feat: added v2 support

README.md

@@ -4,7 +4,7 @@ emoji: 🎤
 colorFrom: red
 colorTo: purple
 sdk: gradio
-sdk_version: 3.32.0
+sdk_version: 3.34.0
 app_file: app.py
 pinned: true
 license: mit

app.py

@@ -1,7 +1,6 @@
 import os
 import glob
 import json
-import argparse
 import traceback
 import logging
 import gradio as gr
@@ -10,37 +9,48 @@ 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 infer_pack.models import SynthesizerTrnMs256NSFsid, SynthesizerTrnMs256NSFsid_nono
+from 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"  # limit audio length in huggingface spaces
+limitation = os.getenv("SYSTEM") == "spaces"
 
 def create_vc_fn(tgt_sr, net_g, vc, if_f0, file_index):
     def vc_fn(
-        input_audio,
+        vc_input, 
+        vc_upload,
+        tts_text,
+        tts_voice,
         f0_up_key,
+        vc_transform,
         f0_method,
         index_rate,
-        tts_mode,
-        tts_text,
-        tts_voice
+        filter_radius,
+        resample_sr,
+        rms_mix_rate,
+        protect,
     ):
         try:
-            if tts_mode:
-                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)
-            else:
-                if input_audio is None:
+            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 = input_audio
+                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
@@ -49,31 +59,102 @@ def create_vc_fn(tgt_sr, net_g, vc, if_f0, file_index):
                     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)
             times = [0, 0, 0]
             f0_up_key = int(f0_up_key)
             audio_opt = vc.pipeline(
                 hubert_model,
                 net_g,
-                0,
+                vc_transform,
                 audio,
+                vc_input,
                 times,
                 f0_up_key,
                 f0_method,
                 file_index,
                 index_rate,
                 if_f0,
+                filter_radius,
+                tgt_sr,
+                resample_sr,
+                rms_mix_rate,
+                version,
+                protect,
                 f0_file=None,
             )
-            print(
-                f"[{datetime.now().strftime('%Y-%m-%d %H:%M')}]: npy: {times[0]}, f0: {times[1]}s, infer: {times[2]}s"
-            )
-            return (tgt_sr, audio_opt)
+            info = f"[{datetime.now().strftime('%Y-%m-%d %H:%M')}]: npy: {times[0]}, f0: {times[1]}s, infer: {times[2]}s"
+            print(info)
+            return info, (tgt_sr, audio_opt)
         except:
             info = traceback.format_exc()
             print(info)
             return info, (None, None)
     return vc_fn
 
+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 = {
+            '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"
+        else:
+            # Spotify doesnt work.
+            # Need to find other solution soon.
+            ''' 
+            command = f"spotdl download {url} --output dl_audio/.wav"
+            result = subprocess.run(command.split(), stdout=subprocess.PIPE)
+            print(result.stdout.decode())
+            audio_path = "dl_audio/spotify_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(
@@ -88,11 +169,107 @@ def load_hubert():
         hubert_model = hubert_model.float()
     hubert_model.eval()
 
-def change_to_tts_mode(tts_mode):
-    if tts_mode:
-        return gr.Audio.update(visible=False), gr.Textbox.update(visible=True), gr.Dropdown.update(visible=True)
+def change_audio_mode(vc_audio_mode):
+    if vc_audio_mode == "Input path":
+        return (
+            # Input & Upload
+            gr.Textbox.update(visible=True),
+            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.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.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.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 gr.Audio.update(visible=True), gr.Textbox.update(visible=False), gr.Dropdown.update(visible=False)
+        return (
+            # Input & Upload
+            gr.Textbox.update(visible=False),
+            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)
+        )
 
