "" + word_info.word
transcript += "speaker {}: {}".format(tag, speaker)
#for result in response.results:
#transcript += result.alternatives[0].transcript
storage_client = storage.Client()
bucket_name = storage_client.get_bucket(bucket_name)
transcript_filename = audio_file_name.split(
'.')[0] + '_transcript' + '.txt'
blob_transcript_file = bucket_name.blob(transcript_filename)
blob_transcript_file.upload_from_string(transcript)
#delete_blob(bucket_name, destination_blob_name)
return transcript
def transcriberDetail(blob_name, main):
# check if already Inserted to ssrDictionary using audio name/blobname
flagDntInst = 0
# mycursor = mydb.cursor()
mydb._open_connection()
sql = "select audioName from ssrDictionary where audioName='" + blob_name + "' LIMIT 2"
mycursor.execute(sql)
myresult = mycursor.fetchall()
for x in myresult:
flagDntInst = 1
posts = []
# urll = 'gs://bucketgcssr/SSR_8102019114925.wav'
urll = 'gs://bucketgcssr/' + blob_name
from google.cloud import speech_v1p1beta1 as speech # GCP api
client = speech.SpeechClient()
audio = speech.types.RecognitionAudio(uri=urll)
config = speech.types.RecognitionConfig(
encoding=speech.enums.RecognitionConfig.AudioEncoding.LINEAR16,
language_code='ur-PK', # language code
# speaker diaraziation not working for urdu for now
enable_speaker_diarization=True,
diarization_speaker_count=2, # Speak count not working for urdu now
sample_rate_hertz=48000, # audio sampel rage
audio_channel_count=1) # number of chanel used in aud
operation = client.long_running_recognize(config, audio)
response = operation.result(timeout=10000)
transcrip = ""
confidence = 0
for result in response.results:
alternative = result.alternatives[0]
transcrip = format(alternative.transcript)
confidence = alternative.confidence
main.append({
'transcrip': transcrip,
'blob_name': blob_name,
'confidence': confidence
})
for word_info in alternative.words:
confidence = format(word_info.confidence)
word = word_info.word
start_time = word_info.start_time
end_time = word_info.end_time
posts.append({
'word': word,
'start_time': start_time.seconds + start_time.nanos * 1e-9,
'end_time': end_time.seconds + end_time.nanos * 1e-9,
'confidence': confidence
})
# insertion to Mysql For WordDictionary here
if flagDntInst == 0:
sql = "INSERT INTO ssrDictionary (words,audioName, confidance,endTime,startTime) VALUES (%s, %s, %s, %s, %s)"
val = (word, blob_name, confidence,
end_time.seconds + end_time.nanos * 1e-9,
start_time.seconds + start_time.nanos * 1e-9)
mycursor.execute(sql, val)
mydb.commit()
mydb.close()
return posts
def transcribe_gcs(gcs_uri, num_speakers):
"""Asynchronously transcribes the audio file specified by the gcs_uri."""
# Imports the Google Cloud client library
#from google.cloud import speech
from google.cloud import speech_v1p1beta1 as speech
# Instantiates a client
client = speech.SpeechClient()
# Construct a recognition metadata object
metadata = speech.RecognitionMetadata()
metadata.interaction_type = speech.RecognitionMetadata.InteractionType.DISCUSSION
metadata.recording_device_type = (
speech.RecognitionMetadata.RecordingDeviceType.OTHER_INDOOR_DEVICE
)
metadata.audio_topic = "court trial hearing"
metadata.original_mime_type = "audio/mp3"
audio = speech.RecognitionAudio(uri=gcs_uri)
config = speech.RecognitionConfig(
encoding=speech.RecognitionConfig.AudioEncoding.FLAC,
sample_rate_hertz=16000,
language_code="en-US",
enable_automatic_punctuation=True,
enable_speaker_diarization=True,
diarization_speaker_count=num_speakers,
# Enhanced models cost more than standard models.
use_enhanced=True,
model="video",
enable_word_time_offsets=True,
)
# Detects speech in the audio file -- long audio file
operation = client.long_running_recognize(config=config, audio=audio)
print("Waiting for operation to complete...")
