39 lines
1.1 KiB
Python
39 lines
1.1 KiB
Python
import logging
|
|
from typing import Iterator
|
|
|
|
import numpy as np
|
|
|
|
log = logging.getLogger(__name__)
|
|
|
|
DEFAULT_VOICE = "af_heart"
|
|
DEFAULT_LANG = "a" # American English
|
|
|
|
|
|
class TTSEngine:
|
|
"""Wraps Kokoro TTS for fast streaming text-to-speech."""
|
|
|
|
def __init__(self):
|
|
from kokoro import KPipeline
|
|
|
|
self.pipeline = KPipeline(lang_code=DEFAULT_LANG)
|
|
self.voice = DEFAULT_VOICE
|
|
self.sample_rate = 24000
|
|
|
|
def set_voice(self, voice: str, lang_code: str = "a"):
|
|
"""Change the voice."""
|
|
from kokoro import KPipeline
|
|
|
|
self.voice = voice
|
|
self.pipeline = KPipeline(lang_code=lang_code)
|
|
log.info(f"Voice set to: {voice} (lang: {lang_code})")
|
|
|
|
def synthesize_stream(self, text: str) -> Iterator[np.ndarray]:
|
|
"""Yield audio chunks as they are generated.
|
|
|
|
Each chunk is a float32 numpy array at self.sample_rate (24kHz).
|
|
Kokoro internally splits text into sentences and yields per-sentence audio.
|
|
"""
|
|
for _gs, _ps, audio in self.pipeline(text, voice=self.voice):
|
|
if audio is not None and len(audio) > 0:
|
|
yield audio
|