import gradio as gr from pydub import AudioSegment from pydub.silence import detect_nonsilent 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 # from faster_whisper import WhisperModel # from TTS.api import TTS import base64 import pickle import json import soundfile as SF print("Gradio version:", gr.__version__) warnings.filterwarnings("ignore") # Helper to convert file to base64 def file_to_base64_audio(file_path, mime_type="audio/wav"): with open(file_path, "rb") as f: data = f.read() b64 = base64.b64encode(data).decode() return f"data:{mime_type};base64,{b64}" # === Effects Definitions === def apply_normalize(audio): return audio.normalize() def apply_noise_reduction(audio): samples, frame_rate = audiosegment_to_array(audio) reduced = nr.reduce_noise(y=samples, sr=frame_rate) return array_to_audiosegment(reduced, frame_rate, channels=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_frame_rate = int(audio.frame_rate * (2 ** (semitones / 12))) samples = np.array(audio.get_array_of_samples()) resampled = np.interp(np.arange(0, len(samples), 2 ** (semitones / 12)), np.arange(len(samples)), samples).astype(np.int16) return AudioSegment(resampled.tobytes(), frame_rate=new_frame_rate, sample_width=audio.sample_width, channels=audio.channels) 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, channels=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_val = 2 ** (bit_depth) - 1 downsampled = np.round(samples / (32768 / max_val)).astype(np.int16) return array_to_audiosegment(downsampled, audio.frame_rate // 2, channels=audio.channels) # === 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 ) # === Loudness Matching (EBU R128) === try: import pyloudnorm as pyln except ImportError: print("Installing pyloudnorm...") import subprocess subprocess.run(["pip", "install", "pyloudnorm"]) import pyloudnorm as pyln def match_loudness(audio_path, target_lufs=-14.0): meter = pyln.Meter(44100) wav = AudioSegment.from_file(audio_path).set_frame_rate(44100) samples = np.array(wav.get_array_of_samples()).astype(np.float64) / 32768.0 loudness = meter.integrated_loudness(samples) gain_db = target_lufs - loudness adjusted = wav + gain_db out_path = os.path.join(tempfile.gettempdir(), "loudness_output.wav") adjusted.export(out_path, format="wav") return out_path # Define eq_map at the global scope eq_map = { "Pop": [(200, 500, -3), (2000, 4000, +4)], "EDM": [(60, 250, +6), (8000, 12000, +3)], "Rock": [(1000, 3000, +4), (7000, 10000, -3)], "Hip-Hop": [(20, 100, +6), (7000, 10000, -4)], "Acoustic": [(100, 300, -3), (4000, 8000, +2)], "Metal": [(100, 500, -4), (2000, 5000, +6), (7000, 12000, -3)], "Trap": [(80, 120, +6), (3000, 6000, -4)], "LoFi": [(20, 200, +3), (1000, 3000, -2)], "Jazz": [(100, 400, +2), (1500, 3000, +1)], "Classical": [(200, 1000, +1), (3000, 6000, +2)], "Chillhop": [(50, 200, +3), (2000, 5000, +1)], "Ambient": [(100, 500, +4), (6000, 12000, +2)], "Jazz Piano": [(100, 1000, +3), (2000, 5000, +2)], "Trap EDM": [(60, 120, +6), (2000, 5000, -3)], "Indie Rock": [(150, 400, +2), (2000, 5000, +3)], "Lo-Fi Jazz": [(80, 200, +3), (2000, 4000, -1)], "R&B": [(100, 300, +4), (2000, 4000, +3)], "Soul": [(80, 200, +3), (1500, 3500, +4)], "Funk": [(80, 200, +5), (1000, 3000, +3)], "Default": [] } # Auto-EQ per Genre function def auto_eq(audio, genre="Pop"): from scipy.signal import butter, sosfilt def band_eq(samples, sr, lowcut, highcut, gain): sos = butter(10, [lowcut, highcut], btype='band', output='sos', fs=sr) filtered = sosfilt(sos, samples) return samples + gain * filtered samples, sr = audiosegment_to_array(audio) samples = samples.astype(np.float64) for band in eq_map.get(genre, []): low, high, gain = band samples = band_eq(samples, sr, low, high, gain) return array_to_audiosegment(samples.astype(np.int16), sr, channels=audio.channels) from scipy.signal import butter, sosfilt def band_eq(samples, sr, lowcut, highcut, gain): sos = butter(10, [lowcut, highcut], btype='band', output='sos', fs=sr) filtered = sosfilt(sos, samples) return samples + gain * filtered samples, sr = audiosegment_to_array(audio) samples = samples.astype(np.float64) for band in eq_map.get(genre, []): low, high, gain = band samples = band_eq(samples, sr, low, high, gain) return