Create app.py

app.py

@@ -0,0 +1,82 @@
+import streamlit as st
+from transformers import WhisperForConditionalGeneration, WhisperProcessor
+from transformers import pipeline
+import librosa
+import torch
+from spleeter.separator import Separator
+from pydub import AudioSegment
+from IPython.display import Audio
+import os
+import accelerate
+
+
+# load song
+output_file = "/content/"
+
+# preprocess and crop audio file
+def audio_preprocess():
+    # separate music and vocal
+    separator = Separator('spleeter:2stems')
+    separator.separate_to_file(input_file, output_file)
+
+    # Crop the audio
+    start_time = 60000  # e.g. 30 seconds, 30000
+    end_time = 110000  # e.g. 40 seconds, 40000
+
+    audio = AudioSegment.from_file('/content/test1/vocals.wav')
+    cropped_audio = audio[start_time:end_time]
+    cropped_audio.export('/content/cropped_vocals.wav', format='wav') # save vocal audio file
+
+
+# ASR transcription
+def asr_model():
+    # load audio file
+    y, sr = librosa.load('cropped_vocals.wav', sr=16000)
+
+    # ASR model
+    MODEL_NAME = "RexChan/ISOM5240-whisper-small-zhhk_1"
+    processor = WhisperProcessor.from_pretrained(MODEL_NAME)
+    model = WhisperForConditionalGeneration.from_pretrained(MODEL_NAME, low_cpu_mem_usage=True)
+
+    model.config.forced_decoder_ids = None
+    model.config.suppress_tokens = []
+    model.config.use_cache = False
+
+    processed_in = processor(y, sampling_rate=sr, return_tensors="pt")
+    gout = model.generate(
+        input_features=processed_in.input_features,
+        output_scores=True, return_dict_in_generate=True
+    )
+    transcription = processor.batch_decode(gout.sequences, skip_special_tokens=True)[0]
+
+    # print result
+    print(f"Song lyrics = {transcription}")
+
+    return transcription
+
+
+# sentiment analysis
+def senti_model(transcription):
+
+    pipe = pipeline("text-classification", model="lxyuan/distilbert-base-multilingual-cased-sentiments-student")
+    final_result = pipe(transcription)
+    print(f"Sentiment Analysis shows that this song is {final_result[0]['label']}. Confident level of this analysis is {final_result[0]['score']*100:.1f}%.")
+
+    return final_result
+
+
+# main
+def main():
+
+    input_file = st.file_uploader("upload a song in mp3 format", type="mp3")
+    if input_file is not None:
+      st.write("File uploaded successfully!")
+    else:
+      st.write("No file uploaded.")
+    audio_preprocess()
+    transcription = asr_model()
+    final_result = senti_model(transcription)
+
+
+if __name__ == '__main__':
+    main()