Create app.py

app.py

@@ -0,0 +1,209 @@
+import os
+import librosa
+import soundfile
+import gradio as gr
+from scipy.io import wavfile
+import subprocess
+from pathlib import Path
+import time
+import random
+from datetime import datetime
+from scipy.io.wavfile import write
+from pydub import AudioSegment
+high_quality = True
+
+
+def svc(audio_src, audio_ref, singing_check):
+  wav1, sr1 = librosa.load(audio_src, sr=24000)
+  soundfile.write("audio_source.wav", wav1, sr1)
+  wav2, sr2 = librosa.load(audio_ref, sr=24000)
+  soundfile.write("audio_reference.wav", wav2, sr2)
+  if singing_check == True:
+    os.system("python infer.py --src_wav_path audio_source.wav --ref_wav_path audio_reference.wav --out_path output.wav")
+
+  else:
+    os.system("python infer.py --src_wav_path audio_source.wav --ref_wav_path audio_reference.wav --out_path output.wav --speech_enroll")
+
+  return "output.wav"
+
+def youtube_downloader(
+    video_identifier,
+    start_time,
+    end_time,
+    is_full_song,
+    output_filename="track.wav",
+    num_attempts=5,
+    url_base="",
+    quiet=False,
+    force=True,
+):
+    if is_full_song:
+        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([video_identifier])
+        audio_path = "dl_audio/youtube_audio.wav"
+        return audio_path
+
+    else:
+        output_path = Path(output_filename)
+        if output_path.exists():
+            if not force:
+                return output_path
+            else:
+                output_path.unlink()
+
+        quiet = "--quiet --no-warnings" if quiet else ""
+        command = f"""
+            yt-dlp {quiet} -x --audio-format wav -f bestaudio -o "{output_filename}" --download-sections "*{start_time}-{end_time}" "{url_base}{video_identifier}"  # noqa: E501
+        """.strip()
+
+        attempts = 0
+        while True:
+            try:
+                _ = subprocess.check_output(command, shell=True, stderr=subprocess.STDOUT)
+            except subprocess.CalledProcessError:
+                attempts += 1
+                if attempts == num_attempts:
+                    return None
+            else:
+                break
+
+        if output_path.exists():
+            return output_path
+        else:
+            return None
+
+def audio_separated(audio_input, progress=gr.Progress()):
+    # start progress
+    progress(progress=0, desc="Starting...")
+    time.sleep(0.1)
+
+    # check file input
+    if audio_input is None:
+        # show progress
+        for i in progress.tqdm(range(100), desc="Please wait..."):
+            time.sleep(0.01)
+
+        return (None, None, 'Please input audio.')
+
+    # create filename
+    filename = str(random.randint(10000,99999))+datetime.now().strftime("%d%m%Y%H%M%S")
+
+    # progress
+    progress(progress=0.10, desc="Please wait...")
+
+    # make dir output
+    os.makedirs("output", exist_ok=True)
+
+    # progress
+    progress(progress=0.20, desc="Please wait...")
+
+    # write
+    if high_quality:
+        write(filename+".wav", audio_input[0], audio_input[1])
+    else:
+        write(filename+".mp3", audio_input[0], audio_input[1])
+
+    # progress
+    progress(progress=0.50, desc="Please wait...")
+
+    # demucs process
+    if high_quality:
+        command_demucs = "python3 -m demucs --two-stems=vocals -d cpu "+filename+".wav -o output"
+    else:
+        command_demucs = "python3 -m demucs --two-stems=vocals --mp3 --mp3-bitrate 128 -d cpu "+filename+".mp3 -o output"
+
+    os.system(command_demucs)
+
+    # progress
+    progress(progress=0.70, desc="Please wait...")
+
+    # remove file audio
+    if high_quality:
+        command_delete = "rm -v ./"+filename+".wav"
+    else:
+        command_delete = "rm -v ./"+filename+".mp3"
+
+    os.system(command_delete)
+
+    # progress
+    progress(progress=0.80, desc="Please wait...")