array_to_audiosegment(samples.astype(np.int16), sr, channels=audio.channels) # === Load Track Helpers === def load_track_local(path, sample_rate, channels=2): sig, rate = torchaudio.load(path) if rate != sample_rate: sig = torchaudio.functional.resample(sig, rate, sample_rate) if channels == 1: sig = sig.mean(0) return sig def save_track(path, wav, sample_rate): path = Path(path) torchaudio.save(str(path), wav, sample_rate) # === Vocal Isolation Helpers === def apply_vocal_isolation(audio_path): model = pretrained.get_model(name='htdemucs') wav = load_track_local(audio_path, model.samplerate, channels=2) ref = wav.mean(0) wav -= ref[:, None] sources = apply_model(model, wav[None])[0] wav += ref[:, None] vocal_track = sources[3].cpu() out_path = os.path.join(tempfile.gettempdir(), "vocals.wav") save_track(out_path, vocal_track, model.samplerate) return out_path # === Stem Splitting Function === def stem_split(audio_path): model = pretrained.get_model(name='htdemucs') wav = load_track_local(audio_path, model.samplerate, channels=2) sources = apply_model(model, wav[None])[0] output_dir = tempfile.mkdtemp() stem_paths = [] for i, name in enumerate(['drums', 'bass', 'other', 'vocals']): path = os.path.join(output_dir, f"{name}.wav") save_track(path, sources[i].cpu(), model.samplerate) stem_paths.append(gr.File(value=path)) return stem_paths # === Process Audio Function โ Fully Featured === def process_audio(audio_file, selected_effects, isolate_vocals, preset_name, export_format): status = "๐ Loading audio..." try: # Load input audio file audio = AudioSegment.from_file(audio_file) status = "๐ Applying effects..." effect_map_real = { "Noise Reduction": apply_noise_reduction, "Compress Dynamic Range": apply_compression, "Add Reverb": apply_reverb, "Pitch Shift": lambda x: apply_pitch_shift(x), "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": lambda x: apply_vocal_distortion(x), "Stage Mode": apply_stage_mode } history = [audio] # For undo functionality for effect_name in selected_effects: if effect_name in effect_map_real: audio = effect_map_real[effect_name](audio) history.append(audio) status = "๐พ Saving final audio..." with tempfile.NamedTemporaryFile(delete=False, suffix=f".{export_format.lower()}") as f: if isolate_vocals: temp_input = os.path.join(tempfile.gettempdir(), "input.wav") audio.export(temp_input, format="wav") vocal_path = apply_vocal_isolation(temp_input) final_audio = AudioSegment.from_wav(vocal_path) else: final_audio = audio output_path = f.name final_audio.export(output_path, format=export_format.lower()) waveform_image = show_waveform(output_path) genre = detect_genre(output_path) session_log = generate_session_log(audio_file, selected_effects, isolate_vocals, export_format, genre) status = "๐ Done!" return output_path, waveform_image, session_log, genre, status, history except Exception as e: status = f"โ Error: {str(e)}" return None, None, status, "", status, [] # Waveform preview 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", dpi=100) plt.close() buf.seek(0) return Image.open(buf) except Exception: return None # Genre detection stub def detect_genre(audio_path): try: y, sr = torchaudio.load(audio_path) return "Speech" except Exception: return "Unknown" # Session log generator def generate_session_log(audio_path, effects, isolate_vocals, export_format, genre): return json.dumps({ "timestamp": str(datetime.datetime.now()), "filename": os.path.basename(audio_path), "effects_applied": effects, "isolate_vocals": isolate_vocals, "export_format": export_format, "detected_genre": genre }, indent=2) # Preset Choices (30+ options) 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", "EQ Match"], "8-bit Retro": ["Bitcrusher", "Echo", "Mono Downmix"], "๐ Clean Vocal": ["Noise Reduction", "Normalize", "High Pass Filter (80Hz)"], "๐งช Vocal Distortion": ["Vocal Distortion", "Reverb", "Compress Dynamic Range"], "๐ถ Singer's Harmony": ["Harmony", "Stereo Widening", "Pitch Shift"], "๐ซ ASMR Vocal": ["Auto Gain", "Low-Pass Filter (3000Hz)", "Noise Gate"], "๐ผ Stage Mode": ["Reverb", "Bass Boost", "Limiter"], "๐ต Auto-Tune Style": ["Pitch Shift (+1 semitone)", "Normalize", "Treble Boost"], "๐ค