 if __name__ == '__main__':
     load_hubert()
@@ -121,10 +298,19 @@ if __name__ == '__main__':
             tgt_sr = cpt["config"][-1]
             cpt["config"][-3] = cpt["weight"]["emb_g.weight"].shape[0]  # n_spk
             if_f0 = cpt.get("f0", 1)
-            if if_f0 == 1:
-                net_g = SynthesizerTrnMs256NSFsid(*cpt["config"], is_half=config.is_half)
-            else:
-                net_g = SynthesizerTrnMs256NSFsid_nono(*cpt["config"])
+            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"])
+                nodel_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"])
+                nodel_version = "V2"
             del net_g.enc_q
             print(net_g.load_state_dict(cpt["weight"], strict=False))
             net_g.eval().to(config.device)
@@ -134,18 +320,13 @@ if __name__ == '__main__':
                 net_g = net_g.float()
             vc = VC(tgt_sr, config)
             print(f"Model loaded: {model_name}")
-            models.append((model_name, model_title, model_author, model_cover, create_vc_fn(tgt_sr, net_g, vc, if_f0, model_index)))
+            models.append((model_name, model_title, model_author, model_cover, nodel_version, create_vc_fn(tgt_sr, net_g, vc, if_f0, model_index)))
         categories.append([category_title, category_folder, description, models])
     with gr.Blocks() as app:
         gr.Markdown(
-            "# <center> RVC Genshin Impact\n"
-            "## <center> The input audio should be clean and pure voice without background music.\n"
-            "## <center> [Recommended to use google colab to use all genshin model & feature](https://colab.research.google.com/drive/110kiMZTdP6Ri1lY9-NbQf17GVPPhHyeT?usp=sharing)\n"
-            "### <center> I limit the number of models to 15 due to an error caused by exceeding the available memory. (16 GB limit)\n"
-            "### <center> This project was inspired by [zomehwh](https://huggingface.co/spaces/zomehwh/rvc-models) and [ardha27](https://huggingface.co/spaces/ardha27/rvc-models)\n"
-            "[![image](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/110kiMZTdP6Ri1lY9-NbQf17GVPPhHyeT?usp=sharing)\n\n"
-            "[![Original RVC Repo](https://badgen.net/badge/icon/github?icon=github&label=Original%20Repo)](https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI)"
-            "[![RVC Inference Repo](https://badgen.net/badge/icon/github?icon=github&label)](https://github.com/ArkanDash/rvc-inference)"
+            "# <center> RVC Genshin Impact Inference\n"
+            "#### From [Retrieval-based-Voice-Conversion](https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI)\n"
+            "[![Original Repo](https://badgen.net/badge/icon/github?icon=github&label=Original%20Repo)](https://github.com/ArkanDash/Multi-Model-RVC-Inference)"
         )
         for (folder_title, folder, description, models) in categories:
             with gr.TabItem(folder_title):
@@ -154,44 +335,159 @@ if __name__ == '__main__':
                 with gr.Tabs():
                     if not models:
                         gr.Markdown("# <center> No Model Loaded.")
-                        gr.Markdown("## <center> Please added the model or fix your model path.")
+                        gr.Markdown("## <center> Please add model or fix your model path.")
                         continue
-                    with gr.Tabs():
-                        for (name, title, author, cover, vc_fn) in models:
-                            with gr.TabItem(name):
-                                with gr.Row():
-                                    gr.Markdown(
-                                        '<div align="center">'
-                                        f'<div>{title}</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>'
+                    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=["Upload audio", "TTS Audio"], allow_custom_value=False, value="Upload audio")
+                                    # Input and Upload
+                                    vc_input = gr.Textbox(label="Input audio path", visible=False)
+                                    vc_upload = gr.Audio(label="Upload audio file", 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():
+                                    spk_item = gr.Slider(
+                                        minimum=0,
+                                        maximum=2333,
+                                        step=1,
+                                        label="Speaker ID",
+                                        info="(Default: 0)",
+                                        value=0,
+                                        interactive=True,
+                                    )
+                                    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="PM is fast, Harvest is good but extremely slow (Default: PM)",
+                                        choices=["pm", "harvest"],
+                                        value="pm",
+                                        interactive=True,
+                                    )
+                                    index_rate1 = gr.Slider(
+                                        minimum=0,
+                                        maximum=1,
+                                        label="Retrieval feature ratio",
+                                        info="(Default: 0.6)",
+                                        value=0.6,
+                                        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=0.35,
+                                        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
                                     )
-                                with gr.Row():
-                                    with gr.Column():
-                                        vc_input = gr.Audio(label="Input audio"+' (less than 20 seconds)' if limitation else '')
-                                        vc_transpose = gr.Number(label="Transpose", value=0, info='Type "12" to change from male to female voice. Type "-12" to change female to male voice')
-                                        vc_f0method = gr.Radio(
-                                            label="Pitch extraction algorithm",
-                                            choices=["pm", "harvest"],
-                                            value="pm",
-                                            interactive=True,
-                                            info="PM is fast but Harvest is better for low frequencies. (Default: PM)"
-                                        )
-                                        vc_index_ratio = gr.Slider(
-                                            minimum=0,
-                                            maximum=1,
-                                            label="Retrieval feature ratio",
-                                            value=0.6,
-                                            interactive=True,
-                                            info="(Default: 0.6)"
-                                        )
-                                        tts_mode = gr.Checkbox(label="tts (use edge-tts as input)", value=False)
-                                        tts_text = gr.Textbox(visible=False,label="TTS text (100 words limitation)" if limitation else "TTS text")
-                                        tts_voice = gr.Dropdown(label="Edge-tts speaker", choices=voices, visible=False, allow_custom_value=False, value="en-US-AnaNeural-Female")
-                                        vc_submit = gr.Button("Generate", variant="primary")
-                                    with gr.Column():
-                                        vc_output = gr.Audio(label="Output Audio")
-                            vc_submit.click(vc_fn, [vc_input, vc_transpose, vc_f0method, vc_index_ratio, tts_mode, tts_text, tts_voice], [vc_output])
-                            tts_mode.change(change_to_tts_mode, [tts_mode], [vc_input, tts_text, tts_voice])
+                                    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_input, 
+                                vc_upload,
+                                tts_text,
+                                tts_voice,
+                                spk_item,
+                                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_combine.click(
+                            fn=combine_vocal_and_inst,
+                            inputs=[vc_output, vc_volume, vc_split_model],
+                            outputs=[vc_combined_output]
+                        )
+                        vc_audio_mode.change(
+                            fn=change_audio_mode,
+                            inputs=[vc_audio_mode],
+                            outputs=[
+                                vc_input, 
+                                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

