Upload 4 files

app.py

@@ -0,0 +1,516 @@
+import os
+import glob
+import json
+import traceback
+import logging
+import gradio as gr
+import numpy as np
+import librosa
+import torch
+import asyncio
+import edge_tts
+import yt_dlp
+import ffmpeg
+import subprocess
+import sys
+import io
+import wave
+from datetime import datetime
+from fairseq import checkpoint_utils
+from lib.infer_pack.models import (
+    SynthesizerTrnMs256NSFsid,
+    SynthesizerTrnMs256NSFsid_nono,
+    SynthesizerTrnMs768NSFsid,
+    SynthesizerTrnMs768NSFsid_nono,
+)
+from vc_infer_pipeline import VC
+from config import Config
+config = Config()
+logging.getLogger("numba").setLevel(logging.WARNING)
+limitation = os.getenv("SYSTEM") == "spaces"
+
+audio_mode = []
+f0method_mode = []
+f0method_info = ""
+if limitation is True:
+    audio_mode = ["Upload audio", "TTS Audio"]
+    f0method_mode = ["pm", "harvest"]
+    f0method_info = "PM is fast, Harvest is good but extremely slow. (Default: PM)"
+else:
+    audio_mode = ["Input path", "Upload audio", "Youtube", "TTS Audio"]
+    f0method_mode = ["pm", "harvest", "crepe"]
+    f0method_info = "PM is fast, Harvest is good but extremely slow, and Crepe effect is good but requires GPU (Default: PM)"
+
+def create_vc_fn(model_name, tgt_sr, net_g, vc, if_f0, version, file_index):
+    def vc_fn(
+        vc_audio_mode,
+        vc_input, 
+        vc_upload,
+        tts_text,
+        tts_voice,
+        f0_up_key,
+        f0_method,
+        index_rate,
+        filter_radius,
+        resample_sr,
+        rms_mix_rate,
+        protect,
+    ):
+        try:
+            print(f"Converting using {model_name}...")
+            if vc_audio_mode == "Input path" or "Youtube" and vc_input != "":
+                audio, sr = librosa.load(vc_input, sr=16000, mono=True)
+            elif vc_audio_mode == "Upload audio":
+                if vc_upload is None:
+                    return "You need to upload an audio", None
+                sampling_rate, audio = vc_upload
+                duration = audio.shape[0] / sampling_rate
+                if duration > 20 and limitation:
+                    return "Please upload an audio file that is less than 20 seconds. If you need to generate a longer audio file, please use Colab.", None
+                audio = (audio / np.iinfo(audio.dtype).max).astype(np.float32)
+                if len(audio.shape) > 1:
+                    audio = librosa.to_mono(audio.transpose(1, 0))
+                if sampling_rate != 16000:
+                    audio = librosa.resample(audio, orig_sr=sampling_rate, target_sr=16000)
+            elif vc_audio_mode == "TTS Audio":
+                if len(tts_text) > 100 and limitation:
+                    return "Text is too long", None
+                if tts_text is None or tts_voice is None:
+                    return "You need to enter text and select a voice", None
+                asyncio.run(edge_tts.Communicate(tts_text, "-".join(tts_voice.split('-')[:-1])).save("tts.mp3"))
+                audio, sr = librosa.load("tts.mp3", sr=16000, mono=True)
+                vc_input = "tts.mp3"
+            times = [0, 0, 0]
+            f0_up_key = int(f0_up_key)
+            audio_opt = vc.pipeline(
+                hubert_model,
+                net_g,
+                0,
+                audio,
+                vc_input,
+                times,
+                f0_up_key,
+                f0_method,
+                file_index,
+                # file_big_npy,
+                index_rate,
+                if_f0,
+                filter_radius,
+                tgt_sr,
+                resample_sr,
+                rms_mix_rate,
+                version,
+                protect,
+                f0_file=None,
+            )
+            info = f"[{datetime.now().strftime('%Y-%m-%d %H:%M')}]: npy: {times[0]}, f0: {times[1]}s, infer: {times[2]}s"
+            print(f"{model_name} | {info}")
+            return info, (tgt_sr, audio_opt)
+        except:
+            info = traceback.format_exc()
+            print(info)
+            return info, None
+    return vc_fn
+
+def load_model():
+    categories = []
+    with open("weights/folder_info.json", "r", encoding="utf-8") as f:
+        folder_info = json.load(f)
+    for category_name, category_info in folder_info.items():
+        if not category_info['enable']:
+            continue
+        category_title = category_info['title']
+        category_folder = category_info['folder_path']
+        description = category_info['description']
+        models = []
+        with open(f"weights/{category_folder}/model_info.json", "r", encoding="utf-8") as f:
+            models_info = json.load(f)
+        for character_name, info in models_info.items():
+            if not info['enable']:
+                continue
+            model_title = info['title']
+            model_name = info['model_path']
+            model_author = info.get("author", None)
+            model_cover = f"weights/{category_folder}/{character_name}/{info['cover']}"
+            model_index = f"weights/{category_folder}/{character_name}/{info['feature_retrieval_library']}"
+            cpt = torch.load(f"weights/{category_folder}/{character_name}/{model_name}", map_location="cpu")
+            tgt_sr = cpt["config"][-1]
+            cpt["config"][-3] = cpt["weight"]["emb_g.weight"].shape[0]  # n_spk
+            if_f0 = cpt.get("f0", 1)
+            version = cpt.get("version", "v1")
+            if version == "v1":
+                if if_f0 == 1:
+                    net_g = SynthesizerTrnMs256NSFsid(*cpt["config"], is_half=config.is_half)
+                else:
