使用Python和Microsoft Azure语音转文本实现字幕/字幕化

Question

我一直在尝试使用Python中的Microsoft Azure语音识别服务制作字幕，但是无法弄清楚。我已经按照其他人在此处提供的提示获取了单个单词，但即使将其格式化为.srt或.vtt，也似乎很复杂。

以下是代码：

import azure.cognitiveservices.speech as speechsdk


def speech_recognize_continuous_from_file():
    """performs continuous speech recognition with input from an audio file"""
    # <SpeechContinuousRecognitionWithFile>
    speech_key, service_region = "{api-key}", "{serive-region}"
    speech_config = speechsdk.SpeechConfig(subscription=speech_key, region=service_region)
    
    audio_filename = "{for example: video.wav}"
    audio_config = speechsdk.audio.AudioConfig(filename=audio_filename)
    
    speech_config.speech_recognition_language="en-US"
    speech_config.request_word_level_timestamps()

    speech_config.enable_dictation()
    speech_config.output_format = speechsdk.OutputFormat(1)
    
    speech_recognizer = speechsdk.SpeechRecognizer(speech_config=speech_config, audio_config=audio_config)

    done = False
    
    results = []
    
    transcript = []
    words = []
    
    def handle_final_result(evt):
        import json
        results = json.loads(evt.result.json)
        transcript.append(results['DisplayText'])
    confidence_list_temp = [item.get('Confidence') for item in results['NBest']]
    max_confidence_index = confidence_list_temp.index(max(confidence_list_temp))
        words.extend(results['NBest'][max_confidence_index]['Words'])
    def stop_cb(evt):
        """callback that stops continuous recognition upon receiving an event `evt`"""
        print('CLOSING on {}'.format(evt))
        speech_recognizer.stop_continuous_recognition()
        nonlocal done
        done = True
        print("Transcript display list:\n")
        print(transcript)
        print("\nWords\n")
        print(words)
        print("\n")


    speech_recognizer.recognized.connect(handle_final_result)
    # Connect callbacks to the events fired by the speech recognizer
    speech_recognizer.recognizing.connect(lambda evt: format(evt))
    speech_recognizer.recognized.connect(lambda evt: format(evt))
    speech_recognizer.session_started.connect(lambda evt: format(evt))
    speech_recognizer.session_stopped.connect(lambda evt: format(evt))
    speech_recognizer.canceled.connect(lambda evt: format(evt))
    # stop continuous recognition on either session stopped or canceled events
    speech_recognizer.session_stopped.connect(stop_cb)
    speech_recognizer.canceled.connect(stop_cb)

    # Start continuous speech recognition
    speech_recognizer.start_continuous_recognition()
    while not done:
        time.sleep(.5)

    
    with open('Azure_Raw.txt','w') as f:
        f.write('\n'.join(results))

sample_long_running_recognize(storage_uri)

我在字幕方面唯一找到的其他“教程”是一个Google Cloud的，它给出了我想要的结果（是的，我已经亲测过），但Azure显然完全不像G-cloud：https://medium.com/searce/generate-srt-file-subtitles-using-google-clouds-speech-to-text-api-402b2f1da3bd

所以基本上：我如何将大约3秒的语音文本转换为.srt格式，就像这样：

1
00:00:00,000 --> 00:00:03,000
This is the first sentence that

2
00:00:03,000 --> 00:00:06,000
continues after 3 seconds or so

- Python Dong

Answer 1

所以如果你仔细观察，Azure语音服务的JSON输出与其他服务的输出略有不同。

对于提到的配置，在选出最佳匹配后，输出结果如下：

[{'Duration': 3900000, 'Offset': 500000, 'Word': "what's"},
 {'Duration': 1300000, 'Offset': 4500000, 'Word': 'the'},
 {'Duration': 2900000, 'Offset': 5900000, 'Word': 'weather'},
 {'Duration': 4800000, 'Offset': 8900000, 'Word': 'like'}]

有三个输出 - Word, Duration和offset

• Duration - 单词的拼写时间，单位为100纳秒
• Offset - 从视频开始计算，以100纳秒为单位的秒数

您将需要利用此信息来构建时间轴

import azure.cognitiveservices.speech as speechsdk
import os
import time
import pprint
import json
import srt
import datetime

 
path = os.getcwd()
# Creates an instance of a speech config with specified subscription key and service region.
# Replace with your own subscription key and region identifier from here: https://aka.ms/speech/sdkregion
speech_key, service_region = "<>", "<>"
speech_config = speechsdk.SpeechConfig(subscription=speech_key, region=service_region)

