import gradio as gr from pydub import AudioSegment import numpy as np import tempfile import os import noisereduce as nr import torch from demucs import pretrained from demucs.apply import apply_model import torchaudio from pathlib import Path import matplotlib.pyplot as plt from io import BytesIO from PIL import Image import zipfile import datetime import librosa import warnings import json import pickle import soundfile as sf warnings.filterwarnings("ignore") ### Helper Functions ### def audiosegment_to_array(audio): return np.array(audio.get_array_of_samples()), audio.frame_rate def array_to_audiosegment(samples, frame_rate, channels=1): return AudioSegment( samples.tobytes(), frame_rate=int(frame_rate), sample_width=samples.dtype.itemsize, channels=channels ) def save_audiosegment_to_temp(audio: AudioSegment, suffix=".wav"): with tempfile.NamedTemporaryFile(delete=False, suffix=suffix) as f: audio.export(f.name, format=suffix.lstrip('.')) return f.name def load_audiofile_to_numpy(path): samples, sr = sf.read(path, dtype="int16") if samples.ndim > 1 and samples.shape[1] > 2: samples = samples[:, :2] return samples, sr def show_waveform(audio_file): try: audio = AudioSegment.from_file(audio_file) samples = np.array(audio.get_array_of_samples()) plt.figure(figsize=(10,2)) plt.plot(samples[:10000], color="skyblue") plt.axis('off') buf = BytesIO() plt.savefig(buf, format='png', bbox_inches='tight') plt.close() buf.seek(0) return Image.open(buf) except Exception: return None ### Effects ### def apply_normalize(audio): return audio.normalize() def apply_noise_reduction(audio): samples, sr = audiosegment_to_array(audio) reduced = nr.reduce_noise(y=samples, sr=sr) return array_to_audiosegment(reduced, sr, audio.channels) def apply_compression(audio): return audio.compress_dynamic_range() def apply_reverb(audio): reverb = audio - 10 return audio.overlay(reverb, position=1000) def apply_pitch_shift(audio, semitones=-2): new_fr = int(audio.frame_rate * (2 ** (semitones / 12))) return audio._spawn(audio.raw_data, overrides={"frame_rate": new_fr}).set_frame_rate(audio.frame_rate) def apply_echo(audio, delay_ms=500, decay=0.5): echo = audio - 10 return audio.overlay(echo, position=delay_ms) def apply_stereo_widen(audio, pan_amount=0.3): left = audio.pan(-pan_amount) right = audio.pan(pan_amount) return AudioSegment.from_mono_audiosegments(left, right) def apply_bass_boost(audio, gain=10): return audio.low_pass_filter(100).apply_gain(gain) def apply_treble_boost(audio, gain=10): return audio.high_pass_filter(4000).apply_gain(gain) def apply_limiter(audio, limit_dB=-1): limiter = audio._spawn(audio.raw_data, overrides={"frame_rate": audio.frame_rate}) return limiter.apply_gain(limit_dB) def apply_auto_gain(audio, target_dB=-20): change = target_dB - audio.dBFS return audio.apply_gain(change) def apply_vocal_distortion(audio, intensity=0.3): samples = np.array(audio.get_array_of_samples()).astype(np.float32) distorted = samples + intensity * np.sin(samples * 2 * np.pi / 32768) return array_to_audiosegment(distorted.astype(np.int16), audio.frame_rate, audio.channels) def apply_harmony(audio, shift_semitones=4): shifted_up = apply_pitch_shift(audio, shift_semitones) shifted_down = apply_pitch_shift(audio, -shift_semitones) return audio.overlay(shifted_up).overlay(shifted_down) def apply_stage_mode(audio): processed = apply_reverb(audio) processed = apply_bass_boost(processed, gain=6) return apply_limiter(processed, limit_dB=-2) def apply_bitcrush(audio, bit_depth=8): samples = np.array(audio.get_array_of_samples()) max_value = 2 ** bit_depth - 1 downsampled = np.round(samples / (32768 / max_value)).astype(np.int16) return array_to_audiosegment(downsampled, audio.frame_rate // 2, audio.channels) ### Presets ### preset_choices = { "Default": [], "Clean Podcast": ["Noise Reduction", "Normalize"], "Podcast Mastered": ["Noise Reduction", "Normalize", "Compress Dynamic Range"], "Radio Ready": ["Bass Boost", "Treble Boost", "Limiter"], "Music