import os import sys import time import traceback import inspect import logging import shutil import subprocess import numpy as np import soundfile as sf import librosa import gradio as gr import scipy.signal as signal from datetime import datetime # ========================================== # 1. SETUP & IMPORTS # ========================================== print(">>> System Startup: RVC Pro Max...") try: import imageio_ffmpeg import static_ffmpeg from rvc_python.infer import RVCInference print("Libraries loaded successfully.") except ImportError as e: print(f"Import Error: {e}") sys.exit(1) # Setup FFmpeg try: static_ffmpeg.add_paths() ffmpeg_exe = imageio_ffmpeg.get_ffmpeg_exe() os.environ["PATH"] += os.pathsep + os.path.dirname(ffmpeg_exe) except Exception as e: print(f"FFmpeg Warning: {e}") TEMP_DIR = "/tmp/rvc_temp" os.makedirs(TEMP_DIR, exist_ok=True) os.environ["TEMP"] = TEMP_DIR os.environ["TMPDIR"] = TEMP_DIR # ========================================== # 2. AUDIO PROCESSING (DSP) # ========================================== def log_message(message): timestamp = datetime.now().strftime("%H:%M:%S") return f"[{timestamp}] {message}" def apply_clarity_eq(y, sr): # Apply EQ to fix nasal sound and boost clarity try: # 1. Low-Cut (remove rumble < 60Hz) sos_hp = signal.butter(4, 60, 'hp', fs=sr, output='sos') y = signal.sosfilt(sos_hp, y) # 2. Cut Nasal Frequencies (around 1000Hz) sos_mid = signal.butter(2, [800, 1200], 'bandstop', fs=sr, output='sos') y_filtered = signal.sosfilt(sos_mid, y) # Mix: 70% original, 30% filtered y = (y * 0.7) + (y_filtered * 0.3) # 3. High Boost (Air/Clarity > 5000Hz) sos_high = signal.butter(2, 5000, 'hp', fs=sr, output='sos') y_high = signal.sosfilt(sos_high, y) y = y + (y_high * 0.15) return y except Exception as e: print(f"EQ Error: {e}") return y def preprocess_audio(input_path): try: y, sr = librosa.load(input_path, sr=None) if y.ndim > 1: y = librosa.to_mono(y) y = librosa.util.normalize(y) processed_path = os.path.join(TEMP_DIR, "preprocessed.wav") sf.write(processed_path, y, sr) return processed_path, f"Pre-process OK (SR: {sr}Hz)" except Exception as e: return input_path, f"Pre-process Error: {e}" def post_process_audio(input_path, clarity_boost=True): try: y, sr = librosa.load(input_path, sr=None) if clarity_boost: y = apply_clarity_eq(y, sr) y = librosa.util.normalize(y) * 0.95 output_path = input_path.replace(".wav", "_final.wav") sf.write(output_path, y, sr) return output_path except Exception: return input_path def cleanup_temp(): try: for f in os.listdir(TEMP_DIR): os.remove(os.path.join(TEMP_DIR, f)) except Exception: pass # ========================================== # 3. CORE INFERENCE LOGIC # ========================================== def rvc_process_pipeline( audio_path, model_file, index_file, pitch_change, f0_method, index_rate, protect_val, filter_radius, resample_sr, envelope_mix, hop_length, enable_clarity ): logs = [] logs.append(log_message("Starting conversion...")) if not audio_path: return None, "Error: No audio file." if not model_file: return None, "Error: No model file." try: cleanup_temp() model_path = model_file.name index_path = index_file.name if index_file else None # Pre-process clean_audio, msg = preprocess_audio(audio_path) logs.append(log_message(msg)) # Load Model logs.append(log_message(f"Model: {os.path.basename(model_path)}")) rvc = RVCInference(device="cpu") rvc.load_model(model_path) output_temp = os.path.join(TEMP_DIR, f"rvc_out_{int(time.time())}.wav") # Params kwargs = { "input_path": clean_audio, "output_path": output_temp, "pitch": int(pitch_change), "method": f0_method, "index_path": index_path, "index_rate": float(index_rate), "protect": float(protect_val), "filter_radius": int(filter_radius), "resample_sr": int(resample_sr), "rms_mix_rate": float(envelope_mix), "hop_length": int(hop_length) } # Filter invalid params based on installed library version sig = inspect.signature(rvc.infer_file) valid_keys = sig.parameters.keys() final_kwargs = {} for k, v in kwargs.items(): if k in valid_keys: final_kwargs[k] = v elif k == "pitch" and "f0_up_key" in valid_keys: final_kwargs["f0_up_key"] = v elif k == "method" and "f0_method" in valid_keys: final_kwargs["f0_method"] = v logs.append(log_message(f"Method: {f0_method}")) start_time = time.time() rvc.infer_file(**final_kwargs) # Post-process final_output = output_temp if enable_clarity and os.path.exists(output_temp): logs.append(log_message("Applying clarity filter...")) final_output = post_process_audio(output_temp, clarity_boost=True) duration = time.time() - start_time logs.append(log_message(f"Done! ({duration:.2f}s)")) # SAFE STRING JOINING separator = chr(10) log_text = separator.join(logs) return final_output, log_text except Exception as e: separator = chr(10) err_msg = f"Error: {traceback.format_exc()}" print(err_msg) return None, err_msg # ========================================== # 4. GRADIO UI # ========================================== custom_css = """ #run_btn {background: linear-gradient(90deg, #FF5722 0%, #FF8A65 100%); color: white; border: none;} """ with gr.Blocks(title="RVC Pro Persian", theme=gr.themes.Soft(), css=custom_css) as demo: gr.Markdown("## RVC Pro: Professional Voice Converter") with gr.Row(): with gr.Column(): audio_input = gr.Audio(label="Input Audio", type="filepath") with gr.Row(): model_input = gr.File(label="Model (.pth)", file_types=[".pth"]) index_input = gr.File(label="Index (.index)", file_types=[".index"]) algo_dropdown = gr.Dropdown( choices=["rmvpe", "fcpe", "crepe", "harvest", "pm"], value="rmvpe", label="Algorithm" ) pitch_slider = gr.Slider(-24, 24, value=0, step=1, label="Pitch Change") btn_run = gr.Button("Start Conversion", elem_id="run_btn", variant="primary") with gr.Column(): with gr.Accordion("Quality Settings", open=True): enable_clarity = gr.Checkbox(value=True, label="Fix Nasal Sound (Clarity)") index_rate = gr.Slider(0, 1, value=0.4, step=0.05, label="Index Rate") envelope_mix = gr.Slider(0, 1, value=0.25, step=0.05, label="Volume Mix") protect_val = gr.Slider(0, 0.5, value=0.33, step=0.01, label="Protect") filter_radius = gr.Slider(0, 7, value=3, step=1, label="Filter Radius") resample_sr = gr.Slider(0, 48000, value=0, step=1000, label="Resample SR") hop_len = gr.Slider(1, 512, value=128, step=1, label="Hop Length") output_audio = gr.Audio(label="Final Output", type="filepath") logs = gr.Textbox(label="Logs", lines=5) btn_run.click( rvc_process_pipeline, inputs=[ audio_input, model_input, index_input, pitch_slider, algo_dropdown, index_rate, protect_val, filter_radius, resample_sr, envelope_mix, hop_len, enable_clarity ], outputs=[output_audio, logs] ) if __name__ == "__main__": demo.queue().launch(server_name="0.0.0.0", server_port=7860)