@@ -3,6 +3,18 @@ import torch
 from multiprocessing import cpu_count
 
 
+def config_file_change_fp32():
+    for config_file in ["32k.json", "40k.json", "48k.json"]:
+        with open(f"configs/{config_file}", "r") as f:
+            strr = f.read().replace("true", "false")
+        with open(f"configs/{config_file}", "w") as f:
+            f.write(strr)
+    with open("trainset_preprocess_pipeline_print.py", "r") as f:
+        strr = f.read().replace("3.7", "3.0")
+    with open("trainset_preprocess_pipeline_print.py", "w") as f:
+        f.write(strr)
+
+
 class Config:
     def __init__(self):
         self.device = "cuda:0"
@@ -36,7 +48,7 @@ class Config:
             action="store_true",
             help="Do not open in browser automatically",
         )
-        parser.add_argument('--api', action="store_true", default=False)
+        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
@@ -47,7 +59,7 @@ class Config:
             cmd_opts.colab,
             cmd_opts.noparallel,
             cmd_opts.noautoopen,
-            cmd_opts.api,
+            cmd_opts.api
         )
 
     def device_config(self) -> tuple:
@@ -63,15 +75,7 @@ class Config:
             ):
                 print("16系/10系显卡和P40强制单精度")
                 self.is_half = False
-                for config_file in ["32k.json", "40k.json", "48k.json"]:
-                    with open(f"configs/{config_file}", "r") as f:
-                        strr = f.read().replace("true", "false")
-                    with open(f"configs/{config_file}", "w") as f:
-                        f.write(strr)
-                with open("trainset_preprocess_pipeline_print.py", "r") as f:
-                    strr = f.read().replace("3.7", "3.0")
-                with open("trainset_preprocess_pipeline_print.py", "w") as f:
-                    f.write(strr)
+                config_file_change_fp32()
             else:
                 self.gpu_name = None
             self.gpu_mem = int(
@@ -90,10 +94,12 @@ class Config:
             print("没有发现支持的N卡, 使用MPS进行推理")
             self.device = "mps"
             self.is_half = False
+            config_file_change_fp32()
         else:
             print("没有发现支持的N卡, 使用CPU进行推理")
             self.device = "cpu"
             self.is_half = False
+            config_file_change_fp32()
 
         if self.n_cpu == 0:
             self.n_cpu = cpu_count()
@@ -117,4 +123,4 @@ class Config:
             x_center = 30
             x_max = 32
 
-        return x_pad, x_query, x_center, x_max
+        return x_pad, x_query, x_center, x_max

infer_pack/models.py

@@ -61,7 +61,7 @@ class TextEncoder256(nn.Module):
         return m, logs, x_mask
 
 
-class TextEncoder256Sim(nn.Module):
+class TextEncoder768(nn.Module):
     def __init__(
         self,
         out_channels,
@@ -81,14 +81,14 @@ class TextEncoder256Sim(nn.Module):
         self.n_layers = n_layers
         self.kernel_size = kernel_size
         self.p_dropout = p_dropout
-        self.emb_phone = nn.Linear(256, hidden_channels)
+        self.emb_phone = nn.Linear(768, hidden_channels)
         self.lrelu = nn.LeakyReLU(0.1, inplace=True)
         if f0 == True:
             self.emb_pitch = nn.Embedding(256, hidden_channels)  # pitch 256
         self.encoder = attentions.Encoder(
             hidden_channels, filter_channels, n_heads, n_layers, kernel_size, p_dropout
         )
-        self.proj = nn.Conv1d(hidden_channels, out_channels, 1)
+        self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
 