Production": ["Reverb", "Stereo Widening", "Pitch Shift"], "ASMR Creator": ["Noise Gate", "Auto Gain", "Low-Pass Filter"], "Voiceover Pro": ["Vocal Isolation", "TTS", "EQ Match"], "8-bit Retro": ["Bitcrusher", "Echo", "Mono Downmix"], "๐ŸŽ™ Clean Vocal": ["Noise Reduction", "Normalize", "High Pass Filter (80Hz)"], "๐Ÿงช Vocal Distortion": ["Vocal Distortion", "Reverb", "Compress Dynamic Range"], "๐ŸŽถ Singer's Harmony": ["Harmony", "Stereo Widening", "Pitch Shift"], "๐ŸŒซ ASMR Vocal": ["Auto Gain", "Low-Pass Filter (3000Hz)", "Noise Gate"], "๐ŸŽผ Stage Mode": ["Reverb", "Bass Boost", "Limiter"], } preset_names = list(preset_choices.keys()) ### Main processing ### def process_audio(audio_file, selected_effects, isolate_vocals, preset_name, export_format): try: audio = AudioSegment.from_file(audio_file) effect_map = { "Noise Reduction": apply_noise_reduction, "Compress Dynamic Range": apply_compression, "Add Reverb": apply_reverb, "Pitch Shift": apply_pitch_shift, "Echo": apply_echo, "Stereo Widening": apply_stereo_widen, "Bass Boost": apply_bass_boost, "Treble Boost": apply_treble_boost, "Normalize": apply_normalize, "Limiter": lambda x: apply_limiter(x, limit_dB=-1), "Auto Gain": lambda x: apply_auto_gain(x, target_dB=-20), "Vocal Distortion": apply_vocal_distortion, "Stage Mode": apply_stage_mode, "Harmony": apply_harmony, "Bitcrusher": apply_bitcrush, } for effect in selected_effects: if effect in effect_map: audio = effect_map[effect](audio) if isolate_vocals: temp_path = save_audiosegment_to_temp(audio, suffix=".wav") vocal_path = apply_vocal_isolation(temp_path) audio = AudioSegment.from_file(vocal_path) output_path = save_audiosegment_to_temp(audio, suffix='.' + export_format.lower()) samples, sr = load_audiofile_to_numpy(output_path) waveform = show_waveform(output_path) session_log = json.dumps({ "timestamp": str(datetime.datetime.now()), "filename": os.path.basename(audio_file), "effects_applied": selected_effects, "isolate_vocals": isolate_vocals, "export_format": export_format, "detected_genre": "Unknown" }, indent=2) return (samples, sr), waveform, session_log, "Unknown", "๐ŸŽ‰ Done!" except Exception as e: return None, None, f"Error: {e}", "", f"Error: {e}" ### Other necessary functions (batch, AI remaster...) would follow similar patterns. # =================================================== # Now, the Gradio UI: # Paste this after all function definitions above # =================================================== with gr.Blocks() as demo: gr.HTML('

Where Your Audio Meets Intelligence

') gr.Markdown('### Upload, edit, export โ€” powered by AI!') with gr.Tab("๐ŸŽต Single File Studio"): with gr.Row(): with gr.Column(): input_audio = gr.Audio(label="Upload Audio", type="filepath") effect_checkbox = gr.CheckboxGroup(choices=list({e for effects in preset_choices.values() for e in effects}), label="Apply Effects in Order") preset_dropdown = gr.Dropdown(choices=preset_names, label="Select Preset") export_format = gr.Dropdown(choices=["WAV", "MP3"], label="Export Format", value="WAV") isolate_vocals = gr.Checkbox(label="Isolate Vocals After Effects") process_btn = gr.Button("Process Audio") with gr.Column(): processed_audio = gr.Audio(label="Processed Audio", type="numpy") waveform_image = gr.Image(label="Waveform Preview") session_log = gr.Textbox(label="Session Log", lines=6) detected_genre = gr.Textbox(label="Detected Genre") status = gr.Textbox(label="Status", lines=1, value="Ready") def update_effects(preset): return preset_choices.get(preset, []) preset_dropdown.change(update_effects, inputs=preset_dropdown, outputs=effect_checkbox) def run_processing(audio, effects, isolate, preset, fmt): effs = preset_choices.get(preset, []) if preset in preset_choices else effects return process_audio(audio, effs, isolate, preset, fmt) process_btn.click(run_processing, inputs=[input_audio, effect_checkbox, isolate_vocals, preset_dropdown, export_format], outputs=[processed_audio, waveform_image, session_log, detected_genre, status] ) demo.launch()