import subprocess import shutil import time import copy import glob import json import os from pathlib import Path import gradio as gr from threading import Thread from pathlib import Path current_dir = os.path.dirname(os.path.abspath(__file__)) MODELOS = os.path.join(current_dir, "modelos/") SEGMENTS_DIRNAME = os.path.join(current_dir, "segments") INFERENCE_OUTPUT_DIRNAME = os.path.join(current_dir, "inference_output") def slice_audio(filepath): assert os.path.exists(filepath), f"No se ha encontrado {filepath}." filename, extension = os.path.splitext(filepath) filename = filename.split("/")[-1] os.makedirs(SEGMENTS_DIRNAME, exist_ok=True) output_pattern = os.path.join(SEGMENTS_DIRNAME, f"{filename}_%d{extension}") os.system(f"ffmpeg -i {filepath} -f segment -segment_time 35 -c copy {output_pattern}") def get_container_format(filename): command = ["ffprobe", "-v", "error", "-select_streams", "v:0", "-show_entries", "format=format_name", "-of", "default=noprint_wrappers=1:nokey=1", filename] process = subprocess.Popen(command, stdout=subprocess.PIPE, stderr=subprocess.PIPE) output, error = process.communicate() if error: raise ValueError(f"Error: {error.decode()}") return output.decode().strip() def concatenate_segments(foldername, final_filename): foldername = Path(foldername) assert foldername.exists all_segs = [f for f in sorted(foldername.glob("**/*")) if f.is_file()] with open(foldername / "concat_list.txt", "w") as f: for seg in all_segs: f.write('file ' + str(seg.absolute()) + "\n") os.system(f"ffmpeg -f concat -safe 0 -i {foldername}/concat_list.txt -codec copy {foldername}/{final_filename}") def cleanup_dirs(): for dirname in (SEGMENTS_DIRNAME, INFERENCE_OUTPUT_DIRNAME): dir_path = Path(dirname) if dir_path.exists(): shutil.rmtree(dir_path) def get_speakers(): global speakers speakers = [] for _, dirs, _ in os.walk(MODELOS): for folder in dirs: cur_speaker = {} g = glob.glob(os.path.join(MODELOS, folder, 'G_*.pth')) if not len(g): continue cur_speaker["model_path"] = g[0] cur_speaker["model_folder"] = folder cur_speaker["cluster_path"] = "" cfg = glob.glob(os.path.join(MODELOS, folder, '*.json')) if not len(cfg): continue cur_speaker["cfg_path"] = cfg[0] with open(cur_speaker["cfg_path"]) as f: try: cfg_json = json.loads(f.read()) except Exception as e: print("Archivo json malformado en" + folder) for name, i in cfg_json["spk"].items(): cur_speaker["name"] = name cur_speaker["id"] = i if not name.startswith('.'): speakers.append(copy.copy(cur_speaker)) return sorted(speakers, key=lambda x: x["name"].lower()) def run_inference(speaker, seg_path, f0_method, transpose, noise_scale, cluster_ratio): model_path = speaker["model_path"] config_path = speaker["cfg_path"] cluster_path = speaker["cluster_path"] cluster_args = f"-k {cluster_path} -r {cluster_ratio}" if cluster_path and cluster_ratio > 0 else "" inference_cmd = f"svc infer {seg_path.absolute()} -m {model_path} -c {config_path} {cluster_args} -t {transpose} --f0-method {f0_method} -n {noise_scale} -o {INFERENCE_OUTPUT_DIRNAME}/{seg_path.name} --no-auto-predict-f0" result = subprocess.run(inference_cmd.split(), stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True) if result.stderr: if "AttributeError" in result.stderr: return None, gr.Textbox.update("⚠️ Modelo SVC incompatible.") if not list(Path(SEGMENTS_DIRNAME).glob("*")): return None, gr.Textbox.update("⚠️ Error.") def convert(speaker_box, audio): speaker = next((x for x in speakers if x["name"] == speaker_box), None) if not speaker: return None, gr.Textbox.update("⚠️ Selecciona un modelo.") if not audio: return None, gr.Textbox.update("⚠️ Sube un audio.") file_path = os.path.join(os.getcwd(), str(audio)) model_path = os.path.join(os.getcwd(), speaker["model_path"]) config_path = os.path.join(os.getcwd(), speaker["cfg_path"]) cluster_path = os.path.join(os.getcwd(), speaker["cluster_path"]) f0_method = "crepe" transpose = 0 noise_scale = 0.4 cluster_ratio = 0 if os.path.exists(SEGMENTS_DIRNAME) or os.path.exists(INFERENCE_OUTPUT_DIRNAME): cleanup_dirs() slice_audio(file_path) os.makedirs("inference_output", exist_ok=True) all_segs_paths = sorted(Path(SEGMENTS_DIRNAME).glob("*")) ts0 = time.time() for seg_path in all_segs_paths: run_inference(speaker, seg_path, f0_method, transpose, noise_scale, cluster_ratio) final_filename = f"output{Path(file_path).suffix}" concatenate_segments(INFERENCE_OUTPUT_DIRNAME, final_filename) shutil.move(Path(INFERENCE_OUTPUT_DIRNAME, final_filename), Path(final_filename)) cleanup_dirs() os.remove(file_path) ts1 = time.time() tiempo1 = int(ts1 - ts0) return final_filename, gr.Textbox.update("👌 ¡Voz cambiada!", label=f"Tiempo total: {tiempo1} segundos") def clear(): shutil.rmtree(SEGMENTS_DIRNAME, ignore_errors=True) shutil.rmtree(INFERENCE_OUTPUT_DIRNAME, ignore_errors=True) tmp_files = glob.glob("*.tmp") for f in tmp_files: os.remove(f) return gr.Dropdown.update(value="Elige un modelo de voz"), None, gr.Textbox.update("🗑️ Datos borrados.", label=f"Información") css = """ .gradio-container { font-family: 'IBM Plex Sans', sans-serif; } footer { visibility: hidden; display: none; } .center-container { display: flex; flex-direction: column; align-items: center; justify-content: center; } """ with gr.Blocks( css=css, title="VoiceIt! - Pavloh", theme=gr.themes.Soft( primary_hue="cyan", secondary_hue="blue", radius_size="lg", text_size="lg" ).set(loader_color="#0B0F19", shadow_drop='*shadow_drop_lg', block_border_width="3px") ) as pavloh: gr.HTML( """