+                    net_g = SynthesizerTrnMs256NSFsid_nono(*cpt["config"])
+                model_version = "V1"
+            elif version == "v2":
+                if if_f0 == 1:
+                    net_g = SynthesizerTrnMs768NSFsid(*cpt["config"], is_half=config.is_half)
+                else:
+                    net_g = SynthesizerTrnMs768NSFsid_nono(*cpt["config"])
+                model_version = "V2"
+            del net_g.enc_q
+            print(net_g.load_state_dict(cpt["weight"], strict=False))
+            net_g.eval().to(config.device)
+            if config.is_half:
+                net_g = net_g.half()
+            else:
+                net_g = net_g.float()
+            vc = VC(tgt_sr, config)
+            print(f"Model loaded: {character_name} / {info['feature_retrieval_library']} | ({model_version})")
+            models.append((character_name, model_title, model_author, model_cover, model_version, create_vc_fn(model_name, tgt_sr, net_g, vc, if_f0, version, model_index)))
+        categories.append([category_title, category_folder, description, models])
+    return categories
+
+def cut_vocal_and_inst(url, audio_provider, split_model):
+    if url != "":
+        if not os.path.exists("dl_audio"):
+            os.mkdir("dl_audio")
+        if audio_provider == "Youtube":
+            ydl_opts = {
+                'noplaylist': True,
+                'format': 'bestaudio/best',
+                'postprocessors': [{
+                    'key': 'FFmpegExtractAudio',
+                    'preferredcodec': 'wav',
+                }],
+                "outtmpl": 'dl_audio/youtube_audio',
+            }
+            with yt_dlp.YoutubeDL(ydl_opts) as ydl:
+                ydl.download([url])
+            audio_path = "dl_audio/youtube_audio.wav"
+        if split_model == "htdemucs":
+            command = f"demucs --two-stems=vocals {audio_path} -o output"
+            result = subprocess.run(command.split(), stdout=subprocess.PIPE)
+            print(result.stdout.decode())
+            return "output/htdemucs/youtube_audio/vocals.wav", "output/htdemucs/youtube_audio/no_vocals.wav", audio_path, "output/htdemucs/youtube_audio/vocals.wav"
+        else:
+            command = f"demucs --two-stems=vocals -n mdx_extra_q {audio_path} -o output"
+            result = subprocess.run(command.split(), stdout=subprocess.PIPE)
+            print(result.stdout.decode())
+            return "output/mdx_extra_q/youtube_audio/vocals.wav", "output/mdx_extra_q/youtube_audio/no_vocals.wav", audio_path, "output/mdx_extra_q/youtube_audio/vocals.wav"
+    else:
+        raise gr.Error("URL Required!")
+        return None, None, None, None
+
+def combine_vocal_and_inst(audio_data, audio_volume, split_model):
+    if not os.path.exists("output/result"):
+        os.mkdir("output/result")
+    vocal_path = "output/result/output.wav"
+    output_path = "output/result/combine.mp3"
+    if split_model == "htdemucs":
+        inst_path = "output/htdemucs/youtube_audio/no_vocals.wav"
+    else:
+        inst_path = "output/mdx_extra_q/youtube_audio/no_vocals.wav"
+    with wave.open(vocal_path, "w") as wave_file:
+        wave_file.setnchannels(1) 
+        wave_file.setsampwidth(2)
+        wave_file.setframerate(audio_data[0])
+        wave_file.writeframes(audio_data[1].tobytes())
+    command =  f'ffmpeg -y -i {inst_path} -i {vocal_path} -filter_complex [1:a]volume={audio_volume}dB[v];[0:a][v]amix=inputs=2:duration=longest -b:a 320k -c:a libmp3lame {output_path}'
+    result = subprocess.run(command.split(), stdout=subprocess.PIPE)
+    print(result.stdout.decode())
+    return output_path
+
+def load_hubert():
+    global hubert_model
+    models, _, _ = checkpoint_utils.load_model_ensemble_and_task(
+        ["hubert_base.pt"],
+        suffix="",
+    )
+    hubert_model = models[0]
+    hubert_model = hubert_model.to(config.device)
+    if config.is_half:
+        hubert_model = hubert_model.half()
+    else:
+        hubert_model = hubert_model.float()
+    hubert_model.eval()
+
+def change_audio_mode(vc_audio_mode):
+    if vc_audio_mode == "Input path":
+        return (
+            # Input & Upload
+            gr.Textbox.update(visible=True),
+            gr.Checkbox.update(visible=False),
+            gr.Audio.update(visible=False),
+            # Youtube
+            gr.Dropdown.update(visible=False),
+            gr.Textbox.update(visible=False),
+            gr.Dropdown.update(visible=False),
+            gr.Button.update(visible=False),
+            gr.Audio.update(visible=False),
+            gr.Audio.update(visible=False),
+            gr.Audio.update(visible=False),
+            gr.Slider.update(visible=False),
+            gr.Audio.update(visible=False),
+            gr.Button.update(visible=False),
+            # TTS
+            gr.Textbox.update(visible=False),
+            gr.Dropdown.update(visible=False)
+        )
+    elif vc_audio_mode == "Upload audio":
+        return (
+            # Input & Upload
+            gr.Textbox.update(visible=False),
+            gr.Checkbox.update(visible=True),
+            gr.Audio.update(visible=True),
+            # Youtube
+            gr.Dropdown.update(visible=False),
+            gr.Textbox.update(visible=False),
+            gr.Dropdown.update(visible=False),
+            gr.Button.update(visible=False),
+            gr.Audio.update(visible=False),
+            gr.Audio.update(visible=False),
+            gr.Audio.update(visible=False),
+            gr.Slider.update(visible=False),
+            gr.Audio.update(visible=False),
+            gr.Button.update(visible=False),
+            # TTS
+            gr.Textbox.update(visible=False),
