app.py

import gradio as gr
from gradio_client import Client
import torch
import os 
from scipy.io.wavfile import write
hf_token = os.environ.get('HF_TOKEN')
#splt_client = Client("https://fffiloni-splittrack2musicgen.hf.space/")

def split_process(audio, chosen_out_track):
    os.makedirs("out", exist_ok=True)
    write('test.wav', audio[0], audio[1])
    os.system("python3 -m demucs.separate -n mdx_extra_q -j 4 test.wav -o out")
    #return "./out/mdx_extra_q/test/vocals.wav","./out/mdx_extra_q/test/bass.wav","./out/mdx_extra_q/test/drums.wav","./out/mdx_extra_q/test/other.wav"
    if chosen_out_track == "vocals":
        return "./out/mdx_extra_q/test/vocals.wav"
    elif chosen_out_track == "bass":
        return "./out/mdx_extra_q/test/bass.wav"
    elif chosen_out_track == "drums":
        return "./out/mdx_extra_q/test/drums.wav"
    elif chosen_out_track == "other":
        return "./out/mdx_extra_q/test/other.wav"
    elif chosen_out_track == "all-in":
        return "test.wav"

from transformers import pipeline
from transformers.pipelines.audio_utils import ffmpeg_read

import tempfile

MODEL_NAME = "openai/whisper-large-v3-turbo"
BATCH_SIZE = 8
FILE_LIMIT_MB = 1000
YT_LENGTH_LIMIT_S = 3600  # limit to 1 hour YouTube files

device = 0 if torch.cuda.is_available() else "cpu"

pipe = pipeline(
    task="automatic-speech-recognition",
    model=MODEL_NAME,
    chunk_length_s=30,
    device=device,
    token=hf_token
)


#@spaces.GPU
def transcribe(inputs, task):
    if inputs is None:
        raise gr.Error("No audio file submitted! Please upload or record an audio file before submitting your request.")

    text = pipe(inputs, batch_size=BATCH_SIZE, generate_kwargs={"task": task}, return_timestamps=True)["text"]
    return  text

import re

def format_lyrics(text):
    # Remove unwanted subtitle artifacts
    text = re.sub(r"Sous-?titrage Société Radio-Canada", "", text, flags=re.IGNORECASE)

    # Optionally remove repeated whitespace or multiple newlines
    text = re.sub(r'\n+', '\n', text).strip()

    # Use regex to find parts that start with a capital letter and insert a newline
    formatted_text = re.sub(r'(?<!^)([A-Z])', r'\n\1', text)

    # Remove any leading whitespace on each line
    formatted_text = re.sub(r'^[ \t]+', '', formatted_text, flags=re.MULTILINE)

    return formatted_text

    
def infer(audio_input):

    # STEP 1 | Split vocals from the song/audio file
    splt_result = split_process(audio_input, "vocals")
    print(splt_result)

    # STEP 2 | Transcribe 
    # TO-DO : handling errors if JAX demo queue is full
    whisper_result = transcribe(
    				splt_result,	# str (filepath or URL to file) in 'inputs' Audio component
    				"transcribe",	# str  in 'Task' Radio component

    )
    print(whisper_result)

    #return whisper_result[0] # if using JAX

    lyrics = format_lyrics(whisper_result)

    print(lyrics)

    return splt_result, lyrics

css = """
#col-container {max-width: 510px; margin-left: auto; margin-right: auto;}
"""

with gr.Blocks(css=css) as demo:
    with gr.Column(elem_id="col-container"):
        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>""")
        song_in = gr.Audio(label="Song input", type="numpy", sources="upload")
        getlyrics_btn = gr.Button("Get Lyrics !")
        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])

demo.queue().launch()