response = operation.result(timeout=300)
# Writing results to json
result_counter = 0
word_counter = 0
output_json = {}
for result in response.results:
alternative = result.alternatives[0]
output_json[f"{result_counter}_Transcript"] = alternative.transcript
output_json[f"{result_counter}_Confidence"] = alternative.confidence
result_counter += 1
for word_info in alternative.words:
word = word_info.word
start_time = word_info.start_time
end_time = word_info.end_time
speaker_tag = word_info.speaker_tag
output_json[f"{word_counter}_Word"] = word
output_json[f"{word_counter}_start_time"] = start_time.total_seconds()
output_json[f"{word_counter}_end_time"] = end_time.total_seconds()
output_json[f"{word_counter}_speaker_tag"] = speaker_tag
word_counter += 1
with open("{}.json".format(gcs_uri.split('/')[-1][:-5]) , "w+") as file:
json.dump(output_json, file)
print("Dirized and transcribed {}".format(gcs_uri.split('/')[-1]))
def manage_stream(mic, finals, STREAMING_LIMIT):
client = speech.SpeechClient()
config = speech.types.RecognitionConfig(
encoding=speech.enums.RecognitionConfig.AudioEncoding.LINEAR16,
sample_rate_hertz=44100,
language_code='en-US',
max_alternatives=1,
enable_speaker_diarization=True,
enable_automatic_punctuation=True)
streaming_config = speech.types.StreamingRecognitionConfig(
config=config, interim_results=True)
with mic as stream:
while not stream.to_close:
audio_generator = stream.generator()
requests = (speech.types.StreamingRecognizeRequest(
audio_content=content) for content in audio_generator)
responses = client.streaming_recognize(streaming_config, requests)
logging.info("Started new stream")
for response in responses:
logging.info("new response")
if stream.to_close:
break
if get_current_time() - stream.start_time > STREAMING_LIMIT:
stream.start_time = get_current_time()
break
if not response.results:
continue
result = response.results[0]
if not result.alternatives:
continue
if result.is_final:
transcript = result.alternatives[0].transcript
logging.info(transcript)
result_seconds = 0
result_nanos = 0
if result.result_end_time.seconds:
result_seconds = result.result_end_time.seconds
if result.result_end_time.nanos:
result_nanos = result.result_end_time.nanos
stream.end_time = int((result_seconds * 1000) +
(result_nanos / 1000000))
finals.append(
f"Speaker {get_main_speaker(result)}: {transcript}<br/><br/>"
)
stream.last_interim = ""
else:
stream.last_interim = result.alternatives[0].transcript
stream.next_stream()
def transcribe_file(speech_file):
"""Transcribe the given audio file asynchronously."""
from google.cloud import speech_v1p1beta1
from google.cloud.speech_v1p1beta1 import enums
from google.cloud.speech_v1p1beta1 import types
client = speech_v1p1beta1.SpeechClient()
# [START migration_async_request]
with io.open(speech_file, 'rb') as audio_file:
content = audio_file.read()
#convert to mono
from pydub import AudioSegment
AudioSegment.converter = r'C:\FFMpeg'
sound = AudioSegment.from_wav(args.path)
sound = sound.set_frame_rate(44100)
sound = sound.set_channels(1)
sound.export(args.path, format="wav")
#from pydub.utils import mediainfo
#info = mediainfo(args.path)
#print (info['sample_rate'])
from scipy import io
rate = scipy.io.wavfile.read(args.path)
print(rate)
print('Using ', args.path, ', with the below config:')
print("")
print("importing speech_v1p1beta1")
print("language_code='en-US'")
print("use_enhanced=True")
print("enable_automatic_punctuation=False")
print("enable_word_time_offsets=False")
print("profanity_filter=True")
print("sample_rate=48000hz")
print("")
print("Transcript is as follows")
audio = types.RecognitionAudio(content=content)
config = types.RecognitionConfig(
encoding=enums.RecognitionConfig.AudioEncoding.LINEAR16,
sample_rate_hertz=48000,
language_code='en-US',
use_enhanced=True,
# A model must be specified to use enhanced model.
model='phone_call',
enable_automatic_punctuation=False,
enable_word_time_offsets=False,
profanity_filter=True,
#speech_contexts=[speech.types.SpeechContext(
# phrases=['Andy', 'Wisy', 'EEP', 'Project', 'Tom', 'Jeff'],
# )],
)
# [START migration_async_response]
operation = client.long_running_recognize(config, audio)
# [END migration_async_request]
print('Waiting for operation to complete...')
