23 lines
821 B
Python
23 lines
821 B
Python
|
from config.model import MODEL_DIR
|
||
|
|
|
||
|
|
ZIPVOICE = MODEL_DIR / "zipvoice"
|
||
|
|
|
||
|
|
TOKENS = ZIPVOICE / "tokens.txt"
|
||
|
|
ENCODER = ZIPVOICE / "encoder.int8.onnx"
|
||
|
|
DECODER = ZIPVOICE / "decoder.int8.onnx"
|
||
|
|
DATA_DIR = ZIPVOICE / "espeak-ng-data"
|
||
|
|
LEXICON = ZIPVOICE / "lexicon.txt"
|
||
|
|
VOCODER = ZIPVOICE / "vocos_24khz.onnx"
|
||
|
|
|
||
|
|
# Reference voice for zero-shot voice cloning.
|
||
|
|
# REFERENCE_TEXT must match what is spoken in REFERENCE_AUDIO exactly,
|
||
|
|
# otherwise the cloned voice quality will noticeably degrade.
|
||
|
|
REFERENCE_AUDIO = ZIPVOICE / "test_wavs" / "leijun-1.wav"
|
||
|
|
REFERENCE_TEXT = "那还是三十六年前, 一九八七年. 我呢考上了武汉大学的计算机系."
|
||
|
|
|
||
|
|
NUM_THREADS = 2
|
||
|
|
PROVIDER = "cpu"
|
||
|
|
NUM_STEPS = 4 # Generation quality/speed tradeoff (higher = better, slower)
|
||
|
|
|
||
|
|
# Default output filename.
|
||
|
|
OUTPUT_FILE = "output.wav"
|