+            gr.Dropdown.update(visible=False)
+        )
+    elif vc_audio_mode == "Youtube":
+        return (
+            # Input & Upload
+            gr.Textbox.update(visible=False),
+            gr.Checkbox.update(visible=False),
+            gr.Audio.update(visible=False),
+            # Youtube
+            gr.Dropdown.update(visible=True),
+            gr.Textbox.update(visible=True),
+            gr.Dropdown.update(visible=True),
+            gr.Button.update(visible=True),
+            gr.Audio.update(visible=True),
+            gr.Audio.update(visible=True),
+            gr.Audio.update(visible=True),
+            gr.Slider.update(visible=True),
+            gr.Audio.update(visible=True),
+            gr.Button.update(visible=True),
+            # TTS
+            gr.Textbox.update(visible=False),
+            gr.Dropdown.update(visible=False)
+        )
+    elif vc_audio_mode == "TTS Audio":
+        return (
+            # Input & Upload
+            gr.Textbox.update(visible=False),
+            gr.Checkbox.update(visible=False),
+            gr.Audio.update(visible=False),
+            # Youtube
+            gr.Dropdown.update(visible=False),
+            gr.Textbox.update(visible=False),
+            gr.Dropdown.update(visible=False),
+            gr.Button.update(visible=False),
+            gr.Audio.update(visible=False),
+            gr.Audio.update(visible=False),
+            gr.Audio.update(visible=False),
+            gr.Slider.update(visible=False),
+            gr.Audio.update(visible=False),
+            gr.Button.update(visible=False),
+            # TTS
+            gr.Textbox.update(visible=True),
+            gr.Dropdown.update(visible=True)
+        )
+    else:
+        return (
+            # Input & Upload
+            gr.Textbox.update(visible=False),
+            gr.Checkbox.update(visible=True),
+            gr.Audio.update(visible=True),
+            # Youtube
+            gr.Dropdown.update(visible=False),
+            gr.Textbox.update(visible=False),
+            gr.Dropdown.update(visible=False),
+            gr.Button.update(visible=False),
+            gr.Audio.update(visible=False),
+            gr.Audio.update(visible=False),
+            gr.Audio.update(visible=False),
+            gr.Slider.update(visible=False),
+            gr.Audio.update(visible=False),
+            gr.Button.update(visible=False),
+            # TTS
+            gr.Textbox.update(visible=False),
+            gr.Dropdown.update(visible=False)
+        )
+
+def use_microphone(microphone):
+    if microphone == True:
+        return gr.Audio.update(source="microphone")
+    else:
+        return gr.Audio.update(source="upload")
+
+if __name__ == '__main__':
+    load_hubert()
+    categories = load_model()
+    tts_voice_list = asyncio.get_event_loop().run_until_complete(edge_tts.list_voices())
+    voices = [f"{v['ShortName']}-{v['Gender']}" for v in tts_voice_list]
+    with gr.Blocks() as app:
+        gr.Markdown(
+            "<div align='center'>\n\n"+
+            "# Multi Model RVC Inference\n\n"+
+            "[![Repository](https://img.shields.io/badge/Github-Multi%20Model%20RVC%20Inference-blue?style=for-the-badge&logo=github)](https://github.com/ArkanDash/Multi-Model-RVC-Inference)\n\n"+
+            "</div>"
+        )
+        for (folder_title, folder, description, models) in categories:
+            with gr.TabItem(folder_title):
+                if description:
+                    gr.Markdown(f"### <center> {description}")
+                with gr.Tabs():
+                    if not models:
+                        gr.Markdown("# <center> No Model Loaded.")
+                        gr.Markdown("## <center> Please add model or fix your model path.")
+                        continue
+                    for (name, title, author, cover, model_version, vc_fn) in models:
+                        with gr.TabItem(name):
+                            with gr.Row():
+                                gr.Markdown(
+                                    '<div align="center">'
+                                    f'<div>{title}</div>\n'+
+                                    f'<div>RVC {model_version} Model</div>\n'+
+                                    (f'<div>Model author: {author}</div>' if author else "")+
+                                    (f'<img style="width:auto;height:300px;" src="file/{cover}">' if cover else "")+
+                                    '</div>'
+                                )
+                            with gr.Row():
+                                with gr.Column():
+                                    vc_audio_mode = gr.Dropdown(label="Input voice", choices=audio_mode, allow_custom_value=False, value="Upload audio")
+                                    # Input
+                                    vc_input = gr.Textbox(label="Input audio path", visible=False)
+                                    # Upload
+                                    vc_microphone_mode = gr.Checkbox(label="Use Microphone", value=False, visible=True, interactive=True)
+                                    vc_upload = gr.Audio(label="Upload audio file", source="upload", visible=True, interactive=True)
+                                    # Youtube
+                                    vc_download_audio = gr.Dropdown(label="Provider", choices=["Youtube"], allow_custom_value=False, visible=False, value="Youtube", info="Select provider (Default: Youtube)")
+                                    vc_link = gr.Textbox(label="Youtube URL", visible=False, info="Example: https://www.youtube.com/watch?v=Nc0sB1Bmf-A", placeholder="https://www.youtube.com/watch?v=...")
