For speech-to-text extraction you can try Silero [1].

Free software (AGPL-3.0 License), fast, highly accurate and extremely simple to deploy (I have no affiliation with them).

[1] https://github.com/snakers4/silero-models