app.py

from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC
import torch
import gradio as gr
import librosa 
import os
import subprocess

# Install system dependencies
subprocess.run(["apt-get", "update"], check=True)
subprocess.run(["apt-get", "install", "-y", "espeak"], check=True)

# load model and processor
processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-lv-60-espeak-cv-ft")
model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-lv-60-espeak-cv-ft")
 
# define prediction function
def audio2phoneme(audio_path):
  audio, sr = librosa.load(audio_path, sr=16000)
  input_values = processor(audio, return_tensors="pt", padding=True).input_values
  with torch.no_grad():
    logits = model(input_values).logits
  predicted_ids = torch.argmax(logits, dim=-1)
  transcription = processor.batch_decode(predicted_ids)
  return ' '.join(transcription)

app = gr.Interface(
    fn=audio2phoneme,
    inputs=gr.Audio(sources=["upload","microphone"], type="filepath"),
    outputs=gr.Textbox(label="Phoneme Transcription", show_copy_button=True, show_label=True),
    description="Get phonemes from audio",
    title="Audio to Phoneme Transcription using facebook/wav2vec2-lv-60-espeak-cv",
    )

# start space
app.launch(share=True)