app.py

import torch
from transformers import SpeechEncoderDecoder, Wav2Vec2Processor
import gradio as gr
import scipy.signal as sps

def read_file(wav):
    sample_rate, signal = wav                                                                                                                        
    signal = signal.mean(-1)                                                                                                                              
    number_of_samples = round(len(signal) * float(16000) / sample_rate)                                                                                   
    resampled_signal = sps.resample(signal, number_of_samples)
    return resampled_signal
    
def parse_transcription(wav_file):
    speech = read_file(wav_file)
    input_values = processor(speech, sampling_rate=16_000, return_tensors="pt").input_values
    logits = model(input_values).logits
    predicted_ids = torch.argmax(logits, dim=-1)
    transcription = processor.decode(predicted_ids[0], skip_special_tokens=True)
    return transcription
    
    
processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-xls-r-300m-en-to-15", use_auth_token="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")
model = SpeechEncoderDecoder.from_pretrained("facebook/wav2vec2-xls-r-300m-en-to-15", use_auth_token="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")
    

#input_ = gr.inputs.Audio(source="microphone", type="file") 
input_ = gr.inputs.Audio(source="microphone", type="numpy") 
gr.Interface(parse_transcription, inputs = input_,  outputs="text", 
             analytics_enabled=False, show_tips=False, enable_queue=True).launch(inline=False);