Integrate full support for SpeechRecognition

Problem

1. audio time greater than 1 minute
   google speech recognition can accept file that is less or equal than 1 minutes and only 50 requests are allowed per day.
   google speech recognition(not google cloud stt) is almost deprecated at all
2. Exception is handled poorly
   If audio file exceeds 1 minute then it throws error, but user can't know why this error occured.
3. No Test for audio transcription

Solution

• Integrate full support for SpeechRecognition
  • Since this project pursues to be lightweight, avoid custom logics and mapping argument to speech_recognition would be great.
  • I will add **kwargs that contains engine, model, api_key, regions, ... which are required from SpeechRecognition library
    and match it to SpeechRecongition Library using

  recognize_method = getattr(recognizer, f"recognize_{engine}")
  

  • skip for options with offline dependency to be lightweight and align with image_converter
• Exception handling
• Add test codes
  • skip for github_actions(CI)

Related Issue

• #326 tried to add whisper directly, but as #1284 noted, mapping SpeechRecognition's support would be simpler and support more models as well.
• #1275 wants custom support for audio transcription as well.
• #74 If the time of the audio exceeds quota(60 seconds per request, 50 requests per day),
  recognize_google() from SpeechRecognition throws an error, but the user can't know where it came from.

Suggestion

Asynchronous transcription and chunking would be great feature,
but I am concerned that it might not align with the project's philosophy. "Lightweight"

Start by reading the SpeechRecognition README and the related issues #326, #1284, #1275, and #74 to understand the proposed engine mapping, quotas, and exception behavior. Identify the project's audio transcription entry point, then define the supported non-offline engines and add transcription and error-handling tests; done means the supported SpeechRecognition options and failures are covered.