+                                    vc_split_model = gr.Dropdown(label="Splitter Model", choices=["htdemucs", "mdx_extra_q"], allow_custom_value=False, visible=False, value="htdemucs", info="Select the splitter model (Default: htdemucs)")
+                                    vc_split = gr.Button("Split Audio", variant="primary", visible=False)
+                                    vc_vocal_preview = gr.Audio(label="Vocal Preview", visible=False)
+                                    vc_inst_preview = gr.Audio(label="Instrumental Preview", visible=False)
+                                    vc_audio_preview = gr.Audio(label="Audio Preview", visible=False)
+                                    # TTS
+                                    tts_text = gr.Textbox(visible=False, label="TTS text", info="Text to speech input")
+                                    tts_voice = gr.Dropdown(label="Edge-tts speaker", choices=voices, visible=False, allow_custom_value=False, value="en-US-AnaNeural-Female")
+                                with gr.Column():
+                                    vc_transform0 = gr.Number(label="Transpose", value=0, info='Type "12" to change from male to female voice. Type "-12" to change female to male voice')
+                                    f0method0 = gr.Radio(
+                                        label="Pitch extraction algorithm",
+                                        info=f0method_info,
+                                        choices=f0method_mode,
+                                        value="pm",
+                                        interactive=True
+                                    )
+                                    index_rate1 = gr.Slider(
+                                        minimum=0,
+                                        maximum=1,
+                                        label="Retrieval feature ratio",
+                                        info="(Default: 0.7)",
+                                        value=0.7,
+                                        interactive=True,
+                                    )
+                                    filter_radius0 = gr.Slider(
+                                        minimum=0,
+                                        maximum=7,
+                                        label="Apply Median Filtering",
+                                        info="The value represents the filter radius and can reduce breathiness.",
+                                        value=3,
+                                        step=1,
+                                        interactive=True,
+                                    )
+                                    resample_sr0 = gr.Slider(
+                                        minimum=0,
+                                        maximum=48000,
+                                        label="Resample the output audio",
+                                        info="Resample the output audio in post-processing to the final sample rate. Set to 0 for no resampling",
+                                        value=0,
+                                        step=1,
+                                        interactive=True,
+                                    )
+                                    rms_mix_rate0 = gr.Slider(
+                                        minimum=0,
+                                        maximum=1,
+                                        label="Volume Envelope",
+                                        info="Use the volume envelope of the input to replace or mix with the volume envelope of the output. The closer the ratio is to 1, the more the output envelope is used",
+                                        value=1,
+                                        interactive=True,
+                                    )
+                                    protect0 = gr.Slider(
+                                        minimum=0,
+                                        maximum=0.5,
+                                        label="Voice Protection",
+                                        info="Protect voiceless consonants and breath sounds to prevent artifacts such as tearing in electronic music. Set to 0.5 to disable. Decrease the value to increase protection, but it may reduce indexing accuracy",
+                                        value=0.5,
+                                        step=0.01,
+                                        interactive=True,
+                                    )
+                                with gr.Column():
+                                    vc_log = gr.Textbox(label="Output Information", interactive=False)
+                                    vc_output = gr.Audio(label="Output Audio", interactive=False)
+                                    vc_convert = gr.Button("Convert", variant="primary")
+                                    vc_volume = gr.Slider(
+                                        minimum=0,
+                                        maximum=10,
+                                        label="Vocal volume",
+                                        value=4,
+                                        interactive=True,
+                                        step=1,
+                                        info="Adjust vocal volume (Default: 4}",
+                                        visible=False
+                                    )
+                                    vc_combined_output = gr.Audio(label="Output Combined Audio", visible=False)
+                                    vc_combine =  gr.Button("Combine",variant="primary", visible=False)
+                        vc_convert.click(
