updated application file

app.py

@@ -10,6 +10,7 @@ asr = pipeline('automatic-speech-recognition', model='facebook/wav2vec2-large-96
 tokenizer = RobertaTokenizerFast.from_pretrained("arpanghoshal/EmoRoBERTa")
 model = TFRobertaForSequenceClassification.from_pretrained("arpanghoshal/EmoRoBERTa")
 emo = pipeline('sentiment-analysis', model='arpanghoshal/EmoRoBERTa')
+pos = pipeline("token-classification", model="vblagoje/bert-english-uncased-finetuned-pos")
 
 def transcribe_and_describe(audio):
 
@@ -17,9 +18,13 @@ def transcribe_and_describe(audio):
 
   text = asr(audio)['text']
 
+  tagged_text = pos(text)
+  filler_words = [entry['word'] for entry in tagged_text if entry['entity'] == 'INTJ']
+  filler_word_pr =  len(filler_words) / len(tagged_text)
+
   flatness = pd.DataFrame(librosa.feature.spectral_flatness(y=audio).T).describe().T
   loudness = pd.DataFrame(librosa.feature.rms(audio).T).describe().T
-  time, frequency, confidence, activation = crepe.predict(audio, sr)
+  time, frequency, confidence, activation = crepe.predict(audio, sr, viterbi=True)
   frequency = pd.DataFrame(frequency.T).describe().T
 
   mean_spectral_flatness = flatness.loc[0, 'mean'] 
@@ -33,13 +38,14 @@ def transcribe_and_describe(audio):
 
   emotion = emo(text)[0]['label']
 
-  return (text, words_per_minute, mean_pitch, pitch_std, mean_volume, volume_std, mean_spectral_flatness, spectral_flatness_std,  emotion)
+  return (text, f"{filler_word_pr:.2f}", f"{words_per_minute:.2f}", f"{mean_pitch:.2f}", f"{pitch_std:.2f}", f"{mean_volume:.2f}", f"{volume_std:.2f}", f"{mean_spectral_flatness:.2f}", f"{spectral_flatness_std:.2f}",  emotion)
 
 gr.Interface(
     fn=transcribe_and_describe, 
     inputs=gr.Audio(source="microphone", type="filepath"), 
     outputs=[
         gr.Text(label="Transcription"),
+        fr.Text(label="Filler Word Percent"),
         gr.Text(label="Rate of Speech (WPM)"), 
         gr.Text(label="Mean Pitch (Hz)"), 
         gr.Text(label="Pitch Variation (Hz)"),