yashsrivastava's picture
Upload app.py
17da9ca
#!/usr/bin/env python
# coding: utf-8
# In[ ]:
import soundfile as sf
import torch
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
import argparse
from glob import glob
import torchaudio
import subprocess
import gradio as gr
resampler = torchaudio.transforms.Resample(48_000, 16_000)
def get_filename(wav_file):
filename_local = wav_file.split('/')[-1][:-4]
filename_new = '/tmp/'+filename_local+'_16.wav'
subprocess.call(["sox {} -r {} -b 16 -c 1 {}".format(wav_file, str(16000), filename_new)], shell=True)
return filename_new
def parse_transcription(wav_file):
# load pretrained model
processor = Wav2Vec2Processor.from_pretrained("jonatasgrosman/wav2vec2-large-xlsr-53-english")
model = Wav2Vec2ForCTC.from_pretrained("jonatasgrosman/wav2vec2-large-xlsr-53-english")
# load audio
wav_file = get_filename(wav_file.name)
audio_input, sample_rate = sf.read(wav_file)
#test_file = resampler(test_file[0])
# pad input values and return pt tensor
input_values = processor(audio_input, sampling_rate=16_000, return_tensors="pt").input_values
# INFERENCE
# retrieve logits & take argmax
logits = model(input_values).logits
predicted_ids = torch.argmax(logits, dim=-1)
# transcribe
transcription = processor.decode(predicted_ids[0], skip_special_tokens=True)
return transcription
# In[ ]:
import gradio as gr
title = "Speech-to-Text-English"
description = "Upload a English audio clip, and let AI do the hard work of transcribing."
gr.Interface(
parse_transcription,
title=title,
inputs=gr.inputs.Audio(label="Record Audio File", type="file", source = "microphone"),
description=description, outputs = "text").launch(inline = False)