+                            fn=vc_fn, 
+                            inputs=[
+                                vc_audio_mode,
+                                vc_input,
+                                vc_upload,
+                                tts_text,
+                                tts_voice,
+                                vc_transform0,
+                                f0method0,
+                                index_rate1,
+                                filter_radius0,
+                                resample_sr0,
+                                rms_mix_rate0,
+                                protect0,
+                            ], 
+                            outputs=[vc_log ,vc_output]
+                        )
+                        vc_split.click(
+                            fn=cut_vocal_and_inst, 
+                            inputs=[vc_link, vc_download_audio, vc_split_model], 
+                            outputs=[vc_vocal_preview, vc_inst_preview, vc_audio_preview, vc_input]
+                        )
+                        vc_combine.click(
+                            fn=combine_vocal_and_inst,
+                            inputs=[vc_output, vc_volume, vc_split_model],
+                            outputs=[vc_combined_output]
+                        )
+                        vc_microphone_mode.change(
+                            fn=use_microphone,
+                            inputs=vc_microphone_mode,
+                            outputs=vc_upload
+                        )
+                        vc_audio_mode.change(
+                            fn=change_audio_mode,
+                            inputs=[vc_audio_mode],
+                            outputs=[
+                                vc_input,
+                                vc_microphone_mode,
+                                vc_upload,
+                                vc_download_audio,
+                                vc_link,
+                                vc_split_model,
+                                vc_split,
+                                vc_vocal_preview,
+                                vc_inst_preview,
+                                vc_audio_preview,
+                                vc_volume,
+                                vc_combined_output,
+                                vc_combine,
+                                tts_text,
+                                tts_voice
+                            ]
+                        )
+        app.queue(concurrency_count=1, max_size=20, api_open=config.api).launch(share=config.colab)

config.py

@@ -0,0 +1,117 @@
+import argparse
+import sys
+import torch
+from multiprocessing import cpu_count
+
+class Config:
+    def __init__(self):
+        self.device = "cuda:0"
+        self.is_half = True
+        self.n_cpu = 0
+        self.gpu_name = None
+        self.gpu_mem = None
+        (
+            self.python_cmd,
+            self.listen_port,
+            self.colab,
+            self.noparallel,
+            self.noautoopen,
+            self.api
+        ) = self.arg_parse()
+        self.x_pad, self.x_query, self.x_center, self.x_max = self.device_config()
+
+    @staticmethod
+    def arg_parse() -> tuple:
+        exe = sys.executable or "python"
+        parser = argparse.ArgumentParser()
+        parser.add_argument("--port", type=int, default=7865, help="Listen port")
+        parser.add_argument("--pycmd", type=str, default=exe, help="Python command")
+        parser.add_argument("--colab", action="store_true", help="Launch in colab")
+        parser.add_argument(
+            "--noparallel", action="store_true", help="Disable parallel processing"
+        )
+        parser.add_argument(
+            "--noautoopen",
+            action="store_true",
+            help="Do not open in browser automatically",
+        )
+        parser.add_argument("--api", action="store_true", help="Launch with api")
+        cmd_opts = parser.parse_args()
+
+        cmd_opts.port = cmd_opts.port if 0 <= cmd_opts.port <= 65535 else 7865
+
+        return (
+            cmd_opts.pycmd,
+            cmd_opts.port,
+            cmd_opts.colab,
+            cmd_opts.noparallel,
+            cmd_opts.noautoopen,
+            cmd_opts.api
+        )
+
+    # has_mps is only available in nightly pytorch (for now) and MasOS 12.3+.
+    # check `getattr` and try it for compatibility
+    @staticmethod
+    def has_mps() -> bool:
+        if not torch.backends.mps.is_available():
+            return False
+        try:
+            torch.zeros(1).to(torch.device("mps"))
+            return True
+        except Exception:
+            return False
+
+    def device_config(self) -> tuple:
+        if torch.cuda.is_available():
+            i_device = int(self.device.split(":")[-1])
+            self.gpu_name = torch.cuda.get_device_name(i_device)
+            if (
+                ("16" in self.gpu_name and "V100" not in self.gpu_name.upper())
+                or "P40" in self.gpu_name.upper()
+                or "1060" in self.gpu_name
+                or "1070" in self.gpu_name
+                or "1080" in self.gpu_name
+            ):
+                print("Found GPU", self.gpu_name, ", force to fp32")
+                self.is_half = False
+            else:
+                print("Found GPU", self.gpu_name)
+            self.gpu_mem = int(
+                torch.cuda.get_device_properties(i_device).total_memory
+                / 1024
+                / 1024
+                / 1024
+                + 0.4
+            )
+        elif self.has_mps():
+            print("No supported Nvidia GPU found, use MPS instead")
