Create separwator.py

separwator.py

@@ -0,0 +1,601 @@
+import os
+import torch
+import logging
+import yt_dlp
+import spaces
+import gradio as gr
+from audio_separator.separator import Separator
+
+
+
+device = "cuda" if torch.cuda.is_available() else "cpu"
+use_autocast = device == "cuda"
+
+#=========================#
+#     Roformer Models     #
+#=========================#
+roformer_models = {
+    'BS-Roformer-Viperx-1297': 'model_bs_roformer_ep_317_sdr_12.9755.ckpt',
+    'BS-Roformer-Viperx-1296': 'model_bs_roformer_ep_368_sdr_12.9628.ckpt',
+    'BS-Roformer-Viperx-1053': 'model_bs_roformer_ep_937_sdr_10.5309.ckpt',
+    'Mel-Roformer-Viperx-1143': 'model_mel_band_roformer_ep_3005_sdr_11.4360.ckpt',
+    'BS-Roformer-De-Reverb': 'deverb_bs_roformer_8_384dim_10depth.ckpt',
+    'Mel-Roformer-Crowd-Aufr33-Viperx': 'mel_band_roformer_crowd_aufr33_viperx_sdr_8.7144.ckpt',
+    'Mel-Roformer-Denoise-Aufr33': 'denoise_mel_band_roformer_aufr33_sdr_27.9959.ckpt',
+    'Mel-Roformer-Denoise-Aufr33-Aggr' : 'denoise_mel_band_roformer_aufr33_aggr_sdr_27.9768.ckpt',
+    'Mel-Roformer-Karaoke-Aufr33-Viperx': 'mel_band_roformer_karaoke_aufr33_viperx_sdr_10.1956.ckpt',
+    'MelBand Roformer Kim | Inst V1 by Unwa' : 'melband_roformer_inst_v1.ckpt',
+    'MelBand Roformer Kim | Inst V2 by Unwa' : 'melband_roformer_inst_v2.ckpt',
+    'MelBand Roformer Kim | InstVoc Duality V1 by Unwa' : 'melband_roformer_instvoc_duality_v1.ckpt',
+    'MelBand Roformer Kim | InstVoc Duality V2 by Unwa' : 'melband_roformer_instvox_duality_v2.ckpt',
+}
+
+#=========================#
+#      MDX23C Models      #
+#=========================#
+mdx23c_models = [
+    'MDX23C_D1581.ckpt',
+    'MDX23C-8KFFT-InstVoc_HQ.ckpt',
+    'MDX23C-8KFFT-InstVoc_HQ_2.ckpt',
+]
+
+#=========================#
+#     MDXN-NET Models     #
+#=========================#
+mdxnet_models = [
+    'UVR-MDX-NET-Inst_full_292.onnx',
+    'UVR-MDX-NET_Inst_187_beta.onnx',
+    'UVR-MDX-NET_Inst_82_beta.onnx',
+    'UVR-MDX-NET_Inst_90_beta.onnx',
+    'UVR-MDX-NET_Main_340.onnx',
+    'UVR-MDX-NET_Main_390.onnx',
+    'UVR-MDX-NET_Main_406.onnx',
+    'UVR-MDX-NET_Main_427.onnx',
+    'UVR-MDX-NET_Main_438.onnx',
+    'UVR-MDX-NET-Inst_HQ_1.onnx',
+    'UVR-MDX-NET-Inst_HQ_2.onnx',
+    'UVR-MDX-NET-Inst_HQ_3.onnx',
+    'UVR-MDX-NET-Inst_HQ_4.onnx',
+    'UVR-MDX-NET-Inst_HQ_5.onnx',
+    'UVR_MDXNET_Main.onnx',
+    'UVR-MDX-NET-Inst_Main.onnx',
+    'UVR_MDXNET_1_9703.onnx',
+    'UVR_MDXNET_2_9682.onnx',
+    'UVR_MDXNET_3_9662.onnx',
+    'UVR-MDX-NET-Inst_1.onnx',
+    'UVR-MDX-NET-Inst_2.onnx',
+    'UVR-MDX-NET-Inst_3.onnx',
+    'UVR_MDXNET_KARA.onnx',
+    'UVR_MDXNET_KARA_2.onnx',
+    'UVR_MDXNET_9482.onnx',
+    'UVR-MDX-NET-Voc_FT.onnx',
+    'Kim_Vocal_1.onnx',
+    'Kim_Vocal_2.onnx',
+    'Kim_Inst.onnx',
+    'Reverb_HQ_By_FoxJoy.onnx',
+    'UVR-MDX-NET_Crowd_HQ_1.onnx',
+    'kuielab_a_vocals.onnx',
+    'kuielab_a_other.onnx',
+    'kuielab_a_bass.onnx',
+    'kuielab_a_drums.onnx',
+    'kuielab_b_vocals.onnx',
+    'kuielab_b_other.onnx',
+    'kuielab_b_bass.onnx',
+    'kuielab_b_drums.onnx',
+]
+
+#========================#
+#     VR-ARCH Models     #
+#========================#
+vrarch_models = [
