app.py

import os
import duckdb
import streamlit as st
from huggingface_hub import hf_hub_download
import pandas as pd
import tempfile
import re

HF_REPO_ID = "stcoats/temp-duckdb-upload"
HF_FILENAME = "ycsep.duckdb"
LOCAL_PATH = "./ycsep.duckdb"

st.set_page_config(layout="wide")
st.title("YCSEP Audio Dataset Viewer")

# Download database if missing
if not os.path.exists(LOCAL_PATH):
    st.write("Downloading from HF Hub...")
    hf_hub_download(
        repo_id=HF_REPO_ID,
        repo_type="dataset",
        filename=HF_FILENAME,
        local_dir=".",
        local_dir_use_symlinks=False
    )
    st.success("Download complete.")

# Connect (only once)
@st.cache_resource(show_spinner=False)
def get_duckdb_connection():
    return duckdb.connect(LOCAL_PATH, read_only=True)

try:
    con = get_duckdb_connection()
    st.success("Connected to DuckDB.")
except Exception as e:
    st.error(f"DuckDB connection failed: {e}")
    st.stop()

# Search
query = st.text_input("Search text (case-insensitive)", "").strip()

if query:
    sql = """
        SELECT id, channel, video_id, video_title, speaker, start_time, end_time, text, pos_tags, upload_date, audio
        FROM data
        WHERE LOWER(text) LIKE LOWER(?)
        LIMIT 100
    """
    df = con.execute(sql, [f"%{query}%"]).df()
else:
    df = con.execute("""
        SELECT id, channel, video_id, video_title, speaker, start_time, end_time, text, pos_tags, upload_date, audio
        FROM data
        LIMIT 100
    """).df()

st.markdown(f"### Showing {len(df)} results")

if len(df) == 0:
    st.warning("No matches found.")

# Show table with inline audio players
for i, row in df.iterrows():
    col1, col2, col3 = st.columns([3, 5, 2])

    col1.markdown(f"**ID:** {row['id']}")
    col1.markdown(f"**Channel:** {row['channel']}")
    col1.markdown(f"**Video ID:** {row['video_id']}")
    col1.markdown(f"**Video Title:** {row['video_title']}")
    col1.markdown(f"**Speaker:** {row['speaker']}")
    col1.markdown(f"**Start Time:** {row['start_time']}")
    col1.markdown(f"**End Time:** {row['end_time']}")
    col1.markdown(f"**Upload Date:** {row['upload_date']}")

    highlighted_text = row['text']
    if query:
        highlighted_text = re.sub(f'({re.escape(query)})', r'<mark>\1</mark>', highlighted_text, flags=re.IGNORECASE)

    col2.markdown(f"**Text:** {highlighted_text}", unsafe_allow_html=True)
    col2.markdown(f"**POS tags:** {row['pos_tags']}")

    audio_data = row["audio"]
    try:
        if isinstance(audio_data, (bytes, bytearray, memoryview)):
            audio_bytes = bytes(audio_data)
        elif isinstance(audio_data, list):  # DuckDB sometimes gives list[int]
            audio_bytes = bytes(audio_data)
        else:
            audio_bytes = None

        if audio_bytes:
            with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmpfile:
                tmpfile.write(audio_bytes)
                tmpfile.flush()
                col3.audio(tmpfile.name, format="audio/mp3")
        else:
            col3.warning("Audio missing or invalid format.")
    except Exception as e:
        col3.error(f"Audio error: {e}")