transcribing things I didn't say

[HXSmc]

it thinks I'm saying "thank you" apparently when I talk it transcribes what I say but when I don't talk it transcribes endless "thank you"

    import Speech_rec as sr
    from RealtimeSTT import AudioToTextRecorder # has to be implemented in the main script

    if __name__ == '__main__':
        record = AudioToTextRecorder()
        wakeup = AudioToTextRecorder(wake_words="jarvis")
        listener = sr.listener(record, wakeup)

        listener.run()

    class listener:
        def __init__(self, recorder, wakeup):
            self.recorder = recorder
            self.recorder.stop()
            self.sleep = True    
            self.wakeup = wakeup

        def listen(self):
            rec = ""
            rec = self.recorder.text() 
            return rec

        def sleep_mode(self):
            self.sleep = True
            print("Sleep mode enabled")
            self.recorder.stop()

        def run(self):
            print("Wait until it says 'speak now'")
            if not self.sleep:
                self.recorder.start()
                command = self.listen()
                print(f"Command:. {command}")
                if'sleep' in command.lower():
                    self.sleep_mode()
                    return None
                else:
                    return command.lower()
            elif self.sleep:
                self.wakeup.start()
                print("Wake up command ('jarvis')")
                command = self.wakeup.text()
                if command != None:
                    self.wakeup.stop()
                    self.sleep = False
                    print(command)
                    print("good morning")
                   self.recorder.start()

this is my whole code (oh it also wakes up on it's own)

[KoljaB]

Use only one recorder. Creating two recorders is overkill because you load the transciption models 2x into your VRAM. Leave out self.recorder.start(). Just use recorder.text(), it will detect when you start speaking. You get "thank you" etc. because recorder.start() initiates recording immediately, then everything you say and also what you NOT say gets transmitted to whisper. And for the parts where you say nothing whisper tends to hallucinate ("thank you" is a common whisper hallucination).

So you should call only the recorder.text() method, then voice activity will detect when to start recording and it will hallucinate way less. If you really want to call recorder.start() you want to make sure you start talking immediately after that.