OpenVoice

Clone any voice with just seconds of audio using OpenVoice.

All examples can be run on GPU servers rented through CLORE.AI Marketplace.

Renting on CLORE.AI

1. Visit CLORE.AI Marketplace

2. Filter by GPU type, VRAM, and price

3. Choose On-Demand (fixed rate) or Spot (bid price)

4. Configure your order:

   • Select Docker image

   • Set ports (TCP for SSH, HTTP for web UIs)

   • Add environment variables if needed

   • Enter startup command

5. Select payment: CLORE, BTC, or USDT/USDC

6. Create order and wait for deployment

Access Your Server

• Find connection details in My Orders

• Web interfaces: Use the HTTP port URL

• SSH: ssh -p <port> root@<proxy-address>

What is OpenVoice?

OpenVoice by MyShell can:

• Clone voices from ~10 seconds of audio

• Control emotion, accent, rhythm

• Cross-lingual voice cloning

• Zero-shot voice conversion

Requirements

Task	Min VRAM	Recommended
Inference	4GB	RTX 3060
Batch processing	6GB	RTX 3070

Quick Deploy

Docker Image:

pytorch/pytorch:2.5.1-cuda12.4-cudnn9-runtime

Ports:

22/tcp
7860/http

Command:

pip install git+https://github.com/myshell-ai/OpenVoice.git gradio && \
python -c "
import gradio as gr
from openvoice import se_extractor
from openvoice.api import ToneColorConverter
import torch

ckpt_converter = 'checkpoints_v2/converter'
device = 'cuda'
tone_color_converter = ToneColorConverter(f'{ckpt_converter}/config.json', device=device)
tone_color_converter.load_ckpt(f'{ckpt_converter}/checkpoint.pth')

def clone(source_audio, reference_audio):
    source_se, _ = se_extractor.get_se(source_audio, tone_color_converter, vad=False)
    target_se, _ = se_extractor.get_se(reference_audio, tone_color_converter, vad=False)

    output_path = 'output.wav'
    tone_color_converter.convert(
        audio_src_path=source_audio,
        src_se=source_se,
        tgt_se=target_se,
        output_path=output_path
    )
    return output_path

demo = gr.Interface(
    fn=clone,
    inputs=[gr.Audio(type='filepath', label='Source'), gr.Audio(type='filepath', label='Target Voice')],
    outputs=gr.Audio(label='Cloned'),
    title='OpenVoice Clone'
)
demo.launch(server_name='0.0.0.0', server_port=7860)
"

Accessing Your Service

After deployment, find your http_pub URL in My Orders:

1. Go to My Orders page

2. Click on your order

3. Find the http_pub URL (e.g., abc123.clorecloud.net)

Use https://YOUR_HTTP_PUB_URL instead of localhost in examples below.

Installation

git clone https://github.com/myshell-ai/OpenVoice.git
cd OpenVoice
pip install -e .

# Download checkpoints
python -c "from huggingface_hub import snapshot_download; snapshot_download(repo_id='myshell-ai/OpenVoice', local_dir='checkpoints')"

Basic Voice Cloning

from openvoice import se_extractor
from openvoice.api import ToneColorConverter
import torch

# Initialize
device = "cuda" if torch.cuda.is_available() else "cpu"
ckpt_converter = 'checkpoints_v2/converter'

tone_color_converter = ToneColorConverter(
    f'{ckpt_converter}/config.json',
    device=device
)
tone_color_converter.load_ckpt(f'{ckpt_converter}/checkpoint.pth')

# Extract speaker embeddings
source_se, _ = se_extractor.get_se("source_audio.wav", tone_color_converter, vad=False)
target_se, _ = se_extractor.get_se("target_voice.wav", tone_color_converter, vad=False)

# Convert voice
tone_color_converter.convert(
    audio_src_path="source_audio.wav",
    src_se=source_se,
    tgt_se=target_se,
    output_path="output.wav"
)

With Text-to-Speech

Generate speech in any voice:

from openvoice import se_extractor
from openvoice.api import ToneColorConverter, BaseSpeakerTTS
from melo.api import TTS

# Initialize TTS
tts = TTS(language='EN', device=device)
speaker_ids = tts.hps.data.spk2id

# Generate base speech
tts.tts_to_file("Hello, this is a test.", speaker_ids['EN-US'], "base.wav")

# Clone to target voice
source_se, _ = se_extractor.get_se("base.wav", tone_color_converter, vad=False)
target_se, _ = se_extractor.get_se("target_voice.wav", tone_color_converter, vad=False)

tone_color_converter.convert(
    audio_src_path="base.wav",
    src_se=source_se,
    tgt_se=target_se,
    output_path="cloned_speech.wav"
)

Multi-Language Support

from melo.api import TTS

# Available languages
languages = ['EN', 'ES', 'FR', 'ZH', 'JP', 'KR']

# English
tts_en = TTS(language='EN', device=device)
tts_en.tts_to_file("Hello world", tts_en.hps.data.spk2id['EN-US'], "en.wav")

# Chinese
tts_zh = TTS(language='ZH', device=device)
tts_zh.tts_to_file("你好世界", tts_zh.hps.data.spk2id['ZH'], "zh.wav")

# Japanese
tts_jp = TTS(language='JP', device=device)
tts_jp.tts_to_file("こんにちは", tts_jp.hps.data.spk2id['JP'], "jp.wav")