response = operation.result(timeout=90)
# Each result is for a consecutive portion of the audio. Iterate through
# them to get the transcripts for the entire audio file.
for result in response.results:
# The first alternative is the most likely one for this portion.
print(('Transcript: {}'.format(result.alternatives[0].transcript)))
print(('Confidence: {}'.format(result.alternatives[0].confidence)))
#Reference: https://cloud.google.com/speech-to-text/docs/async-recognize
#https://google-cloud-python.readthedocs.io/en/0.32.0/storage/blobs.html
from google.cloud import speech_v1p1beta1 as speech
from google.cloud.storage import Blob
from google.cloud import storage
client = speech.SpeechClient()
def diarize(data, context):
speech_file = data['name']
bucket = data['bucket']
print('Bucket {}'.format(bucket))
print('File {}'.format(speech_file))
filename_uri = "gs://" + bucket + "/" + speech_file
print('File name uri {}'.format(filename_uri))
dest_file = speech_file + ".txt"
audio = speech.types.RecognitionAudio(uri=filename_uri)
config = speech.types.RecognitionConfig(
encoding=speech.enums.RecognitionConfig.AudioEncoding.LINEAR16,
sample_rate_hertz=8000,
language_code='en-US',
enable_speaker_diarization=True,
diarization_speaker_count=2)
operation = client.long_running_recognize(config, audio)
print('Waiting for operation to complete...')
def main(input_queue, q_conversation, realTime=True, speech_file=None):
global stop_loop
global queueQA
global arr
global queueTranscripts
global queueSpeakerTags
global q_convo
global queueThread
q_convo = q_conversation
max_conv_length = 1000
stop_loop = multiprocessing.Value('i') # used to stop all processes/loops when 'stop recording' is said
stop_loop = 0
queueQA = multiprocessing.Queue()
queueTranscripts = multiprocessing.Queue()
queueSpeakerTags = multiprocessing.Queue()
queueThread = multiprocessing.Queue()
arr = multiprocessing.Array('i', max_conv_length)
process = multiprocessing.Process(target=calculation,
args=(arr, queueQA, queueTranscripts, queueSpeakerTags, input_queue, stop_loop,
queueThread,))
process.start()
os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = "C:/Users/RedLine/Desktop/Semester 8/FYP/FYP_final/FYP-key.json"
if realTime == True:
print("Starting real time process")
client = speech.SpeechClient()
config = speech.types.RecognitionConfig(
encoding=speech.enums.RecognitionConfig.AudioEncoding.LINEAR16,
sample_rate_hertz=SAMPLE_RATE,
language_code='en-US',
enable_speaker_diarization=True,
enable_automatic_punctuation=True,
max_alternatives=1
# enable_word_time_offsets=True
)
streaming_config = speech.types.StreamingRecognitionConfig(
config=config,
interim_results=True)
mic_manager = ResumableMicrophoneStream(SAMPLE_RATE, CHUNK_SIZE)
print('Say "Quit" or "Exit" to terminate the program.')
with mic_manager as stream:
while not stream.closed:
sys.stdout.write('\n' + str(
STREAMING_LIMIT * stream.restart_counter) + ': NEW REQUEST\n')
stream.audio_input = []
audio_generator = stream.generator()
requests = (speech.types.StreamingRecognizeRequest(
audio_content=content) for content in audio_generator)
responses = client.streaming_recognize(streaming_config,
requests)
# Now, put the transcription responses to use.
listen_print_loop(responses, stream)
if stream.result_end_time > 0:
stream.final_request_end_time = stream.is_final_end_time
stream.result_end_time = 0
stream.last_audio_input = []
stream.last_audio_input = stream.audio_input
stream.audio_input = []
stream.restart_counter = stream.restart_counter + 1
if not stream.last_transcript_was_final:
sys.stdout.write('\n')
stream.new_stream = True
else:
print("Starting Non real time process")
client = speech.SpeechClient()
storage_uri = 'gs://fyp_1/BEP313-Scrum-Meetings1_1.wav'
# Sample rate in Hertz of the audio data sent
# sample_rate_hertz = 16000
# The language of the supplied audio
language_code = "en-US"
# Encoding of audio data sent. This sample sets this explicitly.
# This field is optional for FLAC and WAV audio formats.
encoding = enums.RecognitionConfig.AudioEncoding.LINEAR16
config = {
# "sample_rate_hertz": sample_rate_hertz,
"enable_speaker_diarization": True,
"enable_automatic_punctuation": True,
"language_code": language_code,
"encoding": encoding,
"audio_channel_count": 2,
}
audio = {"uri": storage_uri}
operation = client.long_running_recognize(config, audio)
print(u"Waiting for operation to complete...")