+            self.device = "mps"
+            self.is_half = False
+        else:
+            print("No supported Nvidia GPU found, use CPU instead")
+            self.device = "cpu"
+            self.is_half = False
+
+        if self.n_cpu == 0:
+            self.n_cpu = cpu_count()
+
+        if self.is_half:
+            # 6G显存配置
+            x_pad = 3
+            x_query = 10
+            x_center = 60
+            x_max = 65
+        else:
+            # 5G显存配置
+            x_pad = 1
+            x_query = 6
+            x_center = 38
+            x_max = 41
+
+        if self.gpu_mem != None and self.gpu_mem <= 4:
+            x_pad = 1
+            x_query = 5
+            x_center = 30
+            x_max = 32
+
+        return x_pad, x_query, x_center, x_max

requirements.txt

@@ -0,0 +1,21 @@
+wheel
+setuptools
+ffmpeg
+numba==0.56.4
+numpy==1.23.5
+scipy==1.9.3
+librosa==0.9.1
+fairseq==0.12.2
+faiss-cpu==1.7.3
+gradio==3.36.1
+pyworld==0.3.2
+soundfile>=0.12.1
+praat-parselmouth>=0.4.2
+httpx==0.23.0
+tensorboard
+tensorboardX
+torchcrepe
+onnxruntime
+demucs
+edge-tts
+yt_dlp

vc_infer_pipeline.py

@@ -0,0 +1,443 @@
+import numpy as np, parselmouth, torch, pdb, sys, os
+from time import time as ttime
+import torch.nn.functional as F
+import scipy.signal as signal
+import pyworld, os, traceback, faiss, librosa, torchcrepe
+from scipy import signal
+from functools import lru_cache
+
+now_dir = os.getcwd()
+sys.path.append(now_dir)
+
+bh, ah = signal.butter(N=5, Wn=48, btype="high", fs=16000)
+
+input_audio_path2wav = {}
+
+
+@lru_cache
+def cache_harvest_f0(input_audio_path, fs, f0max, f0min, frame_period):
+    audio = input_audio_path2wav[input_audio_path]
+    f0, t = pyworld.harvest(
+        audio,
+        fs=fs,
+        f0_ceil=f0max,
+        f0_floor=f0min,
+        frame_period=frame_period,
+    )
+    f0 = pyworld.stonemask(audio, f0, t, fs)
+    return f0
+
+
+def change_rms(data1, sr1, data2, sr2, rate):  # 1是输入音频，2是输出音频,rate是2的占比
+    # print(data1.max(),data2.max())
+    rms1 = librosa.feature.rms(
+        y=data1, frame_length=sr1 // 2 * 2, hop_length=sr1 // 2
+    )  # 每半秒一个点
+    rms2 = librosa.feature.rms(y=data2, frame_length=sr2 // 2 * 2, hop_length=sr2 // 2)
+    rms1 = torch.from_numpy(rms1)
+    rms1 = F.interpolate(
+        rms1.unsqueeze(0), size=data2.shape[0], mode="linear"
+    ).squeeze()
+    rms2 = torch.from_numpy(rms2)
+    rms2 = F.interpolate(
+        rms2.unsqueeze(0), size=data2.shape[0], mode="linear"
+    ).squeeze()
+    rms2 = torch.max(rms2, torch.zeros_like(rms2) + 1e-6)
+    data2 *= (
+        torch.pow(rms1, torch.tensor(1 - rate))
+        * torch.pow(rms2, torch.tensor(rate - 1))
+    ).numpy()
+    return data2
+
+
+class VC(object):
+    def __init__(self, tgt_sr, config):
+        self.x_pad, self.x_query, self.x_center, self.x_max, self.is_half = (
+            config.x_pad,
+            config.x_query,
+            config.x_center,
+            config.x_max,
+            config.is_half,
+        )
+        self.sr = 16000  # hubert输入采样率
+        self.window = 160  # 每帧点数
+        self.t_pad = self.sr * self.x_pad  # 每条前后pad时间
+        self.t_pad_tgt = tgt_sr * self.x_pad
+        self.t_pad2 = self.t_pad * 2
+        self.t_query = self.sr * self.x_query  # 查询切点前后查询时间
+        self.t_center = self.sr * self.x_center  # 查询切点位置
+        self.t_max = self.sr * self.x_max  # 免查询时长阈值
+        self.device = config.device
+
+    def get_f0(
+        self,
+        input_audio_path,
+        x,
+        p_len,
+        f0_up_key,
+        f0_method,
+        filter_radius,
+        inp_f0=None,
+    ):
+        global input_audio_path2wav
+        time_step = self.window / self.sr * 1000
+        f0_min = 50
+        f0_max = 1100
+        f0_mel_min = 1127 * np.log(1 + f0_min / 700)
+        f0_mel_max = 1127 * np.log(1 + f0_max / 700)
+        if f0_method == "pm":
+            f0 = (
+                parselmouth.Sound(x, self.sr)
+                .to_pitch_ac(
+                    time_step=time_step / 1000,
+                    voicing_threshold=0.6,
+                    pitch_floor=f0_min,
+                    pitch_ceiling=f0_max,
+                )
+                .selected_array["frequency"]
+            )
+            pad_size = (p_len - len(f0) + 1) // 2
+            if pad_size > 0 or p_len - len(f0) - pad_size > 0:
+                f0 = np.pad(
+                    f0, [[pad_size, p_len - len(f0) - pad_size]], mode="constant"
+                )
+        elif f0_method == "harvest":
+            input_audio_path2wav[input_audio_path] = x.astype(np.double)
+            f0 = cache_harvest_f0(input_audio_path, self.sr, f0_max, f0_min, 10)
+            if filter_radius > 2:
+                f0 = signal.medfilt(f0, 3)
+        elif f0_method == "crepe":
+            model = "full"
+            # Pick a batch size that doesn't cause memory errors on your gpu
+            batch_size = 512
+            # Compute pitch using first gpu
+            audio = torch.tensor(np.copy(x))[None].float()
+            f0, pd = torchcrepe.predict(
+                audio,
+                self.sr,
+                self.window,
+                f0_min,
+                f0_max,
+                model,
+                batch_size=batch_size,
+                device=self.device,
+                return_periodicity=True,
+            )
+            pd = torchcrepe.filter.median(pd, 3)
+            f0 = torchcrepe.filter.mean(f0, 3)
+            f0[pd < 0.1] = 0
+            f0 = f0[0].cpu().numpy()
+        elif f0_method == "rmvpe":
+            if hasattr(self, "model_rmvpe") == False:
+                from rmvpe import RMVPE
