这可能是一个与以下帖子非常相似的问题：
如何从Watson语音到文本输出重建对话？
我尝试应用@zhqiat给出的解决方案。但不幸的是，我得到的输出似乎不正确。
以下是我实施的内容。
# 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)
以下是我从Watson SpeechToText API得到的结果的一小部分。
{
"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，会得到这样的输出。你会看到对话在几个单词后被截断，第二行的文字记录栏中只有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, ...
我试了几种方法来解决这个问题，但都没有得到满意的结果。
据我所见，你只挑选了结果中的第一个元素-
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。
试试这个就行了。感谢其他代码
data = []
for alts in res['results']:
for i in alts['alternatives']:
data.extend(i['timestamps'])
convo = pd.DataFrame(data)