     def forward(self, phone, pitch, lengths):
         if pitch == None:
@@ -102,8 +102,10 @@ class TextEncoder256Sim(nn.Module):
             x.dtype
         )
         x = self.encoder(x * x_mask, x_mask)
-        x = self.proj(x) * x_mask
-        return x, x_mask
+        stats = self.proj(x) * x_mask
+
+        m, logs = torch.split(stats, self.out_channels, dim=1)
+        return m, logs, x_mask
 
 
 class ResidualCouplingBlock(nn.Module):
@@ -638,6 +640,117 @@ class SynthesizerTrnMs256NSFsid(nn.Module):
         return o, x_mask, (z, z_p, m_p, logs_p)
 
 
+class SynthesizerTrnMs768NSFsid(nn.Module):
+    def __init__(
+        self,
+        spec_channels,
+        segment_size,
+        inter_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        spk_embed_dim,
+        gin_channels,
+        sr,
+        **kwargs
+    ):
+        super().__init__()
+        if type(sr) == type("strr"):
+            sr = sr2sr[sr]
+        self.spec_channels = spec_channels
+        self.inter_channels = inter_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.resblock = resblock
+        self.resblock_kernel_sizes = resblock_kernel_sizes
+        self.resblock_dilation_sizes = resblock_dilation_sizes
+        self.upsample_rates = upsample_rates
+        self.upsample_initial_channel = upsample_initial_channel
+        self.upsample_kernel_sizes = upsample_kernel_sizes
+        self.segment_size = segment_size
+        self.gin_channels = gin_channels
+        # self.hop_length = hop_length#
+        self.spk_embed_dim = spk_embed_dim
+        self.enc_p = TextEncoder768(
+            inter_channels,
+            hidden_channels,
+            filter_channels,
+            n_heads,
+            n_layers,
+            kernel_size,
+            p_dropout,
+        )
+        self.dec = GeneratorNSF(
+            inter_channels,
+            resblock,
+            resblock_kernel_sizes,
+            resblock_dilation_sizes,
+            upsample_rates,
+            upsample_initial_channel,
+            upsample_kernel_sizes,
+            gin_channels=gin_channels,
+            sr=sr,
+            is_half=kwargs["is_half"],
+        )
+        self.enc_q = PosteriorEncoder(
+            spec_channels,
+            inter_channels,
+            hidden_channels,
+            5,
+            1,
+            16,
+            gin_channels=gin_channels,
+        )
+        self.flow = ResidualCouplingBlock(
+            inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
+        )
+        self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
+        print("gin_channels:", gin_channels, "self.spk_embed_dim:", self.spk_embed_dim)
+
+    def remove_weight_norm(self):
+        self.dec.remove_weight_norm()
+        self.flow.remove_weight_norm()
+        self.enc_q.remove_weight_norm()
+
+    def forward(
+        self, phone, phone_lengths, pitch, pitchf, y, y_lengths, ds
+    ):  # 这里ds是id，[bs,1]
+        # print(1,pitch.shape)#[bs,t]
+        g = self.emb_g(ds).unsqueeze(-1)  # [b, 256, 1]##1是t，广播的
+        m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
+        z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
+        z_p = self.flow(z, y_mask, g=g)
+        z_slice, ids_slice = commons.rand_slice_segments(
+            z, y_lengths, self.segment_size
+        )
+        # print(-1,pitchf.shape,ids_slice,self.segment_size,self.hop_length,self.segment_size//self.hop_length)
+        pitchf = commons.slice_segments2(pitchf, ids_slice, self.segment_size)
+        # print(-2,pitchf.shape,z_slice.shape)
+        o = self.dec(z_slice, pitchf, g=g)
+        return o, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
+
+    def infer(self, phone, phone_lengths, pitch, nsff0, sid, max_len=None):
+        g = self.emb_g(sid).unsqueeze(-1)
+        m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
+        z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask
+        z = self.flow(z_p, x_mask, g=g, reverse=True)
+        o = self.dec((z * x_mask)[:, :, :max_len], nsff0, g=g)
+        return o, x_mask, (z, z_p, m_p, logs_p)
+
+
 class SynthesizerTrnMs256NSFsid_nono(nn.Module):
     def __init__(
         self,
@@ -740,11 +853,7 @@ class SynthesizerTrnMs256NSFsid_nono(nn.Module):
         return o, x_mask, (z, z_p, m_p, logs_p)
 
