import streamlit as st
from transformers import pipeline
import torch

# -------------------------
# Config / Model names
# -------------------------
EMOTION_MODEL = "cardiffnlp/twitter-roberta-base-emotion"
SENTIMENT_MODEL = "cardiffnlp/twitter-xlm-roberta-base-sentiment"

# -------------------------
# Helpers
# -------------------------
def map_emotion_to_tts_label(emotion_label: str) -> str:
    e = emotion_label.lower()
    if e in {"joy", "happiness", "happy", "amusement", "excited", "excitement", "optimism"}:
        return "happy / energetic"
    if e in {"sadness", "sad", "grief", "disappointed", "disappointment", " melancholy"}:
        return "sad / soft / calm"
    if e in {"anger", "angry", "annoyance", "annoyed", "disgust"}:
        return "angry / intense"
    if e in {"fear", "scared", "nervous", "anxious"}:
        return "scared / tense"
    if e in {"surprise", "surprised"}:
        return "surprised / alert"
    return "neutral / plain"

def map_sentiment_to_tts_label(sentiment_label: str) -> str:
    s = sentiment_label.lower()
    if s == "positive":
        return "positive / warm"
    if s == "negative":
        return "negative / firm"
    return "neutral / plain"

# -------------------------
# Load pipelines (cached)
# -------------------------
@st.cache_resource(show_spinner=False)
def load_pipelines():
    device = 0 if torch.cuda.is_available() else -1

    emotion_pipe = pipeline(
        "text-classification",
        model=EMOTION_MODEL,
        top_k=None,
        device=device
    )

    sentiment_pipe = pipeline(
        "text-classification",
        model=SENTIMENT_MODEL,
        top_k=None,
        device=device
    )

    return emotion_pipe, sentiment_pipe

# -------------------------
# Streamlit UI
# -------------------------
st.set_page_config(page_title="Emotion + Tone Detector", page_icon="👀", layout="centered")
st.title("👀 Emotion & Tone Detector — English / Spanish / French")
st.write(
    "Type a sentence (English / Spanish / French) and click **Analyze**. "
    "Shows emotion + tone + suggested TTS style."
)

emotion_pipe, sentiment_pipe = load_pipelines()

text = st.text_area("✍️ Enter sentence here", height=140, placeholder="Type in English, Spanish, or French...")

if st.button("Analyze"):
    if not text or not text.strip():
        st.warning("Please enter a sentence to analyze.")
    else:
        with st.spinner("Analyzing..."):
            try:
                # Emotion
                emotion_results = emotion_pipe(text, top_k=None)
                if isinstance(emotion_results, dict):
                    emotion_results = [emotion_results]
                emotion_results_sorted = sorted(emotion_results, key=lambda x: x["score"], reverse=True)
                top_emotion = emotion_results_sorted[0]["label"]
                top_emotion_score = emotion_results_sorted[0]["score"]

                # Tone/Sentiment
                sentiment_results = sentiment_pipe(text, top_k=None)
                if isinstance(sentiment_results, dict):
                    sentiment_results = [sentiment_results]
                sentiment_sorted = sorted(sentiment_results, key=lambda x: x["score"], reverse=True)
                top_tone = sentiment_sorted[0]["label"]
                top_tone_score = sentiment_sorted[0]["score"]

                # Map for TTS
                tts_from_emotion = map_emotion_to_tts_label(top_emotion)
                tts_from_tone = map_sentiment_to_tts_label(top_tone)

                # Show results
                st.subheader("🎭 Detected Emotion")
                st.write(f"**{top_emotion}** (confidence **{top_emotion_score:.2f}**)")

                st.subheader("🎵 Detected Tone (Sentiment)")
                st.write(f"**{top_tone}** (confidence **{top_tone_score:.2f}**)")

                st.subheader("🔊 Suggested TTS style (from emotion)")
                st.write(tts_from_emotion)

                st.subheader("🔊 Suggested TTS style (from tone)")
                st.write(tts_from_tone)

                st.subheader("📊 Full emotion scores")
                st.table([{ "label": r["label"], "score": f"{r['score']:.3f}"} for r in emotion_results_sorted])

                st.subheader("📊 Full tone (sentiment) scores")
                st.table([{ "label": r["label"], "score": f"{r['score']:.3f}"} for r in sentiment_sorted])

            except Exception as err:
                st.error("Error during analysis.")
                st.exception(err)