python - 如何使用 Azure Speech to Text 和 Python SDK 获取字级时间戳？

Question

在我在 GitHub 上找到的示例的帮助下，我的代码当前能够读取音频文件并使用 Azure Speech to Text 进行转录。但是，我需要在转录中包含所有单词的时间戳。根据文档，该功能是在 1.5.0 版本中添加的，并通过 request_word_level_timestamps() 方法访问。但即使我打电话了，我也得到了和以前一样的回应。我无法从文档中弄清楚如何使用它。有谁知道它是如何工作的吗？

我使用的是 Python SDK 版本 1.5.1。

import azure.cognitiveservices.speech as speechsdk
import time
from allennlp.predictors.predictor import Predictor
import json 

inputPath = "(inputlocation)"
outputPath = "(outputlocation)"

# Creates an instance of a speech config with specified subscription     key and service region.
# Replace with your own subscription key and service region (e.g., "westus").
speech_key, service_region = "apikey", "region"
speech_config = speechsdk.SpeechConfig(subscription=speech_key,     region=service_region)
speech_config.request_word_level_timestamps()
speech_config.output_format=speechsdk.OutputFormat.Detailed
#print("VALUE: " +     speech_config.get_property(property_id=speechsdk.PropertyId.SpeechServic    eResponse_RequestWordLevelTimestamps))
filename = input("Enter filename: ")

print(speech_config)

try:
    audio_config = speechsdk.audio.AudioConfig(filename= inputPath +     filename)

    # Creates a recognizer with the given settings
    speech_recognizer =     speechsdk.SpeechRecognizer(speech_config=speech_config,     audio_config=audio_config)


def start():
    done = False
    #output = ""
    fileOpened = open(outputPath+ filename[0: len(filename) - 4] + "_MS_recognized.txt", "w+")
    fileOpened.truncate(0)
    fileOpened.close()

    def stop_callback(evt):
        print("Closing on {}".format(evt))
        speech_recognizer.stop_continuous_recognition()
        nonlocal done
        done = True

    def add_to_res(evt):
        #nonlocal output
        #print("Recognized: {}".format(evt.result.text))
        #output = output + evt.result.text + "\n"
        fileOpened = open( outputPath + filename[0: len(filename) - 4] + "_MS_recognized.txt", "a")
        fileOpened.write(evt.result.text + "\n")
        fileOpened.close()
        #print(output)

    # 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.recognized.connect(add_to_res)
    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_callback)
    speech_recognizer.canceled.connect(stop_callback)

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


    # Starts speech recognition, and returns after a single utterance is recognized. The end of a
    # single utterance is determined by listening for silence at the end or until a maximum of 15
    # seconds of audio is processed.  The task returns the recognition text as result. 
    # Note: Since recognize_once() returns only a single utterance, it is suitable only for single
    # shot recognition like command or query. 
    # For long-running multi-utterance recognition, use start_continuous_recognition() instead.

start()

except Exception as e: 
    print("File does not exist")
    #print(e)

结果仅包含 session_id 和结果对象，结果对象包括 result_id、文本和原因。

Answer 1

我引用了你的代码并按照官方教程Quickstart: Recognize speech with the Speech SDK for Python进行操作要编写下面的示例代码，它可以打印每个单词的 Offset 和 Duration 值。我使用了一个名为 whatstheweatherlike.wav 的音频文件来自samples/csharp/sharedcontent/console/whatstheweatherlike.wav GitHub 存储库 Azure-Samples/cognitive-services-speech-sdk。

这是我的示例代码及其结果。

import azure.cognitiveservices.speech as speechsdk

speech_key, service_region = "<your api key>", "<your region>"
speech_config = speechsdk.SpeechConfig(subscription=speech_key, region=service_region)
speech_config.request_word_level_timestamps()

audio_config = speechsdk.audio.AudioConfig(filename='whatstheweatherlike.wav')
speech_recognizer = speechsdk.SpeechRecognizer(speech_config=speech_config, audio_config=audio_config)
result = speech_recognizer.recognize_once()

# print(result.json)
# If without `request_word_level_timestamps`, the result:
# {"DisplayText":"What's the weather like?","Duration":13400000,"Offset":400000,"RecognitionStatus":"Success"}
# Enable `request_word_level_timestamps`, the result includes word level timestamps.
# {"Duration":13400000,"NBest":[{"Confidence":0.9761951565742493,"Display":"What's the weather like?","ITN":"What's the weather like","Lexical":"what's the weather like","MaskedITN":"What's the weather like","Words":[{"Duration":3800000,"Offset":600000,"Word":"what's"},{"Duration":1200000,"Offset":4500000,"Word":"the"},{"Duration":2900000,"Offset":5800000,"Word":"weather"},{"Duration":4700000,"Offset":8800000,"Word":"like"}]},{"Confidence":0.9245584011077881,"Display":"what is the weather like","ITN":"what is the weather like","Lexical":"what is the weather like","MaskedITN":"what is the weather like","Words":[{"Duration":2900000,"Offset":600000,"Word":"what"},{"Duration":700000,"Offset":3600000,"Word":"is"},{"Duration":1300000,"Offset":4400000,"Word":"the"},{"Duration":2900000,"Offset":5800000,"Word":"weather"},{"Duration":4700000,"Offset":8800000,"Word":"like"}]}],"Offset":400000,"RecognitionStatus":"Success"}

import json
stt = json.loads(result.json)
confidences_in_nbest = [item['Confidence'] for item in stt['NBest']]
best_index = confidences_in_nbest.index(max(confidences_in_nbest))
words = stt['NBest'][best_index]['Words']
print(words)

print(f"Word\tOffset\tDuration")
for word in words:
    print(f"{word['Word']}\t{word['Offset']}\t{word['Duration']}")

上面脚本的输出是:

[{'Duration': 3800000, 'Offset': 600000, 'Word': "what's"}, {'Duration': 1200000, 'Offset': 4500000, 'Word': 'the'}, {'Duration': 2900000, 'Offset': 5800000, 'Word': 'weather'}, {'Duration': 4700000, 'Offset': 8800000, 'Word': 'like'}]
Word    Offset  Duration
what's  600000  3800000
the     4500000 1200000
weather 5800000 2900000
like    8800000 4700000

希望有帮助。