import torch
from pydub import AudioSegment
import soundfile as sf
import os

def preprocess_audio(input_audio_path, output_audio_path):
    """
    Converts audio to 16kHz WAV format.
    """
    audio = AudioSegment.from_file(input_audio_path)
    audio = audio.set_frame_rate(16000).set_channels(1)
    audio.export(output_audio_path, format="wav")
    return output_audio_path

def split_audio(audio_path, chunk_length_ms=30000):
    """
    Splits audio into chunks of specified length.
    """
    audio = AudioSegment.from_file(audio_path)
    return [audio[i : i + chunk_length_ms] for i in range(0, len(audio), chunk_length_ms)]

def transcribe_chunk(audio_chunk, chunk_index, whisper_models):
    """
    Transcribes a single audio chunk using the pre-loaded Whisper model.
    """
    temp_path = f"temp_chunk_{chunk_index}.wav"
    audio_chunk.export(temp_path, format="wav")

    audio, _ = sf.read(temp_path)
    inputs = whisper_models["processor"](audio, sampling_rate=16000, return_tensors="pt")
    input_features = inputs.input_features.to(whisper_models["device"])

    predicted_ids = whisper_models["model"].generate(input_features)
    transcription = whisper_models["processor"].batch_decode(predicted_ids, skip_special_tokens=True)[0]

    os.remove(temp_path)
    return transcription

def speech_to_text_long(audio_path, whisper_models):
    """
    Transcribes a long audio file by splitting it into chunks.
    """
    processed_audio_path = "processed_audio.wav"
    preprocess_audio(audio_path, processed_audio_path)

    chunks = split_audio(processed_audio_path)
    transcriptions = [transcribe_chunk(chunk, idx, whisper_models) for idx, chunk in enumerate(chunks)]

    os.remove(processed_audio_path)
    return " ".join(transcriptions)