app.py

import torch
from transformers import Wav2Vec2ForCTC, Wav2Vec2Tokenizer
import torchaudio

def speech_recognition(audio_file_path):
    tokenizer = Wav2Vec2Tokenizer.from_pretrained("facebook/wav2vec2-base-960h")
    model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-base-960h")

    waveform, sample_rate = torchaudio.load(audio_file_path)

    if sample_rate != 16000:
        resampler = torchaudio.transforms.Resample(orig_freq=sample_rate, new_freq=16000)
        waveform = resampler(waveform)

    input_values = tokenizer(waveform.squeeze().numpy(), return_tensors="pt", padding="longest").input_values
    with torch.no_grad():
        logits = model(input_values).logits

    predicted_ids = torch.argmax(logits, dim=-1)
    transcription = tokenizer.batch_decode(predicted_ids)

    return transcription[0]