Python SpeechRecognition - OSError: [Errno -9998] 通道数无效

Question

I've been programming for several years but am quite new to Python, I'm implenting some speech recognition on a Raspberry Pi using https://pypi.org/project/SpeechRecognition/ and streaming the microphone output direction to Wit.ai.

I've been using PocketSphinx to constantly listen for the "wake word" and on my ubuntu VM the code below is working fine

class Recognition:
    def detection(self):
        speech = LiveSpeech(lm=False, keyphrase='alexa', kws_threshold=1e-20)

        print('Listening for wake word.');

        for phrase in speech:
            return self.start()

    def start(self):
        recognizer = sr.Recognizer()

        print('Heard wake word, listening and streaming audio.')

        with sr.Microphone(device_index=2) as source:
            audio = recognizer.listen(source)
        try:
            content = recognizer.recognize_wit(audio, key=WIT_AI_KEY)

            print(content)
        except sr.UnknownValueError:
            print("Unable to understand audio")
        except sr.RequestError as e:
            print("Error from Wit.ai: {}".format(e))


Recognition().detection()

But when running it on my Raspberry Pi, once the "wake word" has been triggered I'm getting the following error from PyAudio: OSError: [Errno -9998] Invalid number of channels

If I run Recognition.start() directly it seems to be working absolutely fine as long as I remove the device_index from the Microphone() ?

Does anyone know what the cause of this might be? As if I remove device_index and still run Recognition().detection() from the start I get a IOError: No Default Input Device Available

Just to note, when recording via arecord on my Raspberry Pi, the audio input/microphone seems to be recording absolutely fine.

Answer 1

For anyone that comes across this in the future, it seems to be an issue with using LiveSpeech and SpeechRecognition python modules and them having their own PyAudio instances.

I've had to change the code to initialise my own PyAudio instance and then use the Decoder sent with PocketSphinx for the wake word and then directly continue the same stream directly to requests for Wit.ai.