Update app.py

app.py

@@ -8,6 +8,8 @@ from scipy.io.wavfile import write
 from transformers import pipeline
 
 
+hf_token = os.environ.get("HF_TOKEN")
+
 MODEL_NAME = "openai/whisper-large-v3-turbo"
 BATCH_SIZE = 8
 
@@ -18,19 +20,22 @@ pipe = pipeline(
     model=MODEL_NAME,
     chunk_length_s=30,
     device=device,
+    token=hf_token,
 )
 
 
-def split_process(audio, chosen_out_track="vocals", progress=gr.Progress(track_tqdm=True)):
+def split_process(audio, chosen_out_track, progress=gr.Progress(track_tqdm=True)):
     if audio is None:
-        raise gr.Error("Upload an audio file first.")
+        raise gr.Error(
+            "No audio file submitted! Please upload or record an audio file before submitting your request."
+        )
 
     os.makedirs("out", exist_ok=True)
 
     progress(0.02, desc="Preparing audio...")
     write("test.wav", audio[0], audio[1])
 
-    progress(0.05, desc="Starting Demucs...")
+    progress(0.05, desc="Starting vocal separation...")
 
     cmd = [
         "python3",
@@ -56,7 +61,7 @@ def split_process(audio, chosen_out_track="vocals", progress=gr.Progress(track_t
     percent_re = re.compile(r"(\d{1,3})%")
     completed_bars = 0
     last_percent = 0
-    max_bars = 4  # mdx_extra_q = bag of 4 Demucs models
+    max_bars = 4
     logs = []
 
     for line in process.stdout:
@@ -93,30 +98,32 @@ def split_process(audio, chosen_out_track="vocals", progress=gr.Progress(track_t
             f"{''.join(logs)[-2000:]}"
         )
 
-    tracks = {
-        "vocals": "./out/mdx_extra_q/test/vocals.wav",
-        "bass": "./out/mdx_extra_q/test/bass.wav",
-        "drums": "./out/mdx_extra_q/test/drums.wav",
-        "other": "./out/mdx_extra_q/test/other.wav",
-        "all-in": "test.wav",
-    }
-
-    output_path = tracks.get(chosen_out_track)
-
-    if output_path is None:
+    if chosen_out_track == "vocals":
+        output_path = "./out/mdx_extra_q/test/vocals.wav"
+    elif chosen_out_track == "bass":
+        output_path = "./out/mdx_extra_q/test/bass.wav"
+    elif chosen_out_track == "drums":
+        output_path = "./out/mdx_extra_q/test/drums.wav"
+    elif chosen_out_track == "other":
+        output_path = "./out/mdx_extra_q/test/other.wav"
+    elif chosen_out_track == "all-in":
+        output_path = "test.wav"
+    else:
         raise gr.Error(f"Unknown output track: {chosen_out_track}")
 
     if not os.path.exists(output_path):
         raise gr.Error(f"Expected output file was not created: {output_path}")
 
-    progress(0.90, desc="Demucs separation complete.")
+    progress(0.90, desc="Vocal separation complete.")
 
     return output_path
 
 
-def transcribe(inputs, task="transcribe"):
+def transcribe(inputs, task):
     if inputs is None:
-        raise gr.Error("No audio file found.")
+        raise gr.Error(
+            "No audio file submitted! Please upload or record an audio file before submitting your request."
+        )
 
     result = pipe(
         inputs,
@@ -128,53 +135,107 @@ def transcribe(inputs, task="transcribe"):
     return result["text"]
 
 
-def infer(audio, progress=gr.Progress(track_tqdm=True)):
+def format_lyrics(text):
+    if not text:
+        return ""
+
+    # Remove unwanted subtitle artifacts
+    text = re.sub(
+        r"Sous-?titrage Société Radio-Canada",
+        "",
+        text,
+        flags=re.IGNORECASE,
+    )
+
+    # Remove repeated newlines
+    text = re.sub(r"\n+", "\n", text).strip()
+
+    # Insert a newline before capital letters, like in the original app
+    formatted_text = re.sub(r"(?<!^)([A-Z])", r"\n\1", text)
+
+    # Remove leading whitespace on each line
+    formatted_text = re.sub(
+        r"^[ \t]+",
+        "",
+        formatted_text,
+        flags=re.MULTILINE,
+    )
+
+    return formatted_text.strip()
+
+
+def infer(audio_input, progress=gr.Progress(track_tqdm=True)):
     progress(0.0, desc="Starting...")
 
-    vocals_path = split_process(
-        audio,
-        chosen_out_track="vocals",
+    # STEP 1 | Split vocals from the song/audio file
+    splt_result = split_process(
+        audio_input,
+        "vocals",
         progress=progress,
     )
 
+    print(splt_result)
+
+    # STEP 2 | Transcribe vocals
     progress(0.92, desc="Transcribing vocals with Whisper...")
-    lyrics = transcribe(vocals_path, task="transcribe")
+
+    whisper_result = transcribe(
+        splt_result,
+        "transcribe",
+    )
+
+    print(whisper_result)
+
+    # STEP 3 | Format lyrics
+    progress(0.98, desc="Formatting lyrics...")
+
+    lyrics = format_lyrics(whisper_result)
+
+    print(lyrics)
 
     progress(1.0, desc="Done.")
 
-    return vocals_path, lyrics
+    return splt_result, lyrics
 
 
 css = """
 #col-container {
-    max-width: 780px;
+    max-width: 510px;
     margin-left: auto;
     margin-right: auto;
 }
-
-a {
-    text-decoration-line: underline;
-    font-weight: 600;
-}
 """
 
 
 with gr.Blocks() as demo:
     with gr.Column(elem_id="col-container"):
-        gr.Markdown(
+        gr.HTML(
+            """
+            <div style="text-align: center; max-width: 700px; margin: 0 auto;">
+                <div
+                    style="
+                        display: inline-flex;
+                        align-items: center;
+                        gap: 0.8rem;
+                        font-size: 1.75rem;
+                    "
+                >
+                    <h1 style="font-weight: 900; margin-bottom: 7px; margin-top: 5px;">
+                        Song To Lyrics
+                    </h1>
+                </div>
+                <p style="margin-bottom: 10px; font-size: 94%">
+                    Send the audio file of your favorite song, and get the lyrics! <br />
+                    Under the hood, we split and get the vocals track from the audio file, then send the vocals to Whisper.
+                </p>
+            </div>
             """
-# Music To Lyrics
-
-Upload a song and get the lyrics.
-
-The app first separates the vocals with Demucs, then transcribes them with Whisper.
-"""
         )
 
         song_in = gr.Audio(
             label="Song input",
             type="numpy",
-            sources=["upload"],
+            sources="upload",
         )
 
         getlyrics_btn = gr.Button("Get Lyrics!")
@@ -182,11 +243,11 @@ The app first separates the vocals with Demucs, then transcribes them with Whisp
         vocals_out = gr.Audio(label="Vocals Only")
         lyrics_res = gr.Textbox(label="Lyrics")
 
-        getlyrics_btn.click(
-            fn=infer,
-            inputs=[song_in],
-            outputs=[vocals_out, lyrics_res],
-        )
+    getlyrics_btn.click(
+        fn=infer,
+        inputs=[song_in],
+        outputs=[vocals_out, lyrics_res],
+    )
 
 
 demo.queue().launch(css=css, ssr_mode=False)