app.py

import gradio as gr
from iman.sad_tfpy10 import *
from autosub import SpeechRecognizer
from autosub import GOOGLE_SPEECH_API_KEY
import soundfile as sf
import io

css = """ 
 textarea { direction: rtl; text-align: right; font-family: Calibri, sans-serif; font-size: 16px;} 
"""


recognizer = SpeechRecognizer(language="fa", rate=16000,api_key=GOOGLE_SPEECH_API_KEY, proxies=None)


seg = Segmenter(ffmpeg_path="ffmpeg",model_path="keras_speech_music_noise_cnn.hdf5" , device="cpu",vad_type="vad")



def process_segment(args):
 segment, wav = args
 start, stop = segment
 # pp = converter((start, stop))
 pp = pcm_to_flac(wav[int(start*16000) : int(stop*16000)])
 tr_beamsearch_lm = recognizer(pp) 
 return start, stop, tr_beamsearch_lm

def pcm_to_flac(pcm_data, sample_rate=16000):
 buffer = io.BytesIO()
 sf.write(buffer, pcm_data, sample_rate, format='FLAC')
 flac_data = buffer.getvalue() 
 return flac_data

    
def transcribe_audio(audio_file):

    text=""
    isig,wav =  seg(audio_file)
    isig = filter_output(isig , max_silence=0.5 ,ignore_small_speech_segments=0.1 , max_speech_len=15 ,split_speech_bigger_than=20)   
    isig = [(a,b) for x,a,b,_,_ in isig]
    print(isig)
    results=[]
    for segment in isig:
         results.append (process_segment((segment, wav)))
    for start, stop, tr_beamsearch_lm in results:
          
          try:
              
              text += ' ' + tr_beamsearch_lm + '\r\n'
              print(start)
              print(stop)
              print(text)
          except:
               pass
    
    return text

# Define the Gradio interface
interface = gr.Interface(
    fn=transcribe_audio,
    inputs=gr.Audio(type="filepath"),  

    outputs=gr.Textbox(label="Transcription", elem_id="output-text",interactive=True),
    title="Persian Audio Transcription",
    description="Upload an audio file or record audio to get the transcription.",
    css=css
)

# Launch the Gradio app
interface.launch()