 
-class SynthesizerTrnMs256NSFsid_sim(nn.Module):
-    """
-    Synthesizer for Training
-    """
-
+class SynthesizerTrnMs768NSFsid_nono(nn.Module):
     def __init__(
         self,
         spec_channels,
@@ -763,9 +872,8 @@ class SynthesizerTrnMs256NSFsid_sim(nn.Module):
         upsample_initial_channel,
         upsample_kernel_sizes,
         spk_embed_dim,
-        # hop_length,
-        gin_channels=0,
-        use_sdp=True,
+        gin_channels,
+        sr=None,
         **kwargs
     ):
         super().__init__()
@@ -787,7 +895,7 @@ class SynthesizerTrnMs256NSFsid_sim(nn.Module):
         self.gin_channels = gin_channels
         # self.hop_length = hop_length#
         self.spk_embed_dim = spk_embed_dim
-        self.enc_p = TextEncoder256Sim(
+        self.enc_p = TextEncoder768(
             inter_channels,
             hidden_channels,
             filter_channels,
@@ -795,8 +903,9 @@ class SynthesizerTrnMs256NSFsid_sim(nn.Module):
             n_layers,
             kernel_size,
             p_dropout,
+            f0=False,
         )
-        self.dec = GeneratorNSF(
+        self.dec = Generator(
             inter_channels,
             resblock,
             resblock_kernel_sizes,
@@ -805,9 +914,16 @@ class SynthesizerTrnMs256NSFsid_sim(nn.Module):
             upsample_initial_channel,
             upsample_kernel_sizes,
             gin_channels=gin_channels,
-            is_half=kwargs["is_half"],
         )
-
+        self.enc_q = PosteriorEncoder(
+            spec_channels,
+            inter_channels,
+            hidden_channels,
+            5,
+            1,
+            16,
+            gin_channels=gin_channels,
+        )
         self.flow = ResidualCouplingBlock(
             inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
         )
@@ -819,28 +935,24 @@ class SynthesizerTrnMs256NSFsid_sim(nn.Module):
         self.flow.remove_weight_norm()
         self.enc_q.remove_weight_norm()
 
-    def forward(
-        self, phone, phone_lengths, pitch, pitchf, y_lengths, ds
-    ):  # y是spec不需要了现在
+    def forward(self, phone, phone_lengths, y, y_lengths, ds):  # 这里ds是id，[bs,1]
         g = self.emb_g(ds).unsqueeze(-1)  # [b, 256, 1]##1是t，广播的
-        x, x_mask = self.enc_p(phone, pitch, phone_lengths)
-        x = self.flow(x, x_mask, g=g, reverse=True)
+        m_p, logs_p, x_mask = self.enc_p(phone, None, phone_lengths)
+        z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
+        z_p = self.flow(z, y_mask, g=g)
         z_slice, ids_slice = commons.rand_slice_segments(
-            x, y_lengths, self.segment_size
+            z, y_lengths, self.segment_size
         )
+        o = self.dec(z_slice, g=g)
+        return o, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
 
-        pitchf = commons.slice_segments2(pitchf, ids_slice, self.segment_size)
-        o = self.dec(z_slice, pitchf, g=g)
-        return o, ids_slice
-
-    def infer(
-        self, phone, phone_lengths, pitch, pitchf, ds, max_len=None
-    ):  # y是spec不需要了现在
-        g = self.emb_g(ds).unsqueeze(-1)  # [b, 256, 1]##1是t，广播的
-        x, x_mask = self.enc_p(phone, pitch, phone_lengths)
-        x = self.flow(x, x_mask, g=g, reverse=True)
-        o = self.dec((x * x_mask)[:, :, :max_len], pitchf, g=g)
-        return o, o
+    def infer(self, phone, phone_lengths, sid, max_len=None):
+        g = self.emb_g(sid).unsqueeze(-1)
+        m_p, logs_p, x_mask = self.enc_p(phone, None, phone_lengths)
+        z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask
+        z = self.flow(z_p, x_mask, g=g, reverse=True)
+        o = self.dec((z * x_mask)[:, :, :max_len], g=g)
+        return o, x_mask, (z, z_p, m_p, logs_p)
 
 
 class MultiPeriodDiscriminator(torch.nn.Module):
@@ -873,6 +985,36 @@ class MultiPeriodDiscriminator(torch.nn.Module):
         return y_d_rs, y_d_gs, fmap_rs, fmap_gs
 
