def speech_to_text_in_a_min(
doc_title='範例1_一分鐘內雲端運算',
title_pattern='nlpno',
wd='/home/slave1/git/Speech2Text_workshop/record',
json_os='/home/slave1/git/Speech2Text_workshop/speech2text-3de4444fd46a.json',
sample_rate_hertz=48000):
'''
* json_os：憑證檔的路徑
* title_pattern：錄音檔的名稱模式
* sample_rate_hertz：錄音的取樣頻率
* doc_title：docx文件名稱
* wd：工作目錄
'''
# 計時
start_time = time.time()
# 從python client端對雲端speech2text服務進行驗證
os.environ['GOOGLE_APPLICATION_CREDENTIALS'] = json_os
client = speech.SpeechClient()
file_list = os.listdir(wd)
# 選出title_pattern的錄音檔
select_wav = []
for i in file_list:
if title_pattern in i:
select_wav.append(wd + '\\' + i)
aa = pd.DataFrame()
for music in select_wav:
# 將 audio錄音檔 讀入進來
with io.open(music, 'rb') as audio_file:
content = audio_file.read()
# 將錄音檔轉換成google 看得懂的格式
audio = types.RecognitionAudio(content=content)
# 設定格式錄音檔
config = types.RecognitionConfig(
encoding=enums.RecognitionConfig.AudioEncoding.LINEAR16,
sample_rate_hertz=sample_rate_hertz,
language_code='cmn-Hant-TW',
enable_word_time_offsets=True)
# 機器學習文字辨識(speech2text)
print('')
response = client.recognize(config, audio)
transcript_list = []
transcript_confidence = []
timerecored = []
# 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:
alternative = result.alternatives[0]
# The first alternative is the most likely one for this portion.
transcript_list.append(alternative.transcript)
transcript_confidence.append(alternative.confidence)
print('Transcript: {}'.format(alternative.transcript))
print('Confidence: {}'.format(alternative.confidence))
# begining and end time of a sentence
sentence_start_time = alternative.words[0].start_time
sentence_end_time = alternative.words[len(alternative.words) -
1].end_time
# make time
sentence_start_time = round(sentence_start_time.seconds +
sentence_start_time.nanos * 1e-9)
sentence_end_time = round(sentence_end_time.seconds +
sentence_end_time.nanos * 1e-9)
# make min
sentence_start_time = str(
datetime.timedelta(seconds=sentence_start_time))
sentence_end_time = str(
datetime.timedelta(seconds=sentence_end_time))
timerecored.append([sentence_start_time, sentence_end_time])
# pandas 建立信心程度資料表
# make df
transcript_df = pd.DataFrame(transcript_list, columns=['文章段句'])
confidence_df = pd.DataFrame(transcript_confidence,
columns=['機器認字信心水準'])
confidence_df['機器認字信心水準'] = round(confidence_df['機器認字信心水準'], 2)
time_df = pd.DataFrame(timerecored, columns=['start', 'end'])
correctness_summary_df = pd.concat(
[transcript_df, confidence_df, time_df], axis=1)
correctness_summary_df = correctness_summary_df.sort_values(
['機器認字信心水準'])
correctness_summary_df['改善順序'] = range(1,
len(correctness_summary_df) + 1)
timer_translist = []
for hah, timer in zip(transcript_list, timerecored):
timer_translist.append(hah + ' ' + '【' + ' to '.join(timer) + '】')
aa = pd.concat([aa, correctness_summary_df])
# 製作文字雲
from speech2text import make_worldcould_report, text_freq
cut_text = make_worldcould_report(data=aa,
pd_text_col='文章段句',
mask_pic=False,
filename='wordcloud',
pic_name='test.png')
words_counts = text_freq(cut_text)
# 計算重要程度
max = words_counts['counts'].describe()['max']
mean = words_counts['counts'].describe()['mean']
# 僅取出max與mean的字詞
words_counts = words_counts[(words_counts['counts'] <= max)
& (words_counts['counts'] >= mean)]
df_count_all = pd.DataFrame()
for index, i in words_counts.iterrows():
df_count = correctness_summary_df[
correctness_summary_df['文章段句'].str.contains(i['word'])]
if not df_count.empty:
df_count['重要性'] = i['counts']
df_count_all = pd.concat([df_count_all, df_count])
# group by
correctness_summary_df = df_count_all.groupby(
['文章段句', '機器認字信心水準', 'start', 'end', '改善順序'],
as_index=False)['重要性'].mean().round(2)
# save to docx
document = Document()
document.add_heading(doc_title, 0)
document.add_paragraph(
'機器認字信心水準' + str(round(correctness_summary_df['機器認字信心水準'].mean(), 2)) +
'\n\n' + '\n\n'.join(timer_translist))
document.add_picture('wordcloud.png', width=Cm(15), height=Cm(13))
document.save(doc_title + '_文章逐字稿.docx')
print('Done')
print('請看os檔案中有沒有兩個檔案，一格個是完整的docx檔案，一個是csv檔案')
print("--- %s seconds ---" % (round(time.time() - start_time, 2)))
return correctness_summary_df.to_excel(doc_title + '_文章認字信心矩陣.xlsx')
def speech_to_text(
gcs_uri='gs://speechfashion/Acc.wav',
doc_title='範例2_一分鐘以上雲端運算',
timeout=None,
path='',
json_os='/home/slave1/git/Speech2Text/damnhow-db8d83229dd4.json',
sample_rate_hertz=96000):
'''
1.產出文章認字信心矩陣csv，提供修改者文句之修正順序
2.產出docx文本，並提供文章機器認字信心水準，供修改者修改
'''
# 計時
start_time = time.time()
os.environ['GOOGLE_APPLICATION_CREDENTIALS'] = json_os
client = speech.SpeechClient()
audio = types.RecognitionAudio(uri=gcs_uri)
config = types.RecognitionConfig(
encoding=enums.RecognitionConfig.AudioEncoding.LINEAR16,
sample_rate_hertz=sample_rate_hertz,
language_code='cmn-Hant-TW',
enable_word_time_offsets=True)
# config=types.StreamingRecognitionConfig(config=config)
# stream = [audio]
# requests = (types.StreamingRecognizeRequest(audio_content=chunk)
# for chunk in stream)
# responses = client.streaming_recognize(config, requests)
operation = client.long_running_recognize(config, audio)
print('機器學習文字辨識中...')