+
+    # progress
+    for i in progress.tqdm(range(80,100), desc="Please wait..."):
+        time.sleep(0.1)
+
+    if high_quality:
+        return "./output/htdemucs/"+filename+"/vocals.wav","./output/htdemucs/"+filename+"/no_vocals.wav","Successfully..."
+    else:
+        return "./output/htdemucs/"+filename+"/vocals.mp3","./output/htdemucs/"+filename+"/no_vocals.mp3","Successfully..."
+
+# mix vocal and non-vocal
+def mix(audio1, audio2):
+  sound1 = AudioSegment.from_file(audio1)
+  sound2 = AudioSegment.from_file(audio2)
+  length = len(sound1)
+  mixed = sound1[:length].overlay(sound2)
+
+  mixed.export("song.wav", format="wav")
+
+  return "song.wav"
+
+app = gr.Blocks()
+
+with app:
+  gr.Markdown("# <center>🥳💕🎶 NeuCo AI歌手，无需训练、一键翻唱</center>")
+  gr.Markdown("## <center>🌟 只需10秒音频，一键翻唱任意歌手的任意歌曲！ </center>")
+  gr.Markdown("### <center>🌊 更多精彩应用，敬请关注[滔滔AI](http://www.talktalkai.com)；滔滔AI，为爱滔滔！💕</center>")
+  with gr.Row():
+    with gr.Column():
+      ydl_url_input  = gr.Textbox(label="音乐视频网址(可直接填写相应的BV号)", value = "https://www.bilibili.com/video/BV...")
+      with gr.Row():
+        start = gr.Number(value=0, label="起始时间 (秒)")
+        end = gr.Number(value=15, label="结束时间 (秒)")
+        check_full = gr.Checkbox(label="是否上传整首歌曲", info="若勾选则不需要填写起止时间", value=False)
+      ydl_url_submit = gr.Button("提取声音文件吧", variant="primary")
+      as_audio_submit = gr.Button("去除背景音吧", variant="primary")
+
+    with gr.Column():
+      ydl_audio_output = gr.Audio(label="歌曲原声")
+      as_audio_input = ydl_audio_output
+      as_audio_vocals = gr.Audio(label="歌曲人声部分", type="filepath")
+      as_audio_no_vocals = gr.Audio(label="歌曲伴奏部分", type="filepath", visible=False)
+      as_audio_message = gr.Textbox(label="Message", visible=False)
+
+  ydl_url_submit.click(fn=youtube_downloader, inputs=[ydl_url_input, start, end, check_full], outputs=[ydl_audio_output])
+  as_audio_submit.click(fn=audio_separated, inputs=[as_audio_input], outputs=[as_audio_vocals, as_audio_no_vocals, as_audio_message], show_progress=True, queue=True)
+
+  with gr.Row():
+    with gr.Column():
+      inp_src = as_audio_vocals
+      inp_ref = gr.Audio(label="请上传您喜欢的参考音频", type="filepath")
+      check_singing = gr.Checkbox(label="参考音频是否为歌曲", info="勾选：参考音频是歌曲；不勾选：参考音频是语音", value=True)
+      btn1 = gr.Button("一键歌声转换吧！", variant="primary")
+      btn2 = gr.Button("加入歌曲伴奏吧！", variant="primary")
+
+    with gr.Column():
+      out_svc = gr.Audio(label="为您生成的专属歌曲", type="filepath", interactive=False)
+      new_song = gr.Audio(label="AI歌手+伴奏", type="filepath")
+
+  btn1.click(svc, [inp_src, inp_ref, check_singing], out_svc)
+  btn2.click(fn=mix, inputs=[out_svc, as_audio_no_vocals], outputs=[new_song])
+
+  gr.Markdown("### <center>注意❗：请不要生成会对个人以及组织造成侵害的内容，此程序仅供科研、学习及个人娱乐使用。</center>")
+  gr.HTML('''
+      <div class="footer">
+                  <p>🌊🏞️🎶 - 江水东流急，滔滔无尽声。 明·顾璘
+                  </p>
+      </div>
+  ''')
+
+app.queue().launch(show_error=True)