# Creates an audio configuration that points to an audio file.
# Replace with your own audio filename.
audio_filename = "sample.wav"
audio_input = speechsdk.audio.AudioConfig(filename=audio_filename)

# Creates a recognizer with the given settings
speech_config.speech_recognition_language="en-US"
speech_config.request_word_level_timestamps()



speech_config.enable_dictation()
speech_config.output_format = speechsdk.OutputFormat(1)

speech_recognizer = speechsdk.SpeechRecognizer(speech_config=speech_config, audio_config=audio_input)

#result = speech_recognizer.recognize_once()
all_results = []
results = []
transcript = []
words = []


#https://learn.microsoft.com/en-us/python/api/azure-cognitiveservices-speech/azure.cognitiveservices.speech.recognitionresult?view=azure-python
def handle_final_result(evt):
    import json
    all_results.append(evt.result.text) 
    results = json.loads(evt.result.json)
    transcript.append(results['DisplayText'])
    confidence_list_temp = [item.get('Confidence') for item in results['NBest']]
    max_confidence_index = confidence_list_temp.index(max(confidence_list_temp))
    words.extend(results['NBest'][max_confidence_index]['Words'])



done = False

def stop_cb(evt):
    print('CLOSING on {}'.format(evt))
    speech_recognizer.stop_continuous_recognition()
    global done
    done= True
    
speech_recognizer.recognized.connect(handle_final_result) 
#Connect callbacks to the events fired by the speech recognizer    
speech_recognizer.recognizing.connect(lambda evt: print('RECOGNIZING: {}'.format(evt)))
speech_recognizer.recognized.connect(lambda evt: print('RECOGNIZED: {}'.format(evt)))
speech_recognizer.session_started.connect(lambda evt: print('SESSION STARTED: {}'.format(evt)))
speech_recognizer.session_stopped.connect(lambda evt: print('SESSION STOPPED {}'.format(evt)))
speech_recognizer.canceled.connect(lambda evt: print('CANCELED {}'.format(evt)))
# stop continuous recognition on either session stopped or canceled events
speech_recognizer.session_stopped.connect(stop_cb)
speech_recognizer.canceled.connect(stop_cb)

speech_recognizer.start_continuous_recognition()

while not done:
    time.sleep(.5)
    
print("Printing all results:")
print(all_results)

speech_to_text_response = words

def convertduration(t):
    x= t/10000
    return int((x / 1000)), (x % 1000)


##-- Code to Create Subtitle --#

#3 Seconds
bin = 3.0
duration = 0 
transcriptions = []
transcript = ""
index,prev=0,0
wordstartsec,wordstartmicrosec=0,0
for i in range(len(speech_to_text_response)):
    #Forms the sentence until the bin size condition is met
    transcript = transcript + " " + speech_to_text_response[i]["Word"]
    #Checks whether the elapsed duration is less than the bin size
    if(int((duration / 10000000)) < bin): 
        wordstartsec,wordstartmicrosec=convertduration(speech_to_text_response[i]["Offset"])
        duration= duration+speech_to_text_response[i]["Offset"]-prev
        prev=speech_to_text_response[i]["Offset"]
                #transcript = transcript + " " + speech_to_text_response[i]["Word"]
    else : 
        index=index+1
        #transcript = transcript + " " + speech_to_text_response[i]["Word"]
        transcriptions.append(srt.Subtitle(index, datetime.timedelta(0, wordstartsec, wordstartmicrosec), datetime.timedelta(0, wordstartsec+bin, 0), transcript))
        duration = 0 
        #print(transcript)
        transcript=""



transcriptions.append(srt.Subtitle(index, datetime.timedelta(0, wordstartsec, wordstartmicrosec), datetime.timedelta(0, wordstartsec+bin, 0), transcript))
subtitles = srt.compose(transcriptions)
with open("subtitle.srt", "w") as f:
    f.write(subtitles)

---

以下是您参考的输出结果：

输出

---

希望这有所帮助 :)

- Satya V

---

现在这是一个非常好的答案！我终于能够看到这个和Google Cloud版本之间的相似之处。它也让我意识到我必须磨练我的JSON知识（我不知道为什么，但我对它有困难，而它是最容易的东西之一）。非常感谢Sathya！ - Python Dong