I’m working on a project where I need to use the speech_recognition module to process audio in real-time. However, from my research, it seems that speech_recognition (which works with pyaudio) only supports input sources like microphones and doesn’t seem to support capturing system output (the audio played through speakers). I need to find a way to capture the system’s output audio so that it can be processed by the recognizer.

I’ve tried using speech_recognition with pyaudio and sounddevice to record audio from the microphone, but neither captures system audio. I've researched into using both loopback methods and directly using output sources with speech_recognition, but most of them seem to be paid services? Any guidance/help would be appreciated.

Here's the code.

import numpy as np
import speech_recognition as sr
from time import sleep

def select_microphone():
    mics = sr.Microphone.list_microphone_names()
    input_mics = []
    for index, name in enumerate(sr.Microphone.list_microphone_names()):
        input_mics.append([index,name])
    
    if not input_mics:
        print("No available microphones.")
        exit(1)
    
    print("Available Microphones:")
    for i, mic in input_mics:
        print(f"{i}: {mic}")
    
    while True:
        try:
            choice = int(input("Select microphone index: "))
            if 0 <= choice < len(input_mics):
                return sr.Microphone(sample_rate=16000, device_index=input_mics[choice][0])
        except ValueError:
            pass
        print("Invalid selection. Try again.")

# Load model quickly
model = whisper.load_model("small")
data_queue = Queue()
transcription = ['']
recorder = sr.Recognizer()

# Select microphone
source = select_microphone()

def record_callback(_, audio: sr.AudioData):
    data_queue.put(audio.get_raw_data())

with source:
    recorder.adjust_for_ambient_noise(source)
recorder.listen_in_background(source, record_callback)

while True:
    try:
        if not data_queue.empty():
            audio_data = b''.join(data_queue.queue)
            data_queue.queue.clear()
            audio_np = np.frombuffer(audio_data, dtype=np.int16).astype(np.float32) / 32768.0
        else:
            sleep(0.25)
    except KeyboardInterrupt:
        break

print("\nFinal Transcription:")
print("\n".join(transcription))