使用IBM Watson语音到文本API的语音识别

Question

这个问题可能与下面的帖子非常相似：

How to reconstruct a conversation from Watson Speech-to-Text output?

我尝试应用@zhqiat给出的解决方案。但不幸的是，我得到的结果似乎不正确

下面是我已经实现的

# credentials from IBM Cloud Service for Speech to text

WATSON_API_KEY = '{API}'
WATSON_STT_URL = '{URL}'

## Implementing the Speech To Text module
import os
import json
import pandas as pd

from ibm_watson import SpeechToTextV1
from ibm_cloud_sdk_core.authenticators import IAMAuthenticator

def watson_batch_stt(filename: str, lang: str, encoding: str) -> str:
    authenticator = IAMAuthenticator(WATSON_API_KEY)
    speech_to_text = SpeechToTextV1(authenticator=authenticator)
    speech_to_text.set_service_url(WATSON_STT_URL)

    with open(filename, 'rb') as audio_file:
        response = speech_to_text.recognize(
            audio=audio_file,
            content_type='audio/{}'.format(os.path.splitext(filename)[1][1:]),
            model=lang + '_NarrowbandModel',
            max_alternatives=0,
            speaker_labels=True,
            inactivity_timeout=-1,
        ).get_result()

    return response


SpeechToTextResults = watson_batch_stt('FILENAME.wav','en-US','UTF-8')
print(json.dumps(SpeechToTextResults))

jsonconvo = json.dumps(SpeechToTextResults)
jsonconvo = json.loads(jsonconvo)
speakers=pd.DataFrame(jsonconvo['speaker_labels']).loc[:,['from','speaker','to']]
convo=pd.DataFrame(jsonconvo['results'][0]['alternatives'][0]['timestamps'])
speakers=speakers.join(convo)

ChangeSpeaker=speakers.loc[speakers['speaker'].shift()!=speakers['speaker']].index


Transcript=pd.DataFrame(columns=['from','to','speaker','transcript'])
for counter in range(0,len(ChangeSpeaker)):
    print(counter)
    currentindex=ChangeSpeaker[counter]
    try:
        nextIndex=ChangeSpeaker[counter+1]-1
        temp=speakers.loc[currentindex:nextIndex,:]
    except:
        temp=speakers.loc[currentindex:,:]



    Transcript=Transcript.append(pd.DataFrame([[temp.head(1)['from'].values[0],temp.tail(1)['to'].values[0],temp.head(1)['speaker'].values[0],temp[0].tolist()]],columns=['from','to','speaker','transcript']))

print(Transcript)

下面是我从Watson SpeechToText API获得的结果的一小部分

{
"result_index": 0,
"results": [
    {
        "final": true,
        "alternatives": [
            {
                "transcript": "thanks for calling Company Name %HESITATION this is Ross ",
                "confidence": 0.73,
                "timestamps": [
                    [
                        "thanks",
                        0.71,
                        1.0
                    ],
                    [
                        "for",
                        1.06,
                        1.23
                    ],
                    [
                        "calling",
                        1.23,
                        1.68
                    ],
                    [
                        "Company",
                        1.68,
                        1.95
                    ],
                    [
                        "Name",
                        1.98,
                        2.35
                    ],
                    [
                        "%HESITATION",
                        2.35,
                        2.61
                    ],
                    [
                        "this",
                        2.61,
                        2.82
                    ],
                    [
                        "is",
                        2.82,
                        2.94
                    ],
                    [
                        "Ross",
                        2.94,
                        3.33
                    ]
                ]
            }
        ]
    },
    {
        "final": true,
        "alternatives": [
            {
                "transcript": "yes Ross %HESITATION I have a new puppy eight eight weeks old ",
                "confidence": 0.88,
                "timestamps": [
                    [
                        "yes",
                        5.42,
                        5.78
                    ],
                    [
                        "Ross",
                        5.78,
                        6.13
                    ],
                    [
                        "%HESITATION",
                        6.23,
                        6.66
                    ],
                    [
                        "I",
                        6.66,
                        6.8
                    ],
                    [
                        "have",
                        6.8,
                        6.99
                    ],
                    [
                        "a",
                        6.99,
                        7.05
                    ],
                    [
                        "new",
                        7.05,
                        7.21
                    ],
                    [
                        "puppy",
                        7.21,
                        7.84
                    ],
                    [
                        "eight",
                        8.31,
                        8.63
                    ],
                    [
                        "eight",
                        8.77,
                        8.98
                    ],
                    [
                        "weeks",
                        8.98,
                        9.27
                    ],
                    [
                        "old",
                        9.27,
                        9.63
                    ]
                ]
            }
        ]
    },
    {
        "final": true,
        "alternatives": [
            {
                "transcript": "%HESITATION it's %HESITATION Australian and miniature Australian shepherd doodle necks ",
                "confidence": 0.77,
                "timestamps": [
                    [
                        "%HESITATION",
                        10.62,
                        11.29
                    ],
                    [
                        "it's",
                        11.82,
                        12.04
                    ],
                    [
                        "%HESITATION",
                        12.04,
                        12.37
                    ],
                    [
                        "Australian",
                        12.45,
                        13.22
                    ],
                    [
                        "and",
                        13.25,
                        13.43
                    ],
                    [
                        "miniature",
                        13.43,
                        13.89
                    ],
                    [
                        "Australian",
                        13.89,
                        14.49
                    ],
                    [
                        "shepherd",
                        14.49,
                        15.13
                    ],
                    [
                        "doodle",
                        15.65,
                        16.01
                    ],
                    [
                        "necks",
                        16.01,
                        16.62
                    ]
                ]
            }
        ]
    },
        

当我尝试应用解决方案时，得到的输出是这样的。您将看到对话在几个单词后被截断，第二行的transcript列中只有nan

   from     to speaker                                         transcript
0  0.71   6.13       0  [thanks, for, calling, Company Name, %HESITATION,...
0  6.23  28.85       1  [nan, nan, nan, nan, nan, nan, nan, nan, nan, ...

我尝试了几种方法来解决这个问题，但没有得到任何令人满意的结果

Answer 1

就我所知，你只是从结果中选取了第一个元素-

convo=pd.DataFrame(jsonconvo['results'][0]['alternatives'][0]['timestamps'])

你需要所有的结果

您得到的nan是由索引不存在的字段引起的not a number错误