+    '1_HP-UVR.pth',
+    '2_HP-UVR.pth',
+    '3_HP-Vocal-UVR.pth',
+    '4_HP-Vocal-UVR.pth',
+    '5_HP-Karaoke-UVR.pth',
+    '6_HP-Karaoke-UVR.pth',
+    '7_HP2-UVR.pth',
+    '8_HP2-UVR.pth',
+    '9_HP2-UVR.pth',
+    '10_SP-UVR-2B-32000-1.pth',
+    '11_SP-UVR-2B-32000-2.pth',
+    '12_SP-UVR-3B-44100.pth',
+    '13_SP-UVR-4B-44100-1.pth',
+    '14_SP-UVR-4B-44100-2.pth',
+    '15_SP-UVR-MID-44100-1.pth',
+    '16_SP-UVR-MID-44100-2.pth',
+    '17_HP-Wind_Inst-UVR.pth',
+    'UVR-De-Echo-Aggressive.pth',
+    'UVR-De-Echo-Normal.pth',
+    'UVR-DeEcho-DeReverb.pth',
+    'UVR-DeNoise-Lite.pth',
+    'UVR-DeNoise.pth',
+    'UVR-BVE-4B_SN-44100-1.pth',
+    'MGM_HIGHEND_v4.pth',
+    'MGM_LOWEND_A_v4.pth',
+    'MGM_LOWEND_B_v4.pth',
+    'MGM_MAIN_v4.pth',
+]
+
+#=======================#
+#     DEMUCS Models     #
+#=======================#
+demucs_models = [
+    'htdemucs_ft.yaml',
+    'htdemucs_6s.yaml',
+    'htdemucs.yaml',
+    'hdemucs_mmi.yaml',
+]
+
+output_format = [
+    'wav',
+    'flac',
+    'mp3',
+    'ogg',
+    'opus',
+    'm4a',
+    'aiff',
+    'ac3'
+]
+
+found_files = []
+logs = []
+out_dir = "./outputs"
+models_dir = "./models"
+extensions = (".wav", ".flac", ".mp3", ".ogg", ".opus", ".m4a", ".aiff", ".ac3")
+
+def download_audio(url, output_dir="ytdl"):
+
+    os.makedirs(output_dir, exist_ok=True)
+
+    ydl_opts = {
+        'format': 'bestaudio/best',
+        'postprocessors': [{
+            'key': 'FFmpegExtractAudio',
+            'preferredcodec': 'wav',
+            'preferredquality': '32',
+        }],
+        'outtmpl': os.path.join(output_dir, '%(title)s.%(ext)s'),
+        'postprocessor_args': [
+            '-acodec', 'pcm_f32le'
+        ],
+    }
+
+    try:
+        with yt_dlp.YoutubeDL(ydl_opts) as ydl:
+            info = ydl.extract_info(url, download=False)
+            video_title = info['title']
+
+            ydl.download([url])
+
+            file_path = os.path.join(output_dir, f"{video_title}.wav")
+
+            if os.path.exists(file_path):
+                return os.path.abspath(file_path)
+            else:
+                raise Exception("Something went wrong")
+
+    except Exception as e:
+        raise Exception(f"Error extracting audio with yt-dlp: {str(e)}")
+
+@spaces.GPU(duration=60)
+def roformer_separator(audio, model_key, out_format, segment_size, override_seg_size, overlap, batch_size, norm_thresh, amp_thresh, progress=gr.Progress(track_tqdm=True)):
+    base_name = os.path.splitext(os.path.basename(audio))[0]
+    roformer_model = roformer_models[model_key]
+    try:
+        separator = Separator(
+            log_level=logging.WARNING,
+            model_file_dir=models_dir,
+            output_dir=out_dir,
+            output_format=out_format,
+            use_autocast=use_autocast,
+            normalization_threshold=norm_thresh,
+            amplification_threshold=amp_thresh,
+            mdxc_params={
+                "segment_size": segment_size,
+                "override_model_segment_size": override_seg_size,
+                "batch_size": batch_size,
+                "overlap": overlap,
+            }
+        )
+    
+        progress(0.2, desc="Loading model...")
