Currently voice recognition accurate rate is worse then Jan even using same *.wav file - Githubissues

Uberi / speech_recognition

Speech recognition module for Python, supporting several engines and APIs, online and offline.

https://pypi.python.org/pypi/SpeechRecognition/

BSD 3-Clause "New" or "Revised" License

8.44k stars 2.4k forks source link

Currently voice recognition accurate rate is worse then Jan even using same *.wav file #478

Open tliang2017 opened 4 years ago

tliang2017 commented 4 years ago

Steps to reproduce

I found voice recognition result has some difference with result from about 3 month ago. I use one same file for recognition. -- file see as in attachment

Expected behaviour

Server will return: "939 popsicle button on the headset if the code on your phone is for 73939 have the call button on the headset if the code" -- same as I got the result on Jan 16, 2020, the accurate rate is very good

Actual behaviour

Server return result is: "939 tap the call button on the headset if the code on your phone is 499 button on the headset if the" -- this is a lot of difference with the result that I got on Jan 16, 2020, and for this time recognition result missing some numbers. I have try a lot of files, all have same observation. Is it any change on back end server to make this issue happen?



System information
------------------
comments: 
Could you help to have a look? 
Thanks in advance

audio file see as in below link:

[Record_Wave-2020-01-16_09_22_36.zip](https://github.com/Uberi/speech_recognition/files/4468754/Record_Wave-2020-01-16_09_22_36.zip)

kot0 commented 4 years ago

I noticed that after 100-200 requests, the recognition quality decreases.