response = operation.result(timeout=timeout) #
aa = pd.DataFrame()
transcript_list = []
transcript_confidence = []
timerecored = []
# Each result is for a consecutive portion of the audio. Iterate through
# them to get the transcripts for the entire audio file.
try:
for result in response.results:
alternative = result.alternatives[0]
# The first alternative is the most likely one for this portion.
transcript_list.append(alternative.transcript)
transcript_confidence.append(alternative.confidence)
print('Transcript: {}'.format(alternative.transcript))
print('Confidence: {}'.format(alternative.confidence))
# begining and end time of a sentence
sentence_start_time = alternative.words[0].start_time
sentence_end_time = alternative.words[len(alternative.words) -
1].end_time
# make time
sentence_start_time = round(sentence_start_time.seconds +
sentence_start_time.nanos * 1e-9)
sentence_end_time = round(sentence_end_time.seconds +
sentence_end_time.nanos * 1e-9)
# make min
sentence_start_time = str(
datetime.timedelta(seconds=sentence_start_time))
sentence_end_time = str(
datetime.timedelta(seconds=sentence_end_time))
timerecored.append([sentence_start_time, sentence_end_time])
# pandas 建立信心程度資料表
# make df
transcript_df = pd.DataFrame(transcript_list, columns=['文章段句'])
confidence_df = pd.DataFrame(transcript_confidence,
columns=['機器認字信心水準'])
confidence_df['機器認字信心水準'] = round(confidence_df['機器認字信心水準'], 2)
time_df = pd.DataFrame(timerecored, columns=['start', 'end'])
correctness_summary_df = pd.concat(
[transcript_df, confidence_df, time_df], axis=1)
correctness_summary_df = correctness_summary_df.sort_values(
['機器認字信心水準'])
correctness_summary_df['改善順序'] = range(
1,
len(correctness_summary_df) + 1)
timer_translist = []
for hah, timer in zip(transcript_list, timerecored):
timer_translist.append(hah + ' ' + '【' + ' to '.join(timer) +
'】')
aa = pd.concat([aa, correctness_summary_df])
# 製作文字雲
from speech2text import make_worldcould_report, text_freq
cut_text = make_worldcould_report(data=aa,
pd_text_col='文章段句',
mask_pic=False,
filename='wordcloud',
pic_name='test.png')
words_counts = text_freq(cut_text)
# 計算重要程度
max = words_counts['counts'].describe()['max']
mean = words_counts['counts'].describe()['mean']
# 僅取出max與mean的字詞
words_counts = words_counts[(words_counts['counts'] <= max)
& (words_counts['counts'] >= mean)]
df_count_all = pd.DataFrame()
for index, i in words_counts.iterrows():
df_count = correctness_summary_df[
correctness_summary_df['文章段句'].str.contains(i['word'])]
if not df_count.empty:
df_count['重要性'] = i['counts']
df_count_all = pd.concat([df_count_all, df_count])
# group by
correctness_summary_df = df_count_all.groupby(
['文章段句', '機器認字信心水準', 'start', 'end', '改善順序'],
as_index=False)['重要性'].mean().round(2)
# save to docx
document = Document()
document.add_heading(doc_title, 0)
document.add_paragraph(
'機器認字信心水準' +
str(round(correctness_summary_df['機器認字信心水準'].mean(), 2)) + '\n\n' +
'\n\n'.join(timer_translist))
document.add_picture('wordcloud.png', width=Cm(15), height=Cm(13))
document.save(path + doc_title + '_文章逐字稿.docx')
print('Done')
print('請看os檔案中有沒有兩個檔案，一格個是完整的docx檔案，一個是csv檔案')
print("--- %s seconds ---" % (round(time.time() - start_time, 2)))
return correctness_summary_df.to_excel(path + doc_title +
'_文章認字信心矩陣.xlsx')
except:
print('錄音檔無法辨識，請檢查該錄音檔是否存在或該錄音檔音質問題')
# 請執行到這邊停住
time_df = pd.DataFrame(timerecored, columns=['start', 'end'])
correctness_summary_df = pd.concat([transcript_df, confidence_df, time_df],
axis=1)
correctness_summary_df = correctness_summary_df.sort_values(['機器認字信心水準'])
correctness_summary_df['改善順序'] = range(1, len(correctness_summary_df) + 1)
timer_translist = []
for hah, timer in zip(transcript_list, timerecored):
timer_translist.append(hah + ' ' + '【' + ' to '.join(timer) + '】')
aa = pd.concat([aa, correctness_summary_df])
# 製作文字雲
from speech2text import make_worldcould_report, text_freq
cut_text = make_worldcould_report(data=aa,
pd_text_col='文章段句',
mask_pic=False,
filename='wordcloud',
pic_name='test.png')
words_counts = text_freq(cut_text)
# 計算重要程度
max = words_counts['counts'].describe()['max']
mean = words_counts['counts'].describe()['mean']
# 僅取出max與mean的字詞
words_counts = words_counts[(words_counts['counts'] <= max)
& (words_counts['counts'] >= mean)]
df_count_all = pd.DataFrame()
for index, i in words_counts.iterrows():