What is the best Python library for continuous speech recognition?

Question

Python

So, the crux of the matter. I work in Python, looking for a way to recognize long-term sequential speech.
The speech itself is audio stories, the goal is transcription back into text.
So far, I have settled on pocketsphinx (Russian dictionary), but it also works frankly poorly.
First, it's too slow.
Secondly, it is good for small phrases with a clear division. When the speech is continuous, things are bad.
What libraries would you recommend to try taking into account my task?

Answer 1

try https://github.com/alphacep/vosk-api
was recently in the news
vosk for local continuous speech recognition that supports Russian

Answer 2

Such things even at Apple and Google suck, I don’t think there is something better

Answer 3

Mozilla recently released a new version of its library with many improvements. It also supports different languages, including Russian. https://hacks.mozilla.org/2019/12/deepspeech-0-6-m...
https://github.com/mozilla/DeepSpeech

Answer 4

I heard about attempts to use this solution: https://cloud.google.com/speech-to-text/
For Russian, you should definitely try the solution from Yandex, it is currently the best in this segment: https://cloud. yandex.ru/services/speechkit
In practice, it works so-so and in production, speech2text models are usually configured to detect specific keywords, and not full speech recognition.