+
+                print("loading rmvpe model")
+                self.model_rmvpe = RMVPE(
+                    "rmvpe.pt", is_half=self.is_half, device=self.device
+                )
+            f0 = self.model_rmvpe.infer_from_audio(x, thred=0.03)
+        f0 *= pow(2, f0_up_key / 12)
+        # with open("test.txt","w")as f:f.write("\n".join([str(i)for i in f0.tolist()]))
+        tf0 = self.sr // self.window  # 每秒f0点数
+        if inp_f0 is not None:
+            delta_t = np.round(
+                (inp_f0[:, 0].max() - inp_f0[:, 0].min()) * tf0 + 1
+            ).astype("int16")
+            replace_f0 = np.interp(
+                list(range(delta_t)), inp_f0[:, 0] * 100, inp_f0[:, 1]
+            )
+            shape = f0[self.x_pad * tf0 : self.x_pad * tf0 + len(replace_f0)].shape[0]
+            f0[self.x_pad * tf0 : self.x_pad * tf0 + len(replace_f0)] = replace_f0[
+                :shape
+            ]
+        # with open("test_opt.txt","w")as f:f.write("\n".join([str(i)for i in f0.tolist()]))
+        f0bak = f0.copy()
+        f0_mel = 1127 * np.log(1 + f0 / 700)
+        f0_mel[f0_mel > 0] = (f0_mel[f0_mel > 0] - f0_mel_min) * 254 / (
+            f0_mel_max - f0_mel_min
+        ) + 1
+        f0_mel[f0_mel <= 1] = 1
+        f0_mel[f0_mel > 255] = 255
+        f0_coarse = np.rint(f0_mel).astype(np.int)
+        return f0_coarse, f0bak  # 1-0
+
+    def vc(
+        self,
+        model,
+        net_g,
+        sid,
+        audio0,
+        pitch,
+        pitchf,
+        times,
+        index,
+        big_npy,
+        index_rate,
+        version,
+        protect,
+    ):  # ,file_index,file_big_npy
+        feats = torch.from_numpy(audio0)
+        if self.is_half:
+            feats = feats.half()
+        else:
+            feats = feats.float()
+        if feats.dim() == 2:  # double channels
+            feats = feats.mean(-1)
+        assert feats.dim() == 1, feats.dim()
+        feats = feats.view(1, -1)
+        padding_mask = torch.BoolTensor(feats.shape).to(self.device).fill_(False)
+
+        inputs = {
+            "source": feats.to(self.device),
+            "padding_mask": padding_mask,
+            "output_layer": 9 if version == "v1" else 12,
+        }
+        t0 = ttime()
+        with torch.no_grad():
+            logits = model.extract_features(**inputs)
+            feats = model.final_proj(logits[0]) if version == "v1" else logits[0]
+        if protect < 0.5 and pitch != None and pitchf != None:
+            feats0 = feats.clone()
+        if (
+            isinstance(index, type(None)) == False
+            and isinstance(big_npy, type(None)) == False
+            and index_rate != 0
+        ):
+            npy = feats[0].cpu().numpy()
+            if self.is_half:
+                npy = npy.astype("float32")
+
+            # _, I = index.search(npy, 1)
+            # npy = big_npy[I.squeeze()]
+
+            score, ix = index.search(npy, k=8)
+            weight = np.square(1 / score)
+            weight /= weight.sum(axis=1, keepdims=True)
+            npy = np.sum(big_npy[ix] * np.expand_dims(weight, axis=2), axis=1)
+
+            if self.is_half:
+                npy = npy.astype("float16")
+            feats = (
+                torch.from_numpy(npy).unsqueeze(0).to(self.device) * index_rate
+                + (1 - index_rate) * feats
+            )
+
+        feats = F.interpolate(feats.permute(0, 2, 1), scale_factor=2).permute(0, 2, 1)
+        if protect < 0.5 and pitch != None and pitchf != None:
+            feats0 = F.interpolate(feats0.permute(0, 2, 1), scale_factor=2).permute(
+                0, 2, 1
+            )
+        t1 = ttime()
+        p_len = audio0.shape[0] // self.window
+        if feats.shape[1] < p_len:
+            p_len = feats.shape[1]
+            if pitch != None and pitchf != None:
+                pitch = pitch[:, :p_len]
+                pitchf = pitchf[:, :p_len]
+
+        if protect < 0.5 and pitch != None and pitchf != None:
+            pitchff = pitchf.clone()
+            pitchff[pitchf > 0] = 1
+            pitchff[pitchf < 1] = protect
+            pitchff = pitchff.unsqueeze(-1)
+            feats = feats * pitchff + feats0 * (1 - pitchff)
+            feats = feats.to(feats0.dtype)
+        p_len = torch.tensor([p_len], device=self.device).long()
+        with torch.no_grad():
+            if pitch != None and pitchf != None:
+                audio1 = (
+                    (net_g.infer(feats, p_len, pitch, pitchf, sid)[0][0, 0])
+                    .data.cpu()
+                    .float()
+                    .numpy()
+                )
+            else:
+                audio1 = (
+                    (net_g.infer(feats, p_len, sid)[0][0, 0]).data.cpu().float().numpy()
+                )
+        del feats, p_len, padding_mask
+        if torch.cuda.is_available():
+            torch.cuda.empty_cache()
+        t2 = ttime()
+        times[0] += t1 - t0
+        times[2] += t2 - t1
+        return audio1
+
+    def pipeline(
+        self,
+        model,
+        net_g,
+        sid,
+        audio,
+        input_audio_path,
+        times,
+        f0_up_key,
+        f0_method,
+        file_index,
+        # file_big_npy,
+        index_rate,
+        if_f0,
+        filter_radius,
+        tgt_sr,
+        resample_sr,
+        rms_mix_rate,
+        version,
+        protect,
+        f0_file=None,
+    ):
+        if (
+            file_index != ""
+            # and file_big_npy != ""
+            # and os.path.exists(file_big_npy) == True
+            and os.path.exists(file_index) == True
+            and index_rate != 0
+        ):
+            try:
