SillyTavern-Extras11 / tts_edge.py
TomatoCocotree
上传
6a62ffb
raw
history blame
803 Bytes
import io
import edge_tts
import asyncio
def get_voices():
voices = asyncio.run(edge_tts.list_voices())
return voices
async def _iterate_chunks(audio):
async for chunk in audio.stream():
if chunk["type"] == "audio":
yield chunk["data"]
async def _async_generator_to_list(async_gen):
result = []
async for item in async_gen:
result.append(item)
return result
def generate_audio(text: str, voice: str, rate: int) -> bytes:
sign = '+' if rate > 0 else '-'
rate = f'{sign}{abs(rate)}%'
audio = edge_tts.Communicate(text=text, voice=voice, rate=rate)
chunks = asyncio.run(_async_generator_to_list(_iterate_chunks(audio)))
buffer = io.BytesIO()
for chunk in chunks:
buffer.write(chunk)
return buffer.getvalue()