102 lines
2.8 KiB
Python
102 lines
2.8 KiB
Python
#encoding = utf8
|
|
import logging
|
|
import asyncio
|
|
import time
|
|
|
|
import edge_tts
|
|
import numpy as np
|
|
import soundfile
|
|
import resampy
|
|
import queue
|
|
from io import BytesIO
|
|
from queue import Queue
|
|
from threading import Thread, Event
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
class TTSBase:
|
|
def __init__(self, human):
|
|
self._human = human
|
|
self._thread = None
|
|
self._queue = Queue()
|
|
self._exit_event = None
|
|
self._io_stream = BytesIO()
|
|
self._sample_rate = 16000
|
|
self._chunk = self._sample_rate // self._human.get_fps()
|
|
|
|
self._exit_event = Event()
|
|
self._thread = Thread(target=self._on_run)
|
|
self._exit_event.set()
|
|
self._thread.start()
|
|
logging.info('tts start')
|
|
|
|
def _on_run(self):
|
|
logging.info('tts run')
|
|
while self._exit_event.is_set():
|
|
try:
|
|
txt = self._queue.get(block=True, timeout=1)
|
|
except queue.Empty:
|
|
continue
|
|
self._request(txt)
|
|
logging.info('tts exit')
|
|
|
|
def _request(self, txt):
|
|
voice = 'zh-CN-XiaoyiNeural'
|
|
t = time.time()
|
|
asyncio.new_event_loop().run_until_complete(self.__on_request(voice, txt))
|
|
logger.info(f'edge tts time:{time.time() - t : 0.4f}s')
|
|
|
|
self._io_stream.seek(0)
|
|
stream = self.__create_bytes_stream(self._io_stream)
|
|
stream_len = stream.shape[0]
|
|
index = 0
|
|
while stream_len >= self._chunk:
|
|
self._human.push_audio_chunk(stream[index:index + self._chunk])
|
|
stream_len -= self._chunk
|
|
index += self._chunk
|
|
|
|
def __create_bytes_stream(self, io_stream):
|
|
stream, sample_rate = soundfile.read(io_stream)
|
|
logger.info(f'tts audio stream {sample_rate} : {stream.shape}')
|
|
stream = stream.astype(np.float32)
|
|
|
|
if stream.ndim > 1:
|
|
logger.warning(f'tts audio has {stream.shape[1]} channels, only use the first')
|
|
stream = stream[:, 1]
|
|
|
|
if sample_rate != self._sample_rate and stream.shape[0] > 0:
|
|
logger.warning(f'tts audio sample rate is {sample_rate}, resample to {self._sample_rate}')
|
|
stream = resampy.resample(x=stream, sr_orig=sample_rate, sr_new=self._sample_rate)
|
|
|
|
return stream
|
|
|
|
async def __on_request(self, voice, txt):
|
|
communicate = edge_tts.Communicate(txt, voice)
|
|
first = True
|
|
async for chuck in communicate.stream():
|
|
if first:
|
|
first = False
|
|
|
|
if chuck['type'] == 'audio':
|
|
self._io_stream.write(chuck['data'])
|
|
|
|
def stop(self):
|
|
if self._exit_event is None:
|
|
return
|
|
|
|
self._exit_event.clear()
|
|
self._thread.join()
|
|
logging.info('tts stop')
|
|
|
|
def clear(self):
|
|
self._queue.queue.clear()
|
|
|
|
def push_txt(self, txt):
|
|
self._queue.put(txt)
|
|
|
|
|
|
|
|
|
|
|