human/human/audio_mal_handler.py

#encoding = utf8
import logging
import queue
import time
from queue import Queue
from threading import Thread, Event, Condition

import numpy as np

from human.message_type import MessageType
from human_handler import AudioHandler
from utils import melspectrogram

logger = logging.getLogger(__name__)


class AudioMalHandler(AudioHandler):
    def __init__(self, context, handler):
        super().__init__(context, handler)

        self._queue = Queue()
        self._wait = False
        self._condition = Condition()
        self._exit_event = Event()
        self._thread = Thread(target=self._on_run)
        self._exit_event.set()
        self._thread.start()

        self.frames = []
        self.chunk = context.sample_rate // context.fps
        logger.info("AudioMalHandler init")

    def on_message(self, message):
        if message['msg_id'] == MessageType.Video_Render_Queue_Empty:
            with self._condition:
                if self._wait:
                    self._wait = False
                    self._condition.notify()
                print('AudioMalHandler notify')
        elif message['msg_id'] == MessageType.Video_Render_Queue_Full:
            if not self._wait:
                self._wait = True
                print('AudioMalHandler wait')
        else:
            super().on_message(message)

    def on_handle(self, stream, index):
        self._queue.put(stream)

    def _on_run(self):
        logging.info('chunk2mal run')
        while self._exit_event.is_set():
            with self._condition:
                self._condition.wait_for(lambda: not self._wait)
                print('AudioMalHandler run')
            self._run_step()
            time.sleep(0.02)

        logging.info('chunk2mal exit')

    def _run_step(self):
        for _ in range(self._context.batch_size * 2):
            frame, _type = self.get_audio_frame()
            self.frames.append(frame)
            self.on_next_handle((frame, _type), 0)
        # context not enough, do not run network.
        if len(self.frames) <= self._context.stride_left_size + self._context.stride_right_size:
            return

        inputs = np.concatenate(self.frames)  # [N * chunk]
        mel = melspectrogram(inputs)
        # print(mel.shape[0],mel.shape,len(mel[0]),len(self.frames))
        # cut off stride
        left = max(0, self._context.stride_left_size * 80 / 50)
        right = min(len(mel[0]), len(mel[0]) - self._context.stride_right_size * 80 / 50)
        mel_idx_multiplier = 80. * 2 / self._context.fps
        mel_step_size = 16
        i = 0
        mel_chunks = []
        while i < (len(self.frames) - self._context.stride_left_size - self._context.stride_right_size) / 2:
            start_idx = int(left + i * mel_idx_multiplier)
            # print(start_idx)
            if start_idx + mel_step_size > len(mel[0]):
                mel_chunks.append(mel[:, len(mel[0]) - mel_step_size:])
            else:
                mel_chunks.append(mel[:, start_idx: start_idx + mel_step_size])
            i += 1
        self.on_next_handle(mel_chunks, 1)

        # discard the old part to save memory
        self.frames = self.frames[-(self._context.stride_left_size + self._context.stride_right_size):]

    def get_audio_frame(self):
        try:
            frame = self._queue.get(block=True, timeout=0.01)
            type_ = 0
        except queue.Empty:
            frame = np.zeros(self.chunk, dtype=np.float32)
            type_ = 1

        return frame, type_

    def stop(self):
        logging.info('stop')
        if self._exit_event is None:
            return

        self._exit_event.clear()
        if self._thread.is_alive():
            self._thread.join()
        logging.info('chunk2mal stop')

    def pause_talk(self):
        self._queue.queue.clear()