from sclib import SoundcloudAPI, Track
from transformers import pipeline
import gradio as gr
import numpy as np
import librosa
import os

transcriber = pipeline("automatic-speech-recognition", model="ID2223/whisper-small-swedish")

# do not pass a Soundcloud client ID that did not come from this library, but you can save a client_id that this lib found and reuse it
api = SoundcloudAPI()  

def get_audio(url: str):
    filename = url_to_file(url)
    audio_original, sr = librosa.load(filename, sr=16000)
    audio_original = audio_original[:30*sr]
    os.remove(filename)
    return (sr, audio_original)

def url_to_file(url: str) -> str:
    track = api.resolve(url)
    assert type(track) is Track
    filename = "to_transcribe.mp3"
    with open(filename, 'wb+') as file:
        track.write_mp3_to(file)
    return filename

def transcribe(url: str):
    sr, audio_original = get_audio(url)
    audio = audio_original.astype(np.float32)
    audio /= np.max(np.abs(audio))
    return ((sr, audio_original), transcriber({"sampling_rate": sr, "raw": audio})["text"])


demo = gr.Interface(
    transcribe,
    inputs = [gr.Textbox(label="Soundcloud URL", value="https://soundcloud.com/user-668319862/sa-gar-borsen-2024")],
    outputs = ["audio", "text"],
    title="Whisper Soundcloud",
)

demo.launch()