Upload ./app.py with huggingface_hub

app.py

@@ -81,6 +81,13 @@ class Utility(object):
       repo_type="space")
     return
 
+  # Hugging face : Login to Hugging face
+  def _login_hface(self):
+    huggingface_hub.login(self._decrypt_it(self._huggingface_key),
+      add_to_git_credential=True) # non-blocking login
+    self._ph()
+    return
+
   # System Info : Fetch available CPU and RAM of the system
   def fetch_system_info(self):
     s=''
@@ -150,14 +157,14 @@ class Utility(object):
     y = f.encrypt(p)
     return y
 
-  # Capitalize : Capitalizes the first letter of each word in a list.
+  # Capitalizes the first letter of each word in a list.
   def capitalize_first_letter(self, list_of_words):
     capitalized_words = []
     for word in list_of_words:
       capitalized_word = word[0].upper() + word[1:]
       capitalized_words.append(capitalized_word)
     return capitalized_words
-
+  
 # Add method to class
 def add_method(cls):
   def decorator(func):
@@ -171,16 +178,15 @@ def add_method(cls):
 """ This file contains multiple Python classes and responssible to provide Emotions based on the given user input
     Currently it supports emotions like Anger, Joy, Optimism and Sadness"""
 
-from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline
+from transformers import AutoTokenizer, AutoModelForSequenceClassification
 from  matplotlib.colors import LinearSegmentedColormap
 import scipy
 import scipy.special
 import pandas
+import whisper
 
 class SentimentAnalyser(object):
 
-  global utility
-
   # initialize the object
   def __init__(self, name="Sentiment",*args, **kwargs):
     super(SentimentAnalyser, self).__init__(*args, **kwargs)
@@ -195,9 +201,9 @@ class SentimentAnalyser(object):
     utility._ph()
     print(utility.fetch_system_info())
     utility._ph()
-    print(utility.fetch_gpu_info())
+    # print(utility.fetch_gpu_info())
     utility._ph()
-    print(utility.fetch_host_ip())
+    # print(utility.fetch_host_ip())
     utility._ph()
     self._init_model()
     utility._login_hface()
@@ -205,9 +211,10 @@ class SentimentAnalyser(object):
 
   # initalise the model
   def _init_model(self):
-    modelLink = "bhadresh-savani/distilbert-base-uncased-emotion"
+    modelLink = "bhadresh-savani/distilbert-base-uncased-emotion" #"SamLowe/roberta-base-go_emotions"
     self.tokenizer = AutoTokenizer.from_pretrained(modelLink)
     self.model = AutoModelForSequenceClassification.from_pretrained(modelLink)
+    self.whisper_model = whisper.load_model("small")
     return
 
 sentiment = SentimentAnalyser(name="EmotionAnalyser")
@@ -218,16 +225,15 @@ def _predict_sentiment(p):
   inputs = sentiment.tokenizer(p, return_tensors="pt")
   # Pass inputs through model
   outputs = sentiment.model(**inputs)
-  print(outputs)
+  sentiment_map = sentiment.utility.capitalize_first_letter(sentiment.model.config.label2id.keys())
   out_data = outputs[0][0]
   scores = out_data.detach().numpy()
-  print(out_data)
   scores = scipy.special.softmax(scores)
-  sentiment_map = sentiment.utility.capitalize_first_letter(sentiment.model.config.label2id.keys())
+  # sentiment_map = ['Sadness', 'Joy', 'Love', 'Anger', 'Fear' , "Surprise"]
   df_out = pandas.DataFrame([scores], columns=sentiment_map)
   df_out = df_out[['Love' , 'Joy', 'Surprise' , 'Fear', 'Sadness', 'Anger']]
   return df_out
-
+  
 @add_method(SentimentAnalyser)
 def draw_bar_plot(df_data, title='Sentiment Analysis', xlabel='p string', ylabel='Emotion Score'):
   graphCmap=LinearSegmentedColormap.from_list('gr',["g", "w", "r"])
@@ -239,19 +245,41 @@ def draw_bar_plot(df_data, title='Sentiment Analysis', xlabel='p string', ylabel
   return pic
 
 @add_method(SentimentAnalyser)
-def predict_sentiment(p):
-  df_out = _predict_sentiment(p)
+def inference(audio):
+  audio = whisper.load_audio(audio)
+  audio = whisper.pad_or_trim(audio)
+
+  mel = whisper.log_mel_spectrogram(audio).to(sentiment.whisper_model.device)
+
+  _, probs = sentiment.whisper_model.detect_language(mel)
+
+  options = whisper.DecodingOptions(fp16 = False)
+  result = whisper.decode(sentiment.whisper_model, mel, options)
+
+  print(result.text)
+  return result.text
+  
+@add_method(SentimentAnalyser)
+def predict_sentiment(input_text, input_audio = None, audio_text_btn = None):
+  df_out = _predict_sentiment(input_text)
   max_column = df_out.loc[0].idxmax()
   max_value = df_out.loc[0].max()
   title = f'Sentiment Analysis: {max_column}: {round(max_value*100,1)}%'
-  xlabel= f'Input: {p}'
+  xlabel= f'Input: {input_text}'
   pic = draw_bar_plot(df_out, title=title, xlabel=xlabel)
   return pic.get_figure(), df_out.to_json()
 
 import gradio
-in_box = [gradio.Textbox(lines=1, label="Input", placeholder="type text here")]
+whisper_audio = gradio.Audio(label="Audio Input",
+                    source="microphone",
+                    type="filepath")
+whisper_button = gradio.Button("Convert Audio to Text")
+input_text = gradio.Textbox(lines=1, label="Text Input", placeholder="type text here")
+
+in_box = [input_text,whisper_audio,whisper_button]
 out_box = [gradio.Plot(label="Sentiment Score:"),
   gradio.Textbox(lines=4, label="Raw JSON Response:")]
+
 title = "Sentiment Analysis: Understanding the Emotional Tone of Text"
 desc = "Sentiment analysis is a powerful tool that can be used to gain insights into how people feel about the world around them."
 exp = [
@@ -260,10 +288,13 @@ exp = [
  ]
 arti= "<b>DistilBERT is 27 times faster than OpenAI, making it the clear winner for speed-sensitive applications.</b>\n\nWe did a comparision of OpenAI vs DestilBert model (which we are currently using in this space) by running 31 sentences in a loop and found DestilBert is 27 times faster than OpenAI."
 
-gradio.Interface(fn=predict_sentiment,
+with gradio.Interface(fn=predict_sentiment,
   inputs=in_box,
   outputs=out_box,
   title=title,
   description=desc,
   examples=exp,
-  article=arti).launch(debug=True)
+  article=arti) as demo:
+    with gradio.Blocks() as block:
+      whisper_button.click(inference, inputs=[whisper_audio], outputs=[input_text])
+demo.launch(debug=True)