 
+class MultiPeriodDiscriminatorV2(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(MultiPeriodDiscriminatorV2, self).__init__()
+        # periods = [2, 3, 5, 7, 11, 17]
+        periods = [2, 3, 5, 7, 11, 17, 23, 37]
+
+        discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
+        discs = discs + [
+            DiscriminatorP(i, use_spectral_norm=use_spectral_norm) for i in periods
+        ]
+        self.discriminators = nn.ModuleList(discs)
+
+    def forward(self, y, y_hat):
+        y_d_rs = []  #
+        y_d_gs = []
+        fmap_rs = []
+        fmap_gs = []
+        for i, d in enumerate(self.discriminators):
+            y_d_r, fmap_r = d(y)
+            y_d_g, fmap_g = d(y_hat)
+            # for j in range(len(fmap_r)):
+            #     print(i,j,y.shape,y_hat.shape,fmap_r[j].shape,fmap_g[j].shape)
+            y_d_rs.append(y_d_r)
+            y_d_gs.append(y_d_g)
+            fmap_rs.append(fmap_r)
+            fmap_gs.append(fmap_g)
+
+        return y_d_rs, y_d_gs, fmap_rs, fmap_gs
+
+
 class DiscriminatorS(torch.nn.Module):
     def __init__(self, use_spectral_norm=False):
         super(DiscriminatorS, self).__init__()

infer_pack/models_onnx.py

@@ -61,7 +61,7 @@ class TextEncoder256(nn.Module):
         return m, logs, x_mask
 
 
-class TextEncoder256Sim(nn.Module):
+class TextEncoder768(nn.Module):
     def __init__(
         self,
         out_channels,
@@ -81,14 +81,14 @@ class TextEncoder256Sim(nn.Module):
         self.n_layers = n_layers
         self.kernel_size = kernel_size
         self.p_dropout = p_dropout
-        self.emb_phone = nn.Linear(256, hidden_channels)
+        self.emb_phone = nn.Linear(768, hidden_channels)
         self.lrelu = nn.LeakyReLU(0.1, inplace=True)
         if f0 == True:
             self.emb_pitch = nn.Embedding(256, hidden_channels)  # pitch 256
         self.encoder = attentions.Encoder(
             hidden_channels, filter_channels, n_heads, n_layers, kernel_size, p_dropout
         )
-        self.proj = nn.Conv1d(hidden_channels, out_channels, 1)
+        self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
 
     def forward(self, phone, pitch, lengths):
         if pitch == None:
@@ -102,8 +102,10 @@ class TextEncoder256Sim(nn.Module):
             x.dtype
         )
         x = self.encoder(x * x_mask, x_mask)
-        x = self.proj(x) * x_mask
-        return x, x_mask
+        stats = self.proj(x) * x_mask
+
+        m, logs = torch.split(stats, self.out_channels, dim=1)
+        return m, logs, x_mask
 
 
 class ResidualCouplingBlock(nn.Module):
@@ -527,7 +529,7 @@ sr2sr = {
 }
 
 
-class SynthesizerTrnMs256NSFsidO(nn.Module):
+class SynthesizerTrnMsNSFsidM(nn.Module):
     def __init__(
         self,
         spec_channels,
@@ -571,15 +573,26 @@ class SynthesizerTrnMs256NSFsidO(nn.Module):
         self.gin_channels = gin_channels
         # self.hop_length = hop_length#
         self.spk_embed_dim = spk_embed_dim
-        self.enc_p = TextEncoder256(
-            inter_channels,
-            hidden_channels,
-            filter_channels,
-            n_heads,
-            n_layers,
-            kernel_size,
-            p_dropout,
-        )
+        if self.gin_channels == 256:
+            self.enc_p = TextEncoder256(
+                inter_channels,
+                hidden_channels,
+                filter_channels,
+                n_heads,
+                n_layers,
+                kernel_size,
+                p_dropout,
+            )
+        else:
+            self.enc_p = TextEncoder768(
+                inter_channels,
+                hidden_channels,
+                filter_channels,
+                n_heads,
+                n_layers,
+                kernel_size,
+                p_dropout,
+            )
         self.dec = GeneratorNSF(
             inter_channels,
             resblock,
@@ -605,6 +618,7 @@ class SynthesizerTrnMs256NSFsidO(nn.Module):
             inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
         )
         self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
+        self.speaker_map = None
         print("gin_channels:", gin_channels, "self.spk_embed_dim:", self.spk_embed_dim)
 
     def remove_weight_norm(self):
@@ -612,10 +626,24 @@ class SynthesizerTrnMs256NSFsidO(nn.Module):
         self.flow.remove_weight_norm()
         self.enc_q.remove_weight_norm()
 
