Transitioning from PyAudio to SoundDevice for macOS M1 Compatibility
After encountering numerous challenges with PyAudio and PortAudio on the macOS M1 chip, particularly in the context of integrating with AssemblyAI, I decided to explore alternative solutions. My journey led me to adopt SoundDevice, a library that seamlessly operates on macOS without the need for additional external libraries. This decision has significantly streamlined the development process, offering a more stable and efficient audio interfacing solution.
To maintain compatibility with AssemblyAI and leverage SoundDevice's capabilities, I developed CustomMicrophoneStream. This class extends AssemblyAI's extras.MicrophoneStream, replacing PyAudio with SoundDevice for audio capture and processing. This custom implementation not only facilitates a smoother integration with AssemblyAI's real-time transcription services but also ensures compatibility with macOS M1 chips, circumventing the limitations previously encountered with PyAudio.
- MacOS M1 Compatibility: Fully compatible with macOS M1, eliminating the need for PyAudio or PortAudio.
- Seamless AssemblyAI Integration: Designed to integrate effortlessly with AssemblyAI, supporting real-time audio transcription without modifying the core workflow.
- SoundDevice as the Core Audio Library: Leverages SoundDevice for audio streaming, providing a robust and efficient audio capture mechanism.
To utilize CustomMicrophoneStream in your project, ensure you have SoundDevice installed:
pip install sounddevice aai.settings.api_key = os.getenv("ASSEMBLYAI_API_KEY")
# Initialize your custom microphone stream
microphone_stream = CustomMicrophoneStream()
transcriber = aai.RealtimeTranscriber(
on_data=on_data,
on_error=on_error,
sample_rate=44_100,
)
transcriber.connect()
# To check each chuck of audio uncomment the following lines
# mic_thread = threading.Thread(target=microphone_stream.run)
# mic_thread.start()
try:
print("Starting transcription stream.")
transcriber.stream(microphone_stream)
finally:
microphone_stream.close()
#mic_thread.join()class CustomMicrophoneStream:
def __init__(self, sample_rate=44100, chunk_size=4410, file_duration=5):
self.sample_rate = sample_rate
self.chunk_size = chunk_size
self.file_duration = file_duration
self.audio_buffer = queue.Queue()
self.frames_written = 0
self.file_index = 0
self.stream = sd.InputStream(callback=self.audio_callback,
samplerate=self.sample_rate,
channels=1,
blocksize=self.chunk_size)
self.stream.start()
self._open = True # Initialize the _open attribute
print("Audio stream started.")
self.prepare_new_file()
def prepare_new_file(self):
self.current_file = wave.open(f"part_{self.file_index}.wav", 'wb')
self.current_file.setnchannels(1)
self.current_file.setsampwidth(2) # Assuming 16-bit audio
self.current_file.setframerate(self.sample_rate)
self.frames_written = 0
self.file_index += 1
def audio_callback(self, indata, frames, time, status):
if status:
print(f"Stream status: {status}")
# Flatten the array and convert it to bytes
audio_bytes = (indata * 32767).astype(np.int16).tobytes()
#audio_bytes = indata.flatten().tobytes()
self.audio_buffer.put(audio_bytes)
#print(f"Audio callback: Buffer size is now {self.audio_buffer.qsize()} chunks.")
def write_to_file(self):
while not self.audio_buffer.empty():
data = self.audio_buffer.get()
self.current_file.writeframes(data)
self.frames_written += len(data)
# Check if it's time to start a new file
if self.frames_written >= self.sample_rate * self.file_duration:
self.current_file.close()
self.prepare_new_file()
def read(self, size):
# Read size bytes. If not enough data is available, block until enough is available.
requested_frames = size // 2 # 2 bytes per frame (16-bit audio)
frames_to_deliver = b''
while len(frames_to_deliver) < size:
frames_to_deliver += self.audio_buffer.get()
#print(f"Read {len(frames_to_deliver)} bytes from buffer.")
return frames_to_deliver[:size]
def __iter__(self):
return self
def __next__(self):
if self._open:
# Implement fetching the next chunk of audio data. You might need to adjust this logic.
try:
data = self.read(self.chunk_size)
if data:
return data
else:
# End of data or handle accordingly.
raise StopIteration
except queue.Empty:
# End of data or handle accordingly.
raise StopIteration
else:
# Stream is closed, stop iteration.
raise StopIteration
def close(self):
self.stream.stop()
self.stream.close()
self._open = False
print("Audio stream closed.")
self.current_file.close()
def run(self):
try:
while self._open:
self.write_to_file()
finally:
self.close()1 error generated.
error: command '/usr/bin/clang' failed with exit code 1
[end of output]
note: This error originates from a subprocess, and is likely not a problem with pip.
ERROR: Failed building wheel for pyaudio
Failed to build pyaudioAn error occured: Could not connect to the real-time service: [SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: unable to get local issuer certificate (_ssl.c:1007)
Could not import the PyAudio C module 'pyaudio._portaudio'.
Traceback (most recent call last):
File "/Users/sajithmr/coding/assemblyai/venv/lib/python3.10/site-packages/assemblyai/extras.py", line 37, in __init__
import pyaudio
File "/Users/sajithmr/coding/assemblyai/venv/lib/python3.10/site-packages/pyaudio/__init__.py", line 111, in <module>
import pyaudio._portaudio as pa
ImportError: dlopen(/Users/sajithmr/coding/assemblyai/venv/lib/python3.10/site-packages/pyaudio/_portaudio.cpython-310-darwin.so, 0x0002): symbol not found in flat namespace '_PaMacCore_SetupChannelMap'import certifi
import os
os.environ['SSL_CERT_FILE'] = certifi.where()||PaMacCore (AUHAL)|| AUHAL component not found.Traceback (most recent call last):
File "app.py", line 18, in <module>
stream = p.open(format=FORMAT, channels=CHANNELS, rate=RATE, input=True)
File "/Users/sajithmr/miniconda3/envs/myconda/lib/python3.8/site-packages/pyaudio/__init__.py", line 639, in open
stream = PyAudio.Stream(self, *args, **kwargs)
File "/Users/sajithmr/miniconda3/envs/myconda/lib/python3.8/site-packages/pyaudio/__init__.py", line 441, in __init__
self._stream = pa.open(**arguments)
OSError: [Errno -9999] Unanticipated host error
I replaced pyaudio and poraudio with CustomMicrophoneStream class using sounddevice