app.py

import gradio as gr
from transformers import pipeline
import torch
import librosa
import json

max_duration = int(30 * 16000)

def load_model(model_name = "cawoylel/windanam_mms-1b-tts_v2"):
  """
  Function to load model from hugging face
  """
  pipe = pipeline("automatic-speech-recognition", model="cawoylel/windanam_mms-1b-tts_v2")
  return pipe

pipeline = load_model()

def transcribe_audio(sample):
  """
  Transcribe audio
  """
  transcription = pipeline(sample)
  return transcription["text"]

def transcribe(audio_file_mic=None, audio_file_upload=None):
    if audio_file_mic:
        audio_file = audio_file_mic
    elif audio_file_upload:
        audio_file = audio_file_upload
    else:
        return "Please upload an audio file or record one"

    # Make sure audio is 16kHz
    speech, sample_rate = librosa.load(audio_file)
    if sample_rate != 16000:
        speech = librosa.resample(speech, orig_sr=sample_rate, target_sr=16000)
    duration = librosa.get_duration(y=speech, sr=16000)
    if duration > 30:
        speech = speech[:max_duration]
    return transcribe_audio(speech)


description = '''Automatic Speech Recognition with [MMS](https://ai.facebook.com/blog/multilingual-model-speech-recognition/) (Massively Multilingual Speech) by Meta.
Supports [1162 languages](https://dl.fbaipublicfiles.com/mms/misc/language_coverage_mms.html). Read the paper for more details: [Scaling Speech Technology to 1,000+ Languages](https://arxiv.org/abs/2305.13516).'''

iface = gr.Interface(fn=transcribe,
                     inputs=[
                         gr.Audio(source="microphone", type="filepath", label="Record Audio"),
                         gr.Audio(source="upload", type="filepath", label="Upload Audio"),
                         ],
                     outputs=gr.Textbox(label="Transcription"),
                     description=description
                     )
iface.launch()