app.py

import streamlit as st
import torch
from transformers import WhisperProcessor, WhisperForConditionalGeneration
from peft import PeftModel
import numpy as np
import pyaudio

# Tải mô hình
@st.cache_resource
def load_model():
    base_model_id = "openai/whisper-tiny"
    adapter_id = "longhoang2112/whisper-turbo-fine-tuning-adapters"
    
    processor = WhisperProcessor.from_pretrained(base_model_id)
    model = WhisperForConditionalGeneration.from_pretrained(base_model_id)
    try:
        model = PeftModel.from_pretrained(model, adapter_id)
        model.set_active_adapters(adapter_id)
    except:
        st.warning("Adapter loading failed. Using base model.")
    
    device = "cuda" if torch.cuda.is_available() else "cpu"
    model.to(device)
    return processor, model, device

processor, model, device = load_model()

# Ghi âm
def record_audio(duration=5, sample_rate=16000):
    CHUNK = 1024
    FORMAT = pyaudio.paFloat32
    CHANNELS = 1
    p = pyaudio.PyAudio()
    stream = p.open(format=FORMAT, channels=CHANNELS, rate=sample_rate, input=True, frames_per_buffer=CHUNK)
    
    st.write(f"Đang ghi âm... ({duration} giây)")
    frames = []
    for _ in range(0, int(sample_rate / CHUNK * duration)):
        data = stream.read(CHUNK)
        frames.append(np.frombuffer(data, dtype=np.float32))
    
    stream.stop_stream()
    stream.close()
    p.terminate()
    return np.concatenate(frames), sample_rate

# Giao diện
st.title("Whisper Turbo với Adapter")
duration = st.slider("Thời gian ghi âm (giây):", 1, 10, 5)

if st.button("Ghi âm"):
    audio, sample_rate = record_audio(duration)
    input_features = processor(audio, sampling_rate=sample_rate, return_tensors="pt").input_features.to(device)
    with torch.no_grad():
        predicted_ids = model.generate(input_features)
    transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
    st.write("**Kết quả:**", transcription)