-    def forward(self, phone, phone_lengths, pitch, nsff0, sid, max_len=None):
-        g = self.emb_g(sid).unsqueeze(-1)
+    def construct_spkmixmap(self, n_speaker):
+        self.speaker_map = torch.zeros((n_speaker, 1, 1, self.gin_channels))
+        for i in range(n_speaker):
+            self.speaker_map[i] = self.emb_g(torch.LongTensor([[i]]))
+        self.speaker_map = self.speaker_map.unsqueeze(0)
+
+    def forward(self, phone, phone_lengths, pitch, nsff0, g, rnd, max_len=None):
+        if self.speaker_map is not None:  # [N, S]  *  [S, B, 1, H]
+            g = g.reshape((g.shape[0], g.shape[1], 1, 1, 1))  # [N, S, B, 1, 1]
+            g = g * self.speaker_map  # [N, S, B, 1, H]
+            g = torch.sum(g, dim=1)  # [N, 1, B, 1, H]
+            g = g.transpose(0, -1).transpose(0, -2).squeeze(0)  # [B, H, N]
+        else:
+            g = g.unsqueeze(0)
+            g = self.emb_g(g).transpose(1, 2)
+
         m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
-        z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask
+        z_p = (m_p + torch.exp(logs_p) * rnd) * x_mask
         z = self.flow(z_p, x_mask, g=g, reverse=True)
         o = self.dec((z * x_mask)[:, :, :max_len], nsff0, g=g)
         return o
@@ -651,6 +679,36 @@ class MultiPeriodDiscriminator(torch.nn.Module):
         return y_d_rs, y_d_gs, fmap_rs, fmap_gs
 
 
+class MultiPeriodDiscriminatorV2(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(MultiPeriodDiscriminatorV2, self).__init__()
+        # periods = [2, 3, 5, 7, 11, 17]
+        periods = [2, 3, 5, 7, 11, 17, 23, 37]
+
+        discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
+        discs = discs + [
+            DiscriminatorP(i, use_spectral_norm=use_spectral_norm) for i in periods
+        ]
+        self.discriminators = nn.ModuleList(discs)
+
+    def forward(self, y, y_hat):
+        y_d_rs = []  #
+        y_d_gs = []
+        fmap_rs = []
+        fmap_gs = []
+        for i, d in enumerate(self.discriminators):
+            y_d_r, fmap_r = d(y)
+            y_d_g, fmap_g = d(y_hat)
+            # for j in range(len(fmap_r)):
+            #     print(i,j,y.shape,y_hat.shape,fmap_r[j].shape,fmap_g[j].shape)
+            y_d_rs.append(y_d_r)
+            y_d_gs.append(y_d_g)
+            fmap_rs.append(fmap_r)
+            fmap_gs.append(fmap_g)
+
+        return y_d_rs, y_d_gs, fmap_rs, fmap_gs
+
+
 class DiscriminatorS(torch.nn.Module):
     def __init__(self, use_spectral_norm=False):
         super(DiscriminatorS, self).__init__()

requirements.txt

@@ -1,46 +1,26 @@
-numba==0.56.4
-numpy==1.23.5
-scipy==1.9.3
-librosa==0.9.2
-llvmlite==0.39.0
+setuptools
+wheel
+httpx==0.23.0
 fairseq==0.12.2
-faiss-cpu==1.7.0; sys_platform == "darwin"
-faiss-cpu==1.7.2; sys_platform != "darwin"
 gradio
-Cython
-future>=0.18.3
-pydub>=0.25.1
-soundfile>=0.12.1
-ffmpeg-python>=0.2.0
-tensorboardX
-functorch>=2.0.0
-Jinja2>=3.1.2
-json5>=0.9.11
-Markdown
-matplotlib>=3.7.1
-matplotlib-inline>=0.1.6
-praat-parselmouth>=0.4.3
-Pillow>=9.1.1
+ffmpeg
+praat-parselmouth
+pyworld
+numpy==1.23.5
+numba==0.56.4
+librosa==0.9.2
+faiss-cpu==1.7.3
+faiss-gpu
+scipy==1.9.3
 pyworld>=0.3.2
-resampy>=0.4.2
-scikit-learn>=1.2.2
-starlette>=0.26.1
 tensorboard
-tensorboard-data-server
-tensorboard-plugin-wit
-torchgen>=0.0.1
-tqdm>=4.65.0
-tornado>=6.2
-Werkzeug>=2.2.3
-uc-micro-py>=1.0.1
-sympy>=1.11.1
-tabulate>=0.9.0
-PyYAML>=6.0
-pyasn1>=0.4.8
-pyasn1-modules>=0.2.8
-fsspec>=2023.3.0
-absl-py>=1.4.0
-audioread
-uvicorn>=0.21.1
-colorama>=0.4.6
+tensorboardX
+onnxruntime
+pyngrok==4.1.12
+soundfile>=0.12.1
+tqdm>=4.63.1
+torchcrepe
+asyncio
 edge-tts
+demucs
+yt_dlp

vc_infer_pipeline.py

@@ -2,11 +2,50 @@ import numpy as np, parselmouth, torch, pdb
 from time import time as ttime
 import torch.nn.functional as F
 import scipy.signal as signal
-import pyworld, os, traceback, faiss
+import pyworld, os, traceback, faiss, librosa, torchcrepe
 from scipy import signal
+from functools import lru_cache
 