response = operation.result()
result = response.results[-1]
x = result.alternatives[0]
words_info = x.words
tag = 1
tag_prev = 1
speaker = ""
transcript = ""
for word_info in words_info:
if tag_prev == tag:
tag_prev = tag
tag = word_info.speaker_tag
speaker = speaker + " " + word_info.word
elif not (speaker[-1] == "." or speaker[-1] == "?"):
speaker = speaker + " " + word_info.word
else:
transcript += "speaker {}: {}".format(tag_prev, speaker) + '\n'
tag_prev = tag
tag = word_info.speaker_tag
speaker = "" + word_info.word
transcript += "speaker {}: {}".format(tag_prev, speaker)
print("transcript_1\n", transcript)
f = open("transcript_1.txt", "w")
f.write(transcript)
f.close()
f = open("transcript_1.txt")
transcript = f.readlines()
print("transcript_2\n", transcript)
f.close()
output = []
for i in transcript:
x = i.split(': ')
sentence = x[-1]
speaker_tag = x[0][-1]
sentences = re.split(', |\. |\? ', sentence)
for j in sentences:
output.append([j.rstrip(), speaker_tag])
print('x: ', output)
print(process_transcripts(output)[:])
def async_transcribe(audio_file_paths,
bucket_name,
output_tsv_path,
sample_rate,
language_code,
speaker_count=0,
begin_sec=0.0):
"""Transcribe a given audio file using the async GCloud Speech-to-Text API.
The async API has the advantage of being able to handler longer audio without
state reset. Empirically, we've observed that the async calls lead to slightly
better accuracy than streaming calls.
Args:
audio_file_paths: Paths to the audio files as a list of strings in the
correct order.
bucket_name: Name of GCS bucket used for holding objects temporarily.
output_tsv_path: Path to the output TSV file.
sample_rate: Audio sample rate.
language_code: Language code for recognition.
speaker_count: Number of speakers. If 0, speaker diarization will be
disabled.
begin_sec: Transcript begin timestamp in seconds.
"""
tmp_audio_file = tempfile.mktemp(suffix=".flac")
print("Temporary audio file: %s" % tmp_audio_file)
audio_duration_s = concatenate_audio_files(audio_file_paths,
tmp_audio_file)
storage_client = storage.Client()
bucket = storage_client.bucket(bucket_name)
destination_blob_name = os.path.basename(tmp_audio_file)
blob = bucket.blob(destination_blob_name)
print("Uploading %s to GCS bucket %s" % (tmp_audio_file, bucket_name))
blob.upload_from_filename(tmp_audio_file)
gcs_uri = "gs://%s/%s" % (bucket_name, destination_blob_name)
print("Uploaded to GCS URI: %s" % gcs_uri)
client = speech.SpeechClient()
audio = speech.RecognitionAudio(uri=gcs_uri)
enable_speaker_diarization = speaker_count > 0
config = speech.RecognitionConfig(
encoding=speech.RecognitionConfig.AudioEncoding.FLAC,
sample_rate_hertz=sample_rate,
language_code=language_code,
enable_speaker_diarization=enable_speaker_diarization,
diarization_speaker_count=speaker_count)
operation = client.long_running_recognize(config=config, audio=audio)
timeout_s = int(audio_duration_s * 0.25)
print("Waiting for async ASR operation to complete "
"(audio duration: %.3f s; ASR timeout: %d s)..." %
(audio_duration_s, timeout_s))
response = operation.result(timeout=timeout_s)
blob.delete()
os.remove(tmp_audio_file)
utterances = []
for result in response.results:
# The first alternative is the most likely one for this portion.
alt = result.alternatives[0]
utterances.append(alt.transcript)
print(u"Transcript: {}".format(alt.transcript))
diarized_words = [(word.word, word.speaker_tag,
word.start_time.total_seconds(),
word.end_time.total_seconds())
for word in alt.words]
# print("Confidence: {}".format(result.alternatives[0].confidence))
regrouped_utterances = regroup_utterances(utterances, diarized_words)
with open(output_tsv_path, "w" if not begin_sec else "a") as f:
if not begin_sec:
# Write the TSV header.
f.write(tsv_data.HEADER + "\n")
utterance_counter = 0
for (regrouped_utterance, speaker_index, start_time_sec,
end_time_sec) in regrouped_utterances:
utterance_counter += 1
line = "%.3f\t%.3f\t%s\t%s [U%d] [Speaker #%d]" % (
start_time_sec + begin_sec, end_time_sec + begin_sec,
tsv_data.SPEECH_TRANSCRIPT_TIER, regrouped_utterance,
utterance_counter, speaker_index)
print(line)
f.write(line + "\n")
def sample_long_running_recognize(storage_uri):
"""
Transcribe long audio file from Cloud Storage using asynchronous speech
recognition
Args:
storage_uri URI for audio file in Cloud Storage, e.g. gs://[BUCKET]/[FILE]
"""
os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = "C:/Users/RedLine/Desktop/Semester 8/FYP/FYP_final/FYP-key.json"
client = speech.SpeechClient()
# storage_uri = 'gs://cloud-samples-data/speech/brooklyn_bridge.raw'
# Sample rate in Hertz of the audio data sent
# sample_rate_hertz = 16000
# The language of the supplied audio
language_code = "en-US"