+        separator.load_model(model_filename=roformer_model)
+
+        progress(0.7, desc="Separating audio...")
+        separation = separator.separate(audio, f"{base_name}_(Stem1)", f"{base_name}_(Stem2)")
+
+        stems = [os.path.join(out_dir, file_name) for file_name in separation]
+        return stems[1], stems[0]
+    except Exception as e:
+        raise RuntimeError(f"Roformer separation failed: {e}") from e
+
+@spaces.GPU(duration=60)
+def mdxc_separator(audio, model, out_format, segment_size, override_seg_size, overlap, batch_size, norm_thresh, amp_thresh, progress=gr.Progress(track_tqdm=True)):
+    base_name = os.path.splitext(os.path.basename(audio))[0]
+    try:
+        separator = Separator(
+            log_level=logging.WARNING,
+            model_file_dir=models_dir,
+            output_dir=out_dir,
+            output_format=out_format,
+            use_autocast=use_autocast,
+            normalization_threshold=norm_thresh,
+            amplification_threshold=amp_thresh,
+            mdxc_params={
+                "segment_size": segment_size,
+                "override_model_segment_size": override_seg_size,
+                "batch_size": batch_size,
+                "overlap": overlap,
+            }
+        )
+
+        progress(0.2, desc="Loading model...")
+        separator.load_model(model_filename=model)
+
+        progress(0.7, desc="Separating audio...")
+        separation = separator.separate(audio, f"{base_name}_(Stem1)", f"{base_name}_(Stem2)")
+
+        stems = [os.path.join(out_dir, file_name) for file_name in separation]
+        return stems[1], stems[0]
+    except Exception as e:
+        raise RuntimeError(f"MDX23C separation failed: {e}") from e
+
+@spaces.GPU(duration=60)
+def mdxnet_separator(audio, model, out_format, hop_length, segment_size, denoise, overlap, batch_size, norm_thresh, amp_thresh, progress=gr.Progress(track_tqdm=True)):
+    base_name = os.path.splitext(os.path.basename(audio))[0]
+    try:
+        separator = Separator(
+            log_level=logging.WARNING,
+            model_file_dir=models_dir,
+            output_dir=out_dir,
+            output_format=out_format,
+            use_autocast=use_autocast,
+            normalization_threshold=norm_thresh,
+            amplification_threshold=amp_thresh,
+            mdx_params={
+                "hop_length": hop_length,
+                "segment_size": segment_size,
+                "overlap": overlap,
+                "batch_size": batch_size,
+                "enable_denoise": denoise,
+            }
+        )
+
+        progress(0.2, desc="Loading model...")
+        separator.load_model(model_filename=model)
+
+        progress(0.7, desc="Separating audio...")
