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()

# Enable full-text search index on first run (one-time setup if not exists)
con.execute("PRAGMA create_fts_index('data', 'text')")

# 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 text % ?
        LIMIT 100
    """
    df = con.execute(sql, [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.")
else:
    def render_audio_cell(audio_bytes):
        try:
            if isinstance(audio_bytes, (bytes, bytearray, memoryview)):
                data = bytes(audio_bytes)
            elif isinstance(audio_bytes, list):
                data = bytes(audio_bytes)
            else:
                return None
            with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmp:
                tmp.write(data)
                tmp.flush()
                return tmp.name
        except Exception:
            return None

    df["audio_file"] = df["audio"].apply(render_audio_cell)
    df_display = df.drop(columns=["audio"]).copy()

    # Add HTML audio tag column
    def audio_html(path):
        if path:
            return f'<audio controls preload="none" style="height:20px;"> <source src="file://{path}" type="audio/mpeg"> </audio>'
        return ""

    df_display["Audio"] = df["audio_file"].apply(audio_html)

    # Reorder columns
    column_order = ["id", "channel", "video_id", "video_title", "speaker", "start_time", "end_time", "upload_date", "text", "pos_tags", "Audio"]
    df_display = df_display[column_order]

    st.markdown("### Full Table View (Sortable)")
    st.write("Note: Audio is embedded using HTML tags; not all browsers allow playback from local temp paths.")
    st.dataframe(df_display.drop(columns=["Audio"]))

    st.markdown("### Audio Previews")
    for i, row in df_display.iterrows():
        if row["Audio"]:
            st.markdown(f"**{row['speaker']} | {row['text'][:80]}**", unsafe_allow_html=True)
            st.markdown(row["Audio"], unsafe_allow_html=True)