app.py

import gradio as gr
import librosa
from transformers import AutoFeatureExtractor, AutoTokenizer, SpeechEncoderDecoderModel
    
feature_extractor = AutoFeatureExtractor.from_pretrained("facebook/wav2vec2-xls-r-300m-en-to-15", use_auth_token="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")
tokenizer = AutoTokenizer.from_pretrained("facebook/wav2vec2-xls-r-300m-en-to-15", use_auth_token="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", use_fast=False)
model = SpeechEncoderDecoderModel.from_pretrained("facebook/wav2vec2-xls-r-300m-en-to-15", use_auth_token="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")

def process_audio_file(file):
    data, sr = librosa.load(file)
    if sr != 16000:
        data = librosa.resample(data, sr, 16000)
    print(data.shape)
    input_values = feature_extractor(data, return_tensors="pt").input_values
    return input_values
    
def transcribe(file):
    input_values = process_audio_file(file)
    
    sequences = model.generate(input_values)
    
    transcription = tokenizer.batch_decode(sequences, skip_special_tokens=True)
    return transcription[0]
    
iface = gr.Interface(
    fn=transcribe, 
    inputs=gr.inputs.Audio(source="microphone", type='filepath'),
    outputs="text",
)
iface.launch()