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

# Build query
if query:
    search_terms = query.lower().split()
    conditions = " AND ".join(["LOWER(text) LIKE ?" for _ in search_terms])
    sql = f"""
        SELECT id, channel, video_id, speaker, start_time, end_time, upload_date, text, pos_tags, audio
        FROM data
        WHERE {conditions}
        LIMIT 100
    """
    df = con.execute(sql, [f"%{term}%" for term in search_terms]).df()
else:
    df = con.execute("""
        SELECT id, channel, video_id, speaker, start_time, end_time, upload_date, text, pos_tags, 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_html(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 ""
            with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmp:
                tmp.write(data)
                tmp.flush()
                return f'<audio controls style="height:20px;width:100%;"><source src="file://{tmp.name}" type="audio/mpeg"></audio>'
        except Exception:
            return ""

    df["Audio"] = df["audio"].apply(render_audio_html)
    df = df.drop(columns=["audio"])
    df = df[["id", "channel", "video_id", "speaker", "start_time", "end_time", "upload_date", "text", "pos_tags", "Audio"]]

    from st_aggrid import AgGrid, GridOptionsBuilder

    st.markdown("### Results Table (Sortable with Embedded Audio)")

    gb = GridOptionsBuilder.from_dataframe(df.drop(columns=["Audio"]))
    gb.configure_default_column(resizable=True, sortable=True, filter=True)
    grid_options = gb.build()

    AgGrid(df.drop(columns=["Audio"]), gridOptions=grid_options, fit_columns_on_grid_load=True)

    st.markdown("### Audio Controls in Table")
    for i in range(len(df)):
        st.markdown(df.loc[i, "Audio"], unsafe_allow_html=True)