app.py

import gradio as gr
import os
from moviepy.editor import VideoFileClip
from pydub import AudioSegment
import torch
from nemo.collections.asr.models import EncDecCTCModelBPE  # Adjust based on your model type
import wget

MODEL_URL = "https://huggingface.co/Mohammadp/Persian-ASR/resolve/main/conformer_transducer_persian.nemo"
MODEL_PATH = "conformer_transducer_persian.nemo"

# Download model if it doesn't exist
if not os.path.exists(MODEL_PATH):
    print("Downloading model...")
    wget.download(MODEL_URL, MODEL_PATH)
    print("\nModel downloaded successfully.")

# Load the model
model = EncDecCTCModelBPE.restore_from(MODEL_PATH)
print("Model loaded successfully!")
# Constants
SAMPLE_RATE = 16000
MAX_CHUNK_LENGTH_MS = 10 * 1000  # 10 seconds per chunk

# Helper functions
def extract_audio_from_video(video_path):
    """Extracts audio from a video file and saves it as a WAV file."""
    video = VideoFileClip(video_path)
    audio_path = "extracted_audio.wav"
    video.audio.write_audiofile(audio_path)
    return audio_path

def resample_audio(audio_path, target_sample_rate=SAMPLE_RATE):
    """Resamples an audio file to 16kHz."""
    audio = AudioSegment.from_file(audio_path)
    audio = audio.set_frame_rate(target_sample_rate)
    resampled_path = "resampled_audio.wav"
    audio.export(resampled_path, format="wav")
    return resampled_path

def split_audio(audio_path, max_length_ms=MAX_CHUNK_LENGTH_MS):
    """Splits audio into chunks of max_length_ms each."""
    audio = AudioSegment.from_file(audio_path)
    chunks = []
    for i in range(0, len(audio), max_length_ms):
        chunk = audio[i:i + max_length_ms]
        chunk_path = f"chunk_{i // max_length_ms}.wav"
        chunk.export(chunk_path, format="wav")
        chunks.append(chunk_path)
    return chunks

def transcribe_audio(audio_path):
    """Transcribes a single audio file using the NeMo model."""
    return model.transcribe([audio_path])#[0]

def process_audio(audio_path):
    """Processes an audio file: resamples, splits, and transcribes."""
    resampled_path = resample_audio(audio_path)
    chunks = split_audio(resampled_path)
    transcriptions = [transcribe_audio(chunk) for chunk in chunks]
    return " ".join(transcriptions)

def process_video(video_path):
    """Extracts and processes audio from a video file."""
    audio_path = extract_audio_from_video(video_path)
    return process_audio(audio_path)

def process_microphone(audio_path):
    """Processes live-recorded microphone audio."""
    return process_audio(audio_path)

# Gradio Interface
def process_input(video=None, audio=None, microphone=None):
    if video is not None:
        return f"Transcription: {process_video(video)}"
    elif audio is not None:
        return f"Transcription: {process_audio(audio)}"
    elif microphone is not None:
        return f"Transcription: {process_microphone(microphone)}"
    else:
        return "No input provided."

# ** WAV FILE EXAMPLES ONLY **
example_wav_files = [
    "FEmEC4QBSwA_285%20(4).wav",
]

iface = gr.Interface(
    fn=process_input,
    inputs=[
        gr.Video(label="Upload Video"),
        gr.Audio(label="Upload Audio File", type="filepath"),
        gr.Microphone(label="Record from Microphone", type="filepath")
    ],
    outputs="text",
    title="NeMo ASR Transcription Interface",
    description="Upload a video, an audio file, or record from the microphone to transcribe the audio using a trained NeMo model.",
    examples=[[None, wav, None] for wav in example_wav_files]  # **Only WAV examples**
)

iface.launch()