import gradio as gr
import urllib.request
import re
import whisper
import os
import requests
# Whisper: Speech-to-text
model = whisper.load_model("base")
model_med = whisper.load_model("medium")
# Whisper - speech-to-text
def whisper_stt(audio):
print("Inside Whisper TTS")
# 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)
lang = max(probs, key=probs.get)
print(f"Detected language: {max(probs, key=probs.get)}")
# decode the audio
options_transc = whisper.DecodingOptions(fp16 = False, language=lang, task='transcribe') #lang
options_transl = whisper.DecodingOptions(fp16 = False, language='en', task='translate') #lang
result_transc = whisper.decode(model_med, mel, options_transc)
result_transl = whisper.decode(model_med, mel, options_transl)
# print the recognized text
print(f"transcript is : {result_transc.text}")
print(f"translation is : {result_transl.text}")
return result_transc.text, result_transl.text, lang
def engine(audio):
# Get voice query to text
transcribe, translation, lang = whisper_stt(audio)
# Get Video result
video_url, video_id = get_youtube_video(translation)
# Display video
html_out1, html_out2 = display_vid(video_id)
return html_out1, html_out2, translation
def get_youtube_video(query):
search_term = '+'.join(query.split())
html = urllib.request.urlopen("https://www.youtube.com/results?sp=mAEB&search_query="+search_term)
# samples : "https://www.youtube.com/results?sp=mAEB&search_query=openai+whisper+model"
# samples : "https://www.youtube.com/results?sp=mAEB&search_query=gradio+for+ml+model+frontend"
video_ids = re.findall(r"watch\?v=(\S{11})", html.read().decode())
# samples : https://www.youtube.com/watch?v=HbY51mVKrcE
video_url = "https://www.youtube.com/watch?v="+video_ids[0]
return video_url, video_ids
def display_vid(video_ids):
print("******** display_vid ********")
# created embedding
html_out1 = ''
html_out2 = ''
print(f"html output 1 is : {html_out1}")
print(f"html output 2 is : {html_out2}")
return html_out1, html_out2
# not using currently
def set_example_question(sample_question):
print(f"******* Inside Sample Questions ********")
print(f"Sample Question coming from Radio box is : {sample_question}")
print("What is the Return value : {gr.Radio.update(value=sample_question)}")
return gr.Radio.update(value=sample_question) #input_ques.update(example)
demo = gr.Blocks()
with demo:
gr.Markdown("
Voice to Youtube Search
")
gr.Markdown(
"""
Whisper powered voice search app to search and display Youtube Videos on Spaces.
Best Part you don't have to step away from Spaces !
"""
)
with gr.Column():
with gr.Row():
in_audio = gr.Audio(source="microphone", type="filepath", label='Record your voice search query here in English, Spanish or French for best results-')
b1 = gr.Button("Whisper powered Youtube search")
out_textbox = gr.Textbox(label="Whisper Specch to Text transcript")
with gr.Row():
output_vid1 = gr.HTML(label="First Video Result from Youtube", show_label=True)
output_vid2 = gr.HTML(label="Second Video Result from Youtube", show_label=True)
b1.click(engine, inputs=[in_audio], outputs=[output_vid1, output_vid2, out_textbox])
with gr.Row():
gr.Markdown("")
demo.launch(enable_queue=True, debug=True)