+        separation = separator.separate(audio, f"{base_name}_(Stem1)", f"{base_name}_(Stem2)")
+
+        stems = [os.path.join(out_dir, file_name) for file_name in separation]
+        return stems[0], stems[1]
+    except Exception as e:
+        raise RuntimeError(f"MDX-NET separation failed: {e}") from e
+
+@spaces.GPU(duration=60)
+def vrarch_separator(audio, model, out_format, window_size, aggression, tta, post_process, post_process_threshold, high_end_process, batch_size, norm_thresh, amp_thresh, progress=gr.Progress(track_tqdm=True)):
+    base_name = os.path.splitext(os.path.basename(audio))[0]
+    try:
+        separator = Separator(
+            log_level=logging.WARNING,
+            model_file_dir=models_dir,
+            output_dir=out_dir,
+            output_format=out_format,
+            use_autocast=use_autocast,
+            normalization_threshold=norm_thresh,
+            amplification_threshold=amp_thresh,
+            vr_params={
+                "batch_size": batch_size,
+                "window_size": window_size,
+                "aggression": aggression,
+                "enable_tta": tta,
+                "enable_post_process": post_process,
+                "post_process_threshold": post_process_threshold,
+                "high_end_process": high_end_process,
+            }
+        )
+
+        progress(0.2, desc="Loading model...")
+        separator.load_model(model_filename=model)
+
+        progress(0.7, desc="Separating audio...")
+        separation = separator.separate(audio, f"{base_name}_(Stem1)", f"{base_name}_(Stem2)")
+
+        stems = [os.path.join(out_dir, file_name) for file_name in separation]
+        return stems[0], stems[1]
+    except Exception as e:
+        raise RuntimeError(f"VR ARCH separation failed: {e}") from e
+
+@spaces.GPU(duration=60)
+def demucs_separator(audio, model, out_format, shifts, segment_size, segments_enabled, overlap, batch_size, norm_thresh, amp_thresh, progress=gr.Progress(track_tqdm=True)):
+    base_name = os.path.splitext(os.path.basename(audio))[0]
+    try:
+        separator = Separator(
+            log_level=logging.WARNING,
+            model_file_dir=models_dir,
+            output_dir=out_dir,
+            output_format=out_format,
+            use_autocast=use_autocast,
+            normalization_threshold=norm_thresh,
+            amplification_threshold=amp_thresh,
+            demucs_params={
+                "batch_size": batch_size,
+                "segment_size": segment_size,
+                "shifts": shifts,
+                "overlap": overlap,
+                "segments_enabled": segments_enabled,
+            }
+        )
+
+        progress(0.2, desc="Loading model...")
+        separator.load_model(model_filename=model)
+
+        progress(0.7, desc="Separating audio...")
+        separation = separator.separate(audio)
+
+        stems = [os.path.join(out_dir, file_name) for file_name in separation]
+        
+        if model == "htdemucs_6s.yaml":
+            return stems[0], stems[1], stems[2], stems[3], stems[4], stems[5]
+        else:
+            return stems[0], stems[1], stems[2], stems[3], None, None
+    except Exception as e:
+        raise RuntimeError(f"Demucs separation failed: {e}") from e
+
+def update_stems(model):
+    if model == "htdemucs_6s.yaml":
+        return gr.update(visible=True)
+    else:
+        return gr.update(visible=False)
+
+@spaces.GPU(duration=60)
+def roformer_batch(path_input, path_output, model_key, out_format, segment_size, override_seg_size, overlap, batch_size, norm_thresh, amp_thresh):
+    found_files.clear()
+    logs.clear()
+    roformer_model = roformer_models[model_key]
+
+    for audio_files in os.listdir(path_input):
+        if audio_files.endswith(extensions):
+            found_files.append(audio_files)
+    total_files = len(found_files)
+
+    if total_files == 0:
+        logs.append("No valid audio files.")
+        yield "\n".join(logs)
+    else:
+        logs.append(f"{total_files} audio files found")
+        found_files.sort()
+
+        for audio_files in found_files:
+            file_path = os.path.join(path_input, audio_files)
+            base_name = os.path.splitext(os.path.basename(file_path))[0]
+            try:
+                separator = Separator(
+                    log_level=logging.WARNING,
+                    model_file_dir=models_dir,
+                    output_dir=path_output,
+                    output_format=out_format,
+                    use_autocast=use_autocast,
+                    normalization_threshold=norm_thresh,
+                    amplification_threshold=amp_thresh,
+                    mdxc_params={
+                        "segment_size": segment_size,
+                        "override_model_segment_size": override_seg_size,
+                        "batch_size": batch_size,
+                        "overlap": overlap,
+                    }
+                )
+
+                logs.append("Loading model...")
