app.py

import soundfile as sf
import torch
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
import gradio as gr
import sox


def convert(inputfile, outfile):
    sox_tfm = sox.Transformer()
    sox_tfm.set_output_format(
        file_type="wav", channels=1, encoding="signed-integer", rate=16000, bits=16
    )
    sox_tfm.build(inputfile, outfile)
    

model_name = "indonesian-nlp/wav2vec2-indonesian-javanese-sundanese"
processor = Wav2Vec2Processor.from_pretrained(model_name)
model = Wav2Vec2ForCTC.from_pretrained(model_name)

def parse_transcription(wav_file):
    filename = wav_file.name.split('.')[0]
    convert(wav_file.name, filename + "16k.wav")
    speech, _ = sf.read(filename + "16k.wav")
    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



output = gr.outputs.Textbox(label="Indonesian, Javanese or Sundanese")
    
input_ = gr.inputs.Audio(source="microphone", type="file") 
#gr.Interface(parse_transcription, inputs = input_,  outputs="text", 
#             analytics_enabled=False, show_tips=False, enable_queue=True).launch(inline=False);
             
gr.Interface(parse_transcription, inputs = input_,  outputs=[output],
             analytics_enabled=False,
             show_tips=False,
             theme='huggingface',
             layout='vertical',
             title="Multilingual Speech Recognition for Indonesian Languages",
             description="Speech Recognition Live Demo for Indonesian, Javanese and Sundanese",
             enable_queue=True).launch( inline=False)