Sample code for voice transformation

[quantum304]

Hello,
let's say I have 2 audio files : f1.wav and f2.wav
I'd like to generate a new audio file f3.wav that contains the message in f1.wav but with the voice of f2.wav.
Is this possible with openvoice 2 ?
If yes can you please post a sample code ?

Also is it possible to use this with french messages and voices ?

Thanks for help.

PS: an AI tool helped me to generate the following code which works but the generated voice has no link with the voice I provided. It's a female voice while my target voice is a male voice. Furthermore, the generated message is read by an english speech synthesis while the original message is in french. I tried to change the language of the speech synthesis but it doesn't recognize french.

import torch
import numpy as np
from openvoice import se_extractor
from openvoice.api import BaseSpeakerTTS, ToneColorConverter

def clone_voice_en(f1_path, f2_path, output_path='f3.wav'):
# 1. Initialisation des modèles
checkpoint = 'C:/Users/me/Desktop/OV/openvoice/checkpoints/base_speakers/EN'
config_path = f'{checkpoint}/config.json'
model_path = f'{checkpoint}/checkpoint.pth'
device = "cuda" if torch.cuda.is_available() else "cpu"

# 2. Chargement des modèles
base_speaker_tts = BaseSpeakerTTS(f'{checkpoint}/config.json', device=device)
base_speaker_tts.load_ckpt(f'{checkpoint}/checkpoint.pth')

tone_color_converter = ToneColorConverter(
    f'C:/Users/me/Desktop/OV/openvoice/checkpoints/converter/config.json', 
    device=device
)
tone_color_converter.load_ckpt(f'C:/Users/me/Desktop/OV/openvoice/checkpoints/converter/checkpoint.pth')

# 3. Extraction du texte du fichier source (f1)
source_text = extract_text_from_audio_en(f1_path)  # À implémenter

# 4. Extraction des caractéristiques vocales (f2)
reference_speaker = f2_path
target_se, _ = se_extractor.get_se(
    reference_speaker, 
    tone_color_converter, 
    vad=True
)

# 5. Synthèse vocale
src_path = f'{checkpoint}/tmp.wav'
speaker = 'default'

# Génération de la voix de base
base_speaker_tts.tts(
    source_text, 
    src_path, 
    speaker=speaker, 
    language='english', 
    speed=1.0
)

# Application de la couleur tonale
encode_message = "@MyShell"
tone_color_converter.convert(
    audio_src_path=src_path, 
    src_se=target_se, 
    tgt_se=target_se, 
    output_path=output_path,
    message=encode_message
)

print(f"Fichier généré avec succès : {output_path}")

def extract_text_from_audio_en(audio_path):
"""Utilisation de Whisper pour la transcription"""
import whisper
model = whisper.load_model("base")
result = model.transcribe(audio_path, language="fr")
return result["text"]

if name == "main":
clone_voice_en("f1.wav", "f2.wav", "f3.wav")

[dosubot[bot]]

Yes, it is possible to use OpenVoice 2 to transform the message in message.wav to have the voice of voice.wav, and it supports French messages and voices. Here is a sample code snippet that demonstrates how to achieve this:

from openvoice.api import OpenVoice

# Initialize OpenVoice with necessary parameters
openvoice = OpenVoice()

# Define source and target speaker embeddings
src_se = 'path_to_source_speaker_embedding'  # Replace with actual path or embedding
tgt_se = 'path_to_target_speaker_embedding'  # Replace with actual path or embedding

# Convert the message in message.wav to have the voice of voice.wav
output_audio = openvoice.convert(
    audio_src_path='message.wav',
    src_se=src_se,
    tgt_se=tgt_se,
    output_path='output_message.wav',
    tau=0.3,
    message="default"
)

# The converted audio will be saved to 'output_message.wav'

Make sure to replace 'path_to_source_speaker_embedding' and 'path_to_target_speaker_embedding' with the actual paths or embeddings for the source and target speakers. This should help ensure the output voice matches the target voice, including gender and language ^[1].

To continue talking to Dosu, mention @dosu.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Bug Report | Other}

[quantum304]

can you please provide the whole code ?

[quantum304]

I get the following error with this code
`ImportError Traceback (most recent call last)
Cell In[6], line 1
----> 1 from openvoice.api import OpenVoice
3 # Initialize OpenVoice with necessary parameters
4 openvoice = OpenVoice()

ImportError: cannot import name 'OpenVoice' from 'openvoice.api'`