+                index = faiss.read_index(file_index)
+                # big_npy = np.load(file_big_npy)
+                big_npy = index.reconstruct_n(0, index.ntotal)
+            except:
+                traceback.print_exc()
+                index = big_npy = None
+        else:
+            index = big_npy = None
+        audio = signal.filtfilt(bh, ah, audio)
+        audio_pad = np.pad(audio, (self.window // 2, self.window // 2), mode="reflect")
+        opt_ts = []
+        if audio_pad.shape[0] > self.t_max:
+            audio_sum = np.zeros_like(audio)
+            for i in range(self.window):
+                audio_sum += audio_pad[i : i - self.window]
+            for t in range(self.t_center, audio.shape[0], self.t_center):
+                opt_ts.append(
+                    t
+                    - self.t_query
+                    + np.where(
+                        np.abs(audio_sum[t - self.t_query : t + self.t_query])
+                        == np.abs(audio_sum[t - self.t_query : t + self.t_query]).min()
+                    )[0][0]
+                )
+        s = 0
+        audio_opt = []
+        t = None
+        t1 = ttime()
+        audio_pad = np.pad(audio, (self.t_pad, self.t_pad), mode="reflect")
+        p_len = audio_pad.shape[0] // self.window
+        inp_f0 = None
+        if hasattr(f0_file, "name") == True:
+            try:
+                with open(f0_file.name, "r") as f:
+                    lines = f.read().strip("\n").split("\n")
+                inp_f0 = []
+                for line in lines:
+                    inp_f0.append([float(i) for i in line.split(",")])
+                inp_f0 = np.array(inp_f0, dtype="float32")
+            except:
+                traceback.print_exc()
+        sid = torch.tensor(sid, device=self.device).unsqueeze(0).long()
+        pitch, pitchf = None, None
+        if if_f0 == 1:
+            pitch, pitchf = self.get_f0(
+                input_audio_path,
+                audio_pad,
+                p_len,
+                f0_up_key,
+                f0_method,
+                filter_radius,
+                inp_f0,
+            )
+            pitch = pitch[:p_len]
+            pitchf = pitchf[:p_len]
+            if self.device == "mps":
+                pitchf = pitchf.astype(np.float32)
+            pitch = torch.tensor(pitch, device=self.device).unsqueeze(0).long()
+            pitchf = torch.tensor(pitchf, device=self.device).unsqueeze(0).float()
+        t2 = ttime()
+        times[1] += t2 - t1
+        for t in opt_ts:
+            t = t // self.window * self.window
+            if if_f0 == 1:
+                audio_opt.append(
+                    self.vc(
+                        model,
+                        net_g,
+                        sid,
+                        audio_pad[s : t + self.t_pad2 + self.window],
+                        pitch[:, s // self.window : (t + self.t_pad2) // self.window],
+                        pitchf[:, s // self.window : (t + self.t_pad2) // self.window],
+                        times,
+                        index,
+                        big_npy,
+                        index_rate,
+                        version,
+                        protect,
+                    )[self.t_pad_tgt : -self.t_pad_tgt]
+                )
+            else:
+                audio_opt.append(
+                    self.vc(
+                        model,
+                        net_g,
+                        sid,
+                        audio_pad[s : t + self.t_pad2 + self.window],
+                        None,
+                        None,
+                        times,
+                        index,
+                        big_npy,
+                        index_rate,
+                        version,
+                        protect,
+                    )[self.t_pad_tgt : -self.t_pad_tgt]
+                )
+            s = t
+        if if_f0 == 1:
+            audio_opt.append(
+                self.vc(
+                    model,
+                    net_g,
+                    sid,
+                    audio_pad[t:],
+                    pitch[:, t // self.window :] if t is not None else pitch,
+                    pitchf[:, t // self.window :] if t is not None else pitchf,
+                    times,
+                    index,
+                    big_npy,
+                    index_rate,
+                    version,
+                    protect,
+                )[self.t_pad_tgt : -self.t_pad_tgt]
+            )
+        else:
+            audio_opt.append(
+                self.vc(
+                    model,
+                    net_g,
+                    sid,
+                    audio_pad[t:],
+                    None,
+                    None,
+                    times,
+                    index,
+                    big_npy,
+                    index_rate,
+                    version,
+                    protect,
+                )[self.t_pad_tgt : -self.t_pad_tgt]
+            )
+        audio_opt = np.concatenate(audio_opt)
+        if rms_mix_rate != 1:
+            audio_opt = change_rms(audio, 16000, audio_opt, tgt_sr, rms_mix_rate)
+        if resample_sr >= 16000 and tgt_sr != resample_sr:
+            audio_opt = librosa.resample(
+                audio_opt, orig_sr=tgt_sr, target_sr=resample_sr
+            )
+        audio_max = np.abs(audio_opt).max() / 0.99
+        max_int16 = 32768
+        if audio_max > 1:
+            max_int16 /= audio_max
+        audio_opt = (audio_opt * max_int16).astype(np.int16)
+        del pitch, pitchf, sid
+        if torch.cuda.is_available():
+            torch.cuda.empty_cache()
+        return audio_opt