+                yield "\n".join(logs)
+                separator.load_model(model_filename=roformer_model)
+
+                logs.append(f"Separating file: {audio_files}")
+                yield "\n".join(logs)
+                separator.separate(file_path, f"{base_name}_(Stem1)", f"{base_name}_(Stem2)")
+                logs.append(f"File: {audio_files} separated!")
+                yield "\n".join(logs)
+            except Exception as e:
+                raise RuntimeError(f"Roformer batch separation failed: {e}") from e
+
+@spaces.GPU(duration=60)
+def mdx23c_batch(path_input, path_output, model, out_format, segment_size, override_seg_size, overlap, batch_size, norm_thresh, amp_thresh):
+    found_files.clear()
+    logs.clear()
+
+    for audio_files in os.listdir(path_input):
+        if audio_files.endswith(extensions):
+            found_files.append(audio_files)
+    total_files = len(found_files)
+
+    if total_files == 0:
+        logs.append("No valid audio files.")
+        yield "\n".join(logs)
+    else:
+        logs.append(f"{total_files} audio files found")
+        found_files.sort()
+
+        for audio_files in found_files:
+            file_path = os.path.join(path_input, audio_files)
+            base_name = os.path.splitext(os.path.basename(file_path))[0]
+            try:
+                separator = Separator(
+                    log_level=logging.WARNING,
+                    model_file_dir=models_dir,
+                    output_dir=path_output,
+                    output_format=out_format,
+                    use_autocast=use_autocast,
+                    normalization_threshold=norm_thresh,
+                    amplification_threshold=amp_thresh,
+                    mdxc_params={
+                        "segment_size": segment_size,
+                        "override_model_segment_size": override_seg_size,
+                        "batch_size": batch_size,
+                        "overlap": overlap,
+                    }
+                )
+
+                logs.append("Loading model...")
+                yield "\n".join(logs)
+                separator.load_model(model_filename=model)
+
+                logs.append(f"Separating file: {audio_files}")
+                yield "\n".join(logs)
+                separator.separate(file_path, f"{base_name}_(Stem1)", f"{base_name}_(Stem2)")
+                logs.append(f"File: {audio_files} separated!")
+                yield "\n".join(logs)
+            except Exception as e:
+                raise RuntimeError(f"Roformer batch separation failed: {e}") from e
+
+@spaces.GPU(duration=60)
+def mdxnet_batch(path_input, path_output, model, out_format, hop_length, segment_size, denoise, overlap, batch_size, norm_thresh, amp_thresh):
+    found_files.clear()
+    logs.clear()
+
+    for audio_files in os.listdir(path_input):
+        if audio_files.endswith(extensions):
+            found_files.append(audio_files)
+    total_files = len(found_files)
+
+    if total_files == 0:
+        logs.append("No valid audio files.")
+        yield "\n".join(logs)
+    else:
+        logs.append(f"{total_files} audio files found")
+        found_files.sort()
+
+        for audio_files in found_files:
+            file_path = os.path.join(path_input, audio_files)
+            base_name = os.path.splitext(os.path.basename(file_path))[0]
+            try:
+                separator = Separator(
+                    log_level=logging.WARNING,
+                    model_file_dir=models_dir,
+                    output_dir=path_output,
+                    output_format=out_format,
+                    use_autocast=use_autocast,
+                    normalization_threshold=norm_thresh,
+                    amplification_threshold=amp_thresh,
+                    mdx_params={
+                        "hop_length": hop_length,
+                        "segment_size": segment_size,
+                        "overlap": overlap,
+                        "batch_size": batch_size,
+                        "enable_denoise": denoise,
+                    }
+                )
+
+                logs.append("Loading model...")
