app.py

import os
#import numpy as np
import gradio as gr
import whisper
import requests 
import tempfile
from neon_tts_plugin_coqui import CoquiTTS

# Whisper: Speech-to-text
model = whisper.load_model("base")

# LLM : Bloom as inference
API_URL = "https://api-inference.huggingface.co/models/bigscience/bloom"
HF_TOKEN = os.environ["HF_TOKEN"]
headers = {"Authorization": f"Bearer {HF_TOKEN}"}
#Language covered in Bloom : en, fr, esp, arb, hn, portu, Indonesian, Vietnamese, Chinese, tamil, telugu, bengali

# Text-to-Speech
LANGUAGES = list(CoquiTTS.langs.keys())
print(f"Languages for Coqui are: {LANGUAGES}")
#Languages for Coqui are: ['en', 'es', 'fr', 'de', 'pl', 'uk', 'ro', 'hu', 'el', 'bg', 'nl', 'fi', 'sl', 'lv', 'ga']
coquiTTS = CoquiTTS()

# Whisper - speeech-to-text
def whisper_stt(audio):
  # load audio and pad/trim it to fit 30 seconds
  audio = whisper.load_audio(audio)
  audio = whisper.pad_or_trim(audio)
  
  # make log-Mel spectrogram and move to the same device as the model
  mel = whisper.log_mel_spectrogram(audio).to(model.device)
  
  # detect the spoken language
  _, probs = model.detect_language(mel)
  print(f"Detected language: {max(probs, key=probs.get)}")
  
  # decode the audio
  options = whisper.DecodingOptions(fp16 = False)
  result = whisper.decode(model, mel, options)
  
  # print the recognized text
  print(f"transcript is : {result.text}")
  return result.text

# Driver function
def fun_engine(audio) : 
  text1 = whisper_stt(audio)
  #text1 = model.transcribe(audio)["text"]
  text2 = lang_model_response(text1)
  speech = tts(text2, 'en')
  return text1, text2, speech

# LLM - Bloom Response  
def lang_model_response(prompt): 
  print(f"*****Inside lang_model_response - Prompt is :{prompt}")
  if len(prompt) == 0:
    prompt = """Can you help me please?"""
  
  json_ = {"inputs": prompt,
            "parameters":
            {
          "top_p": 0.90, #0.90 default
          "max_new_tokens": 64,
          "temperature": 1.1, #1.1 default
          "return_full_text": True,
          "do_sample": True,
          }, 
          "options": 
          {"use_cache": True,
          "wait_for_model": True, 
          },}
  response = requests.post(API_URL, headers=headers, json=json_)
  print(f"Response  is : {response}")
  output = response.json()
  print(f"output is : {output}") 
  output_tmp = output[0]['generated_text']
  print(f"output_tmp is: {output_tmp}")
  solution = output_tmp.split(".")[1]   
  print(f"Final response after splits is: {solution}")
  return solution

# Coqui - Text-to-Speech
def tts(text, language):
    with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as fp:
        coquiTTS.get_tts(text, fp, speaker = {"language" : language})
        return fp.name

  
gr.Interface(
    title = 'Testing Whisper', 
    fn=fun_engine, 
    inputs=[
        gr.Audio(source="microphone",  type="filepath"), #streaming = True,
       # "state"
    ],
    outputs=[
        "textbox",  "textbox", "audio",
    ],
    live=True).launch()