Spleeter：“spleeter”和 PyAudio 的实时音频人声去除问题

Question

我一直在使用 Python 中的 spleeter 库从音频文件中分离人声，在处理预先录制的音频文件时效果很好。但是，我正在尝试使用带有 PyAudio 的 spleeter 来实现实时音频人声去除，但它似乎没有按预期工作。我已经编写了以下代码，但它没有产生所需的输出。我需要专家的帮助来解决问题。

from spleeter.separator import Separator
import multiprocessing
import pyaudio
import numpy as np

# Global variables
CHUNK_SIZE = 1024
SAMPLING_RATE = 16000
THIN_FACTOR = 0.5
vocals_data = bytes()

# Create PyAudio object
p = pyaudio.PyAudio()

# Define callback function for audio processing
def process_audio(in_data, frame_count, time_info, status):
    global vocals_data
    # Convert input data to numpy array
    audio_array = np.frombuffer(in_data, dtype=np.int16)
    
    # Perform vocal removal on the audio input
    # Pass the audio array as waveform to separate() method
    vocals = Separator('spleeter:2stems').separate(audio_array)
    
    # Convert vocals to audio data
    vocals_data = vocals['vocals'].flatten().astype(np.int16).tobytes()

    # Return processed data for output
    return vocals_data, pyaudio.paContinue

# Open stream for recording
stream = p.open(format=pyaudio.paInt16,
                channels=1,
                rate=SAMPLING_RATE,
                input=True,
                output=True,  # Set output to True for an output stream
                frames_per_buffer=CHUNK_SIZE,
                stream_callback=process_audio)

# Start stream
stream.start_stream()

# Create stream for playback
playback_stream = p.open(format=pyaudio.paInt16,
                         channels=1,
                         rate=SAMPLING_RATE,
                         output=True)

# Play processed data in real-time
while stream.is_active():
    if len(vocals_data) >= CHUNK_SIZE:
        playback_stream.write(vocals_data[:CHUNK_SIZE])
        vocals_data = vocals_data[CHUNK_SIZE:]

# Stop streams
stream.stop_stream()
stream.close()
playback_stream.stop_stream()
playback_stream.close()

# Terminate PyAudio object
p.terminate()

if __name__ == '__main__':
    multiprocessing.freeze_support()

我尝试使用 spleeter 和 PyAudio 实现实时音频人声去除。 我希望代码能够实时将人声与音频输入分开，并毫无问题地播放处理后的音频。

我也尝试过使用 multiprocessing.freeze_support() 但它没有解决问题。 任何帮助或建议将不胜感激！