+                yield "\n".join(logs)
+                separator.load_model(model_filename=model)
+
+                logs.append(f"Separating file: {audio_files}")
+                yield "\n".join(logs)
+                separator.separate(file_path, f"{base_name}_(Stem1)", f"{base_name}_(Stem2)")
+                logs.append(f"File: {audio_files} separated!")
+                yield "\n".join(logs)
+            except Exception as e:
+                raise RuntimeError(f"Roformer batch separation failed: {e}") from e
+
+@spaces.GPU(duration=60)
+def vrarch_batch(path_input, path_output, model, out_format, window_size, aggression, tta, post_process, post_process_threshold, high_end_process, batch_size, norm_thresh, amp_thresh):
+    found_files.clear()
+    logs.clear()
+
+    for audio_files in os.listdir(path_input):
+        if audio_files.endswith(extensions):
+            found_files.append(audio_files)
+    total_files = len(found_files)
+
+    if total_files == 0:
+        logs.append("No valid audio files.")
+        yield "\n".join(logs)
+    else:
+        logs.append(f"{total_files} audio files found")
+        found_files.sort()
+
+        for audio_files in found_files:
+            file_path = os.path.join(path_input, audio_files)
+            base_name = os.path.splitext(os.path.basename(file_path))[0]
+            try:
+                separator = Separator(
+                    log_level=logging.WARNING,
+                    model_file_dir=models_dir,
+                    output_dir=path_output,
+                    output_format=out_format,
+                    use_autocast=use_autocast,
+                    normalization_threshold=norm_thresh,
+                    amplification_threshold=amp_thresh,
+                    vr_params={
+                        "batch_size": batch_size,
+                        "window_size": window_size,
+                        "aggression": aggression,
+                        "enable_tta": tta,
+                        "enable_post_process": post_process,
+                        "post_process_threshold": post_process_threshold,
+                        "high_end_process": high_end_process,
+                    }
+                )
+
+                logs.append("Loading model...")
+                yield "\n".join(logs)
+                separator.load_model(model_filename=model)
+
+                logs.append(f"Separating file: {audio_files}")
+                yield "\n".join(logs)
+                separator.separate(file_path, f"{base_name}_(Stem1)", f"{base_name}_(Stem2)")
+                logs.append(f"File: {audio_files} separated!")
+                yield "\n".join(logs)
+            except Exception as e:
+                raise RuntimeError(f"Roformer batch separation failed: {e}") from e
+
+@spaces.GPU(duration=60)
+def demucs_batch(path_input, path_output, model, out_format, shifts, segment_size, segments_enabled, overlap, batch_size, norm_thresh, amp_thresh):
+    found_files.clear()
+    logs.clear()
+
+    for audio_files in os.listdir(path_input):
+        if audio_files.endswith(extensions):
+            found_files.append(audio_files)
+    total_files = len(found_files)
+
+    if total_files == 0:
+        logs.append("No valid audio files.")
+        yield "\n".join(logs)
+    else:
+        logs.append(f"{total_files} audio files found")
+        found_files.sort()
+
+        for audio_files in found_files:
+            file_path = os.path.join(path_input, audio_files)
+            try:
+                separator = Separator(
+                    log_level=logging.WARNING,
+                    model_file_dir=models_dir,
+                    output_dir=path_output,
+                    output_format=out_format,
+                    use_autocast=use_autocast,
+                    normalization_threshold=norm_thresh,
+                    amplification_threshold=amp_thresh,
+                    demucs_params={
+                        "batch_size": batch_size,
+                        "segment_size": segment_size,
+                        "shifts": shifts,
+                        "overlap": overlap,
+                        "segments_enabled": segments_enabled,
+                    }
+                )
+
+                logs.append("Loading model...")
+                yield "\n".join(logs)
+                separator.load_model(model_filename=model)
+
+                logs.append(f"Separating file: {audio_files}")
+                yield "\n".join(logs)
+                separator.separate(file_path)
+                logs.append(f"File: {audio_files} separated!")
+                yield "\n".join(logs)
+            except Exception as e:
+                raise RuntimeError(f"Roformer batch separation failed: {e}") from e
+