R&B Vocal": ["Noise Reduction", "Bass Boost (100-300Hz)", "Treble Boost (2000-4000Hz)"], "๐ Soul Vocal": ["Noise Reduction", "Bass Boost (80-200Hz)", "Treble Boost (1500-3500Hz)"], "๐บ Funk Groove": ["Bass Boost (80-200Hz)", "Treble Boost (1000-3000Hz)"], "Studio Master": ["Noise Reduction", "Normalize", "Bass Boost", "Treble Boost", "Limiter"], "Podcast Voice": ["Noise Reduction", "Auto Gain", "High Pass Filter (85Hz)"], "Lo-Fi Chill": ["Noise Gate", "Low-Pass Filter (3000Hz)", "Mono Downmix", "Bitcrusher"], "Vocal Clarity": ["Noise Reduction", "EQ Match", "Reverb", "Auto Gain"], "Retro Game Sound": ["Bitcrusher", "Echo", "Mono Downmix"], "Live Stream Optimized": ["Noise Reduction", "Auto Gain", "Saturation", "Normalize"], "Deep Bass Trap": ["Bass Boost (60-120Hz)", "Low-Pass Filter (200Hz)", "Limiter"], "8-bit Voice": ["Bitcrusher", "Pitch Shift (-4 semitones)", "Mono Downmix"], "Pop Vocal": ["Noise Reduction", "Normalize", "EQ Match (Pop)", "Auto Gain"], "EDM Lead": ["Noise Reduction", "Tape Saturation", "Stereo Widening", "Limiter"], "Hip-Hop Beat": ["Bass Boost (60-200Hz)", "Treble Boost (7000-10000Hz)", "Compression"], "ASMR Whisper": ["Noise Gate", "Auto Gain", "Low-Pass Filter (5000Hz)"], "Jazz Piano Clean": ["Noise Reduction", "EQ Match (Jazz Piano)", "Normalize"], "Metal Guitar": ["Noise Reduction", "EQ Match (Metal)", "Compression"], "Podcast Intro": ["Echo", "Reverb", "Pitch Shift (+1 semitone)"], "Vintage Radio": ["Bitcrusher", "Low-Pass Filter (4000Hz)", "Saturation"], "Speech Enhancement": ["Noise Reduction", "High Pass Filter (100Hz)", "Normalize", "Auto Gain"], "Nightcore Speed": ["Pitch Shift (+3 semitones)", "Time Stretch (1.2x)", "Treble Boost"], "Robot Voice": ["Pitch Shift (-12 semitones)", "Bitcrusher", "Low-Pass Filter (2000Hz)"], "Underwater Effect": ["Low-Pass Filter (1000Hz)", "Reverb", "Echo"], "Alien Voice": ["Pitch Shift (+7 semitones)", "Tape Saturation", "Echo"], "Cinematic Voice": ["Reverb", "Limiter", "Bass Boost", "Auto Gain"], "Phone Call Sim": ["Low-Pass Filter (3400Hz)", "Noise Gate", "Compression"], "AI Generated Voice": ["Pitch Shift", "Vocal Distortion"], } preset_names = list(preset_choices.keys()) # Batch Processing def batch_process_audio(files, selected_effects, isolate_vocals, preset_name, export_format): try: output_dir = tempfile.mkdtemp() results = [] session_logs = [] for file in files: processed_path, _, log, _, _ = process_audio(file.name, selected_effects, isolate_vocals, preset_name, export_format)[0:5] results.append(processed_path) session_logs.append(log) zip_path = os.path.join(tempfile.gettempdir(), "batch_output.zip") with zipfile.ZipFile(zip_path, 'w') as zipf: for i, res in enumerate(results): filename = f"processed_{i}.{export_format.lower()}" zipf.write(res, filename) zipf.writestr(f"session_info_{i}.json", session_logs[i]) return zip_path, "๐ฆ ZIP created successfully!" except Exception as e: return None, f"โ Batch processing failed: {str(e)}" # AI Remastering def ai_remaster(audio_path): try: audio = AudioSegment.from_file(audio_path) samples, sr = audiosegment_to_array(audio) reduced = nr.reduce_noise(y=samples, sr=sr) cleaned = array_to_audiosegment(reduced, sr, channels=audio.channels) cleaned_wav_path = os.path.join(tempfile.gettempdir(), "cleaned.wav") cleaned.export(cleaned_wav_path, format="wav") isolated_path = apply_vocal_isolation(cleaned_wav_path) final_path = ai_mastering_chain(isolated_path, genre="Pop", target_lufs=-14.0) return final_path except Exception as e: print(f"Remastering Error: {str(e)}") return None def ai_mastering_chain(audio_path, genre="Pop", target_lufs=-14.0): audio = AudioSegment.from_file(audio_path) audio = auto_eq(audio, genre=genre) audio = match_loudness(audio_path, target_lufs=target_lufs) audio = apply_stereo_widen(audio, pan_amount=0.3) out_path = os.path.join(tempfile.gettempdir(), "mastered_output.wav") audio.export(out_path, format="wav") return out_path # Harmonic Saturation def harmonic_saturation(audio, saturation_type="Tube", intensity=0.2): samples = np.array(audio.get_array_of_samples()).astype(np.float32) if saturation_type == "Tube": saturated = np.tanh(intensity * samples) elif saturation_type == "Tape": saturated = np.where(samples > 0, 1 - np.exp(-intensity * samples), -1 + np.exp(intensity * samples)) elif saturation_type == "Console": saturated = np.clip(samples, -32768, 32768) * intensity elif saturation_type == "Mix Bus": saturated = np.log1p(np.abs(samples)) * np.sign(samples) * intensity else: saturated = samples return array_to_audiosegment(saturated.astype(np.int16), audio.frame_rate, channels=audio.channels) # Vocal Formant Correction def formant_correct(audio, shift=1.0): samples, sr = audiosegment_to_array(audio) corrected = librosa.effects.pitch_shift(samples, sr=sr, n_steps=shift) return array_to_audiosegment(corrected.astype(np.int16), sr, channels=audio.channels) # Voice Swap def clone_voice(source_audio, reference_audio): source = AudioSegment.from_file(source_audio) ref = AudioSegment.from_file(reference_audio) mixed = source.overlay(ref - 10) out_path = os.path.join(tempfile.gettempdir(), "cloned_output.wav") mixed.export(out_path, format="wav") return out_path # Save/Load Mix Session (.aiproj) def save_project(audio, preset, effects): project_data = { "audio": AudioSegment.from_file(audio).raw_data, "preset": preset, "effects": effects } out_path = os.path.join(tempfile.gettempdir(), "project.aiproj") with open(out_path, "wb") as f: pickle.dump(project_data, f) return out_path def load_project(project_file): with open(project_file.name, "rb") as f: data = pickle.load(f) return data["preset"], data["effects"] # Prompt-Based Editing def process_prompt(audio, prompt): return apply_noise_reduction(audio) # Vocal Pitch Correction def auto_tune_vocal(audio_path, target_key="C"): try: audio = AudioSegment.from_file(audio_path) semitones = key_to_semitone(target_key) tuned_audio = apply_pitch_shift(audio, semitones) out_path = os.path.join(tempfile.gettempdir(), "autotuned_output.wav") tuned_audio.export(out_path, format="wav") return out_path except Exception as e: print(f"Auto-Tune Error: {e}") return None def key_to_semitone(key="C"): keys = {"C": 0, "C#": 1, "D": 2, "D#": 3, "E": 4, "F": 5, "F#": 6, "G": 7, "G#": 8, "A": 9, "A#": 10, "B": 11} return keys.get(key, 0) # Loop Section Tool def loop_section(audio_path, start_ms, end_ms, loops=2): audio = AudioSegment.from_file(audio_path) section = audio[start_ms:end_ms] looped = section * loops out_path = os.path.join(tempfile.gettempdir(), "looped_output.wav") looped.export(out_path, format="wav") return out_path # Frequency Spectrum Visualization def visualize_spectrum(audio_path): y, sr = torchaudio.load(audio_path) y_np = y.numpy().flatten() stft = librosa.stft(y_np) db = librosa.amplitude_to_db(abs(stft)) plt.figure(figsize=(10, 4)) img = librosa.display.specshow(db, sr=sr, x_axis="time", y_axis="hz", cmap="magma") plt.colorbar(img, format="%+2.0f dB") plt.title("Frequency Spectrum") plt.tight_layout() buf = BytesIO() plt.savefig(buf, format="png") plt.close() buf.seek(0) return Image.open(buf) # A/B Compare def compare_ab(track1_path, track2_path): return track1_path, track2_path # DAW Template Export def generate_ableton_template(stems): template = { "format": "Ableton Live", "stems": [os.path.basename(s) for s in stems], "effects": ["Reverb", "EQ", "Compression"], "tempo": 128, "title": "Studio Pulse Project" } out_path = os.path.join(tempfile.gettempdir(), "ableton_template.json") with open(out_path, "w") as f: json.dump(template, f, indent=2) return out_path # Export Full Mix ZIP def export_full_mix(stems, final_mix): zip_path = os.path.join(tempfile.gettempdir(), "full_export.zip") with zipfile.ZipFile(zip_path, "w") as zipf: for i, stem in enumerate(stems): zipf.write(stem, f"stem_{i}.wav") zipf.write(final_mix, "final_mix.wav") return zip_path # Text-to-Sound # Main UI with gr.Blocks(css=""" body { font-family: 'Segoe UI', sans-serif; background-color: #1f2937; color: white; padding: 20px; } .studio-header { text-align: center; margin-bottom: 30px; animation: float 3s ease-in-out infinite; } @keyframes float { 0%, 100% { transform: translateY(0); } 50% { transform: translateY(-10px); } } .gr-button { background-color: #2563eb !important; color: white !important; border-radius: 10px; padding: 10px 20px; box-shadow: 0 0 10px #2563eb44; border: none; } """) as demo: gr.HTML('''