 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):
@@ -27,7 +66,17 @@ class VC(object):
         self.t_max = self.sr * self.x_max  # 免查询时长阈值
         self.device = config.device
 
-    def get_f0(self, x, p_len, f0_up_key, f0_method, inp_f0=None):
+    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
@@ -50,15 +99,31 @@ class VC(object):
                     f0, [[pad_size, p_len - len(f0) - pad_size]], mode="constant"
                 )
         elif f0_method == "harvest":
-            f0, t = pyworld.harvest(
-                x.astype(np.double),
-                fs=self.sr,
-                f0_ceil=f0_max,
-                f0_floor=f0_min,
-                frame_period=10,
+            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,
             )
-            f0 = pyworld.stonemask(x.astype(np.double), f0, t, self.sr)
-            f0 = signal.medfilt(f0, 3)
+            pd = torchcrepe.filter.median(pd, 3)
+            f0 = torchcrepe.filter.mean(f0, 3)
+            f0[pd < 0.1] = 0
+            f0 = f0[0].cpu().numpy()
         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
@@ -96,6 +161,8 @@ class VC(object):
         index,
         big_npy,
         index_rate,
+        version,
+        protect
     ):  # ,file_index,file_big_npy
         feats = torch.from_numpy(audio0)
         if self.is_half:
@@ -111,13 +178,14 @@ class VC(object):
         inputs = {
             "source": feats.to(self.device),
             "padding_mask": padding_mask,
-            "output_layer": 9,  # layer 9
+            "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])
-
+            feats = model.final_proj(logits[0]) if version == "v1" else logits[0]
+        if(protect<0.5):
+            feats0=feats.clone()
         if (
             isinstance(index, type(None)) == False
             and isinstance(big_npy, type(None)) == False
@@ -143,6 +211,8 @@ class VC(object):
             )
 
         feats = F.interpolate(feats.permute(0, 2, 1), scale_factor=2).permute(0, 2, 1)
+        if(protect<0.5):
+            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:
@@ -150,23 +220,26 @@ class VC(object):
             if pitch != None and pitchf != None:
                 pitch = pitch[:, :p_len]
                 pitchf = pitchf[:, :p_len]
+
+        if(protect<0.5):
+            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] * 32768)
+                    (net_g.infer(feats, p_len, pitch, pitchf, sid)[0][0, 0])
                     .data.cpu()
                     .float()
                     .numpy()
-                    .astype(np.int16)
                 )
             else:
                 audio1 = (
-                    (net_g.infer(feats, p_len, sid)[0][0, 0] * 32768)
-                    .data.cpu()
-                    .float()
-                    .numpy()
-                    .astype(np.int16)
+                    (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():
@@ -182,6 +255,7 @@ class VC(object):
         net_g,
         sid,
         audio,
+        input_audio_path,
         times,
         f0_up_key,
         f0_method,
@@ -189,6 +263,12 @@ class VC(object):
         # file_big_npy,
         index_rate,
         if_f0,
+        filter_radius,
+        tgt_sr,
+        resample_sr,
+        rms_mix_rate,
+        version,
+        protect,
         f0_file=None,
     ):
         if (
@@ -243,9 +323,19 @@ class VC(object):
         sid = torch.tensor(sid, device=self.device).unsqueeze(0).long()
         pitch, pitchf = None, None
         if if_f0 == 1:
-            pitch, pitchf = self.get_f0(audio_pad, p_len, f0_up_key, f0_method, inp_f0)
+            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()
@@ -265,6 +355,8 @@ class VC(object):
                         index,
                         big_npy,
                         index_rate,
+                        version,
+                        protect
                     )[self.t_pad_tgt : -self.t_pad_tgt]
                 )
             else:
@@ -280,6 +372,8 @@ class VC(object):
                         index,
                         big_npy,
                         index_rate,
+                        version,
+                        protect
                     )[self.t_pad_tgt : -self.t_pad_tgt]
                 )
             s = t
@@ -296,6 +390,8 @@ class VC(object):
                     index,
                     big_npy,
                     index_rate,
+                    version,
+                    protect
                 )[self.t_pad_tgt : -self.t_pad_tgt]
             )
         else:
@@ -311,9 +407,22 @@ class VC(object):
                     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()