# Encoding of audio data sent. This sample sets this explicitly.
# This field is optional for FLAC and WAV audio formats.
encoding = enums.RecognitionConfig.AudioEncoding.LINEAR16
config = {
# "sample_rate_hertz": sample_rate_hertz,
"enable_speaker_diarization": True,
"enable_automatic_punctuation": True,
"language_code": language_code,
"encoding": encoding,
"audio_channel_count": 2,
}
audio = {"uri": storage_uri}
operation = client.long_running_recognize(config, audio)
print(u"Waiting for operation to complete...")
response = operation.result()
result = response.results[-1]
words_info = result.alternatives[0].words
tag = 1
tag_prev = 1
speaker = ""
transcript = ""
for word_info in words_info:
if tag_prev == tag:
tag_prev = tag
tag = word_info.speaker_tag
speaker = speaker + " " + word_info.word
elif not (speaker[-1] == "." or speaker[-1] == "?"):
speaker = speaker + " " + word_info.word
else:
transcript += "speaker {}: {}".format(tag_prev, speaker) + '\n'
tag_prev = tag
tag = word_info.speaker_tag
speaker = "" + word_info.word
transcript += "speaker {}: {}".format(tag_prev, speaker)
print("transcript : ", transcript)
f = open("transcript.txt", "a")
f.write(transcript)
f.close()
import io
import os
# Imports the Google Cloud client library
from google.cloud import speech_v1p1beta1
from google.cloud.speech_v1p1beta1 import enums
from google.cloud.speech_v1p1beta1 import types
# Instantiates a client
client = speech_v1p1beta1.SpeechClient()
print "Using testwav1m, 8k, with the below config:"
print ""
print "importing speech_v1p1beta1"
print "language_code='en-GB'"
print "use_enhanced=True"
print "enable_automatic_punctuation=True"
print "enable_word_time_offsets=True"
print ""
print "Transcript is as follows"
# The name of the audio file to transcribe
file_name = os.path.join(os.path.dirname(__file__), 'resources',
'test_8k_mocktheweek.wav')
# Loads the audio into memory
with io.open(file_name, 'rb') as audio_file:
content = audio_file.read()
audio = types.RecognitionAudio(content=content)
config = types.RecognitionConfig(
encoding=enums.RecognitionConfig.AudioEncoding.LINEAR16,
def sample_long_running_recognize(storage_uri):
"""
Transcribe a long audio file using asynchronous speech recognition
Args:
local_file_path Path to local audio file, e.g. /path/audio.wav
"""
# standard speech client
#client = speech_v1.SpeechClient()
# if utilizing speaker diarization
client = speech_v1p1beta1.SpeechClient()
# local_file_path = 'resources/brooklyn_bridge.raw'
# The language of the supplied audio
language_code = "en-US"
# Enhanced model to use
model = "phone_call"
# Sample rate in Hertz of the audio data sent
sample_rate_hertz = int(sys.argv[2])
# Optional. Specifies the estimated number of speakers in the conversation.
diarization_speaker_count = 2
# Encoding of audio data sent. This sample sets this explicitly.
# This field is optional for FLAC and WAV audio formats.
encoding = enums.RecognitionConfig.AudioEncoding.LINEAR16
config = {
"model": model,
"use_enhanced": True,
"language_code": language_code,
"sample_rate_hertz": sample_rate_hertz,
"encoding": encoding,
"enable_automatic_punctuation": True,
"enable_speaker_diarization": True,
"diarization_speaker_count": diarization_speaker_count,
}
audio = {"uri": storage_uri}
operation = client.long_running_recognize(config, audio)
print(u"Waiting for operation to complete...")
response = operation.result()
outtext = list()
out_text_speaker = list()
out_text_speaker_label = list()
for result in response.results:
# First alternative is the most probable result
alternative = result.alternatives[0]
# print(u"Transcript: {}".format(alternative.transcript))
outtext.append(alternative.transcript)
for word in alternative.words:
print(u"Speaker: {}, Word: {}".format(word.speaker_tag, word.word))
out_text_speaker.append(word.word)
out_text_speaker_label.append(word.speaker_tag)
return outtext, out_text_speaker, out_text_speaker_label
