Hey there! The recognize_google() function you're using is tied to Google's cloud-based speech recognition service—so that's why it only works when you're online. To get offline functionality, we can use CMU PocketSphinx, an open-source offline speech engine that integrates seamlessly with the speech_recognition library you're already using.

Step 1: Install Required Dependencies #

First, you'll need to install the PocketSphinx package (it's not included with speech_recognition by default):

pip install pocketsphinx

Note: On Linux systems, you might need to install system-level dependencies first, like libpulse-dev (for Ubuntu/Debian: sudo apt-get install libpulse-dev).

Step 2: Modified Offline Code #

Here's your updated code that uses offline recognition instead of Google's online API. I've added error handling too, which helps catch cases where speech can't be recognized:

import speech_recognition as sr

r = sr.Recognizer()
with sr.Microphone() as source:
    # Adjust for ambient noise first (boosts recognition accuracy)
    r.adjust_for_ambient_noise(source, duration=0.5)
    print('Say Something!')
    audio = r.listen(source)

print('Done!')

try:
    # Use PocketSphinx for offline speech recognition
    text = r.recognize_sphinx(audio)
    print(f"Recognized: {text}")
except sr.UnknownValueError:
    print("Sorry, I couldn't understand what you said.")
except sr.RequestError as e:
    print(f"Error with the offline recognition engine: {e}")

Key Notes for Offline Usage #

• Accuracy: Offline recognition won't match the precision of Google's online service, especially in noisy environments or with less common vocabulary.
• Language Support: By default, PocketSphinx uses English. To use Chinese (or other languages), specify the language code in the recognize_sphinx call. For example:

  text = r.recognize_sphinx(audio, language='zh-CN')
  

  Most distributions include common language models by default, but if you run into issues, you can download the specific model files from PocketSphinx's official resources.
• Model Tweaks: For better accuracy tailored to your use case, you can customize acoustic/language models, but that requires more advanced setup.

内容的提问来源于stack exchange，提问作者nadhem