Licencia GitHub
Twitter

🗣️ VoiceIt! - Un proyecto de Pavloh

Cambia la voz de audios utilizando modelos pre-entrenados de streamers.

""" ) with gr.Row(elem_id="1").style(equal_height=True): with gr.Column(): d1 = gr.Dropdown([x["name"] for x in get_speakers()], label="📦 Selecciona un modelo", value="Elige un modelo de voz") audio = gr.Audio(label="🗣️ Sube un audio", type="filepath") with gr.Column(): a2 = gr.Audio(label="🔊 Resultado", type="filepath") t1 = gr.Textbox(type="text", label="📄 Información", value="Elige un modelo y un audio para cambiar la voz.") with gr.Row(): b0 = gr.Button("🗑️ Borrar") b1 = gr.Button("🎤 Cambiar voz",variant="primary") b0.click(clear, outputs=[d1, audio, t1]) b1.click(convert, inputs=[d1, audio], outputs=[a2, t1]) with gr.Row(): with gr.Accordion(label="Licencia", open=False): gr.HTML("""

Ten en cuenta que los audios deben contener solamente una voz y estar libres de ruido o música de fondo.

Asegúrate de que el nombre del archivo no contenga espacios ni símbolos raros, utilizando solo caracteres alfanuméricos y guiones bajos (_) para separar palabras si es necesario.

En caso de que tarde más de 5 minutos procesar el audio, es posible que se produzca un error debido a los límites del servidor. En ese caso, tendrás que utilizar VoiceIt localmente usando tus recursos.

Al utilizar este sitio web, aceptas la licencia y condiciones de uso.

""") if __name__ == "__main__": pavloh.queue(concurrency_count=3) pavloh.launch()