import gradio as gr
import torch
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline
import numpy as np
from pyannote.audio import Pipeline
from dotenv import load_dotenv
import os

load_dotenv()

# Check and set device
device = "cuda:0" if torch.cuda.is_available() else "cpu"
print(f"Using device: {device}")
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32

# Model and pipeline setup
model_id = "distil-whisper/distil-small.en"
model = AutoModelForSpeechSeq2Seq.from_pretrained(
    model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True
)
model.to(device)

processor = AutoProcessor.from_pretrained(model_id)

pipe = pipeline(
    "automatic-speech-recognition",
    model=model,
    tokenizer=processor.tokenizer,
    feature_extractor=processor.feature_extractor,
    max_new_tokens=128,
    torch_dtype=torch_dtype,
    device=device,
)

# diarization pipeline (renamed to avoid conflict)
diarization_pipeline = Pipeline.from_pretrained(
    "pyannote/speaker-diarization-3.0", use_auth_token=os.getenv("HF_KEY")
)


def transcribe(audio):
    sr, data = audio
    processed_data = np.array(data).astype(np.float32) / 32767.0
    waveform_tensor = torch.tensor(processed_data[np.newaxis, :])

    # results from the pipeline
    transcription_res = pipe({"sampling_rate": sr, "raw": processed_data})["text"]
    diarization_res = diarization_pipeline(
        {"waveform": waveform_tensor, "sample_rate": sr}
    )

    return transcription_res, diarization_res


demo = gr.Interface(
    fn=transcribe,
    inputs=gr.Audio(sources=["upload", "microphone"]),
    outputs=[
        gr.Textbox(lines=3, info="audio transcription"),
        gr.Textbox(info="speaker diarization"),
    ],
    title="Automatic Speech Recognition 🗣️",
    description="Transcribe your speech to text with distilled whisper",
)

if __name__ == "__main__":
    demo.launch()