File size: 1,659 Bytes
5165e58 0c20337 0c5c249 0c20337 0c5c249 5165e58 0c5c249 5165e58 0c5c249 0c20337 0c5c249 5165e58 0c5c249 0c20337 5165e58 0c5c249 0c20337 0c5c249 0c20337 0c5c249 0c20337 5165e58 0c5c249 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 |
import gradio as gr
import os
from openvoice.api import ToneColorConverter
from openvoice import se_extractor
import torch
import time
import uuid
# Set model paths
ckpt_converter = "checkpoints/converter"
output_dir = "outputs"
os.makedirs(output_dir, exist_ok=True)
# Initialize converter
tone_color_converter = ToneColorConverter(ckpt_converter)
# Load base speaker embedding for style transfer
ref_speaker_embed = None
def clone_and_speak(text, speaker_wav):
if not speaker_wav:
return "Please upload a reference .wav file."
# Generate a unique filename
timestamp = str(int(time.time()))
base_name = f"output_{timestamp}_{uuid.uuid4().hex[:6]}"
output_wav = os.path.join(output_dir, f"{base_name}.wav")
# Extract style from uploaded speaker voice
global ref_speaker_embed
ref_speaker_embed = se_extractor.get_se(speaker_wav, tone_color_converter)
# Generate speech using base model (internal prompt and sampling)
tone_color_converter.infer(
text=text,
speaker_id="openvoice",
language="en",
ref_speaker=speaker_wav,
ref_embed=ref_speaker_embed,
output_path=output_wav,
top_k=10,
temperature=0.3
)
return output_wav
demo = gr.Interface(
fn=clone_and_speak,
inputs=[
gr.Textbox(label="Enter Text"),
gr.Audio(type="filepath", label="Upload a Reference Voice (.wav)")
],
outputs=gr.Audio(label="Synthesized Output"),
title="Text to Voice using OpenVoice",
description="Clone any voice (English) and generate speech using OpenVoice on CPU.",
)
if __name__ == "__main__":
demo.launch()
|