How to Access the Confidence on Pocket Sphinx Transcription

Question

I am converting audio to text using sphinx, and I can't find how to access the confidence score for each word

I am able to access the transcription output, but I can't get the estimated probabilities behind the model. This feels basic, but I can't find the proper documentation. What should I add to the below?

test = sr.AudioFile(audio_file)
Recon = sr.Recognizer()

with test as source:
    test_audio = Recon.record(source)
text = Recon.recognize_sphinx(test_audio,language = 'en-US')```

score 1 · Answer 1 · answered Aug 07 '19 at 07:11

1

Confidence result is not returned by the current version of speech-recognition. If you look at the implementation:

def recognize_sphinx(...):
   ...
   # return results
   hypothesis = decoder.hyp()
   if hypothesis is not None: return hypothesis.hypstr
   raise UnknownValueError()  # no transcriptions available

you will see that only the text result (hypothesis.hypstr) is returned, while the confidence is in hypothesis.prob. A quick workaround would be to copy-paste the entire method after installing pocketsphinx alone:

pip install pocketsphinx

answered Aug 07 '19 at 07:11

Alexander Solovets

2,447
15
22

Can you please elaborate on what do you mean by copy-paste the entire method? – SuperKogito Sep 04 '19 at 15:22
1

I meant to follow the link, copy the source code, and paste it into your project. – Alexander Solovets Sep 05 '19 at 11:31
what about overloading the recognize_sphinx method in your project: you define your own flavour of the method then you do a r.recognize_sphinx = my_own_private_method() – Alexandre Mazel Oct 19 '22 at 15:33
@AlexandreMazel That might work as well. – Alexander Solovets Oct 21 '22 at 01:39

How to Access the Confidence on Pocket Sphinx Transcription

1 Answers1