Emotion Control

OpenVoice V2 supports emotion/style control:

from openvoice.api import BaseSpeakerTTS

# Base TTS with styles
base_speaker_tts = BaseSpeakerTTS(
    f'{ckpt_base}/config.json',
    device=device
)
base_speaker_tts.load_ckpt(f'{ckpt_base}/checkpoint.pth')

# Available styles
styles = ['default', 'whispering', 'cheerful', 'terrified', 'angry', 'sad', 'friendly']

for style in styles:
    base_speaker_tts.tts(
        "This is a test sentence.",
        f"output_{style}.wav",
        speaker='default',
        language='English',
        style=style
    )

Batch Processing

import os
from openvoice import se_extractor
from openvoice.api import ToneColorConverter

tone_color_converter = ToneColorConverter(
    f'{ckpt_converter}/config.json',
    device='cuda'
)
tone_color_converter.load_ckpt(f'{ckpt_converter}/checkpoint.pth')

# Get target voice embedding once
target_se, _ = se_extractor.get_se("target_voice.wav", tone_color_converter, vad=False)

input_dir = "./audio_files"
output_dir = "./cloned"
os.makedirs(output_dir, exist_ok=True)

for filename in os.listdir(input_dir):
    if filename.endswith(('.wav', '.mp3')):
        input_path = os.path.join(input_dir, filename)
        output_path = os.path.join(output_dir, f"cloned_{filename}")

        source_se, _ = se_extractor.get_se(input_path, tone_color_converter, vad=False)

        tone_color_converter.convert(
            audio_src_path=input_path,
            src_se=source_se,
            tgt_se=target_se,
            output_path=output_path
        )
        print(f"Cloned: {filename}")

API Server

from fastapi import FastAPI, UploadFile
from fastapi.responses import FileResponse
from openvoice import se_extractor
from openvoice.api import ToneColorConverter
import tempfile
import shutil

app = FastAPI()

tone_color_converter = ToneColorConverter(
    'checkpoints_v2/converter/config.json',
    device='cuda'
)
tone_color_converter.load_ckpt('checkpoints_v2/converter/checkpoint.pth')

@app.post("/clone")
async def clone_voice(source: UploadFile, target: UploadFile):
    # Save uploaded files
    with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as src_tmp:
        shutil.copyfileobj(source.file, src_tmp)
        src_path = src_tmp.name

    with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tgt_tmp:
        shutil.copyfileobj(target.file, tgt_tmp)
        tgt_path = tgt_tmp.name

    # Extract embeddings
    source_se, _ = se_extractor.get_se(src_path, tone_color_converter, vad=False)
    target_se, _ = se_extractor.get_se(tgt_path, tone_color_converter, vad=False)

    # Convert
    output_path = tempfile.mktemp(suffix=".wav")
    tone_color_converter.convert(
        audio_src_path=src_path,
        src_se=source_se,
        tgt_se=target_se,
        output_path=output_path
    )

    return FileResponse(output_path, media_type="audio/wav")

# Run: uvicorn server:app --host 0.0.0.0 --port 8000

Quality Tips

For Best Results

• Use 10-30 seconds of clear reference audio

• Avoid background noise

• Single speaker only in reference

• Match speaking pace approximately

Audio Preprocessing

import librosa
import soundfile as sf

def preprocess_audio(input_path, output_path, target_sr=22050):
    audio, sr = librosa.load(input_path, sr=target_sr)

    # Trim silence
    audio, _ = librosa.effects.trim(audio, top_db=20)

    # Normalize
    audio = librosa.util.normalize(audio)

    sf.write(output_path, audio, target_sr)
    return output_path

preprocess_audio("raw_reference.wav", "clean_reference.wav")

Comparison with Other Tools

Feature	OpenVoice	RVC	Bark
Reference Audio	10-30s	10+ min	N/A
Training	Not needed	Required	N/A
Speed	Fast	Medium	Slow
Quality	Great	Best	Good
Cross-lingual	Yes	Limited	Yes

Performance

Task	GPU	Time
Extract embedding	RTX 3090	~1s
Convert 10s audio	RTX 3090	~2s
Convert 1 min audio	RTX 3090	~8s

Troubleshooting

Poor Voice Match

• Use longer reference audio

• Ensure clear audio quality

• Check for background noise

Audio Artifacts

• Reduce speed/emphasis settings

• Use consistent audio format

• Check sampling rate match

Out of Memory

• Process shorter clips

• Reduce batch size

• Clear CUDA cache

Cost Estimate

Typical CLORE.AI marketplace rates (as of 2024):

GPU	Hourly Rate	Daily Rate	4-Hour Session
RTX 3060	~$0.03	~$0.70	~$0.12
RTX 3090	~$0.06	~$1.50	~$0.25
RTX 4090	~$0.10	~$2.30	~$0.40
A100 40GB	~$0.17	~$4.00	~$0.70
A100 80GB	~$0.25	~$6.00	~$1.00

Prices vary by provider and demand. Check CLORE.AI Marketplace for current rates.

Save money:

• Use Spot market for flexible workloads (often 30-50% cheaper)

• Pay with CLORE tokens

• Compare prices across different providers

Next Steps

• Bark TTS - Text-to-speech

• RVC Voice Clone - Training-based cloning

• Whisper Transcription - Speech-to-text