app.py

import os
import duckdb
import streamlit as st
from huggingface_hub import hf_hub_download
import pandas as pd
import tempfile
import base64

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):
    with st.spinner("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, exact substring match)", "").strip()

# Escape single quotes in query
query_safe = query.replace("'", "''")

# Build query using exact substring match only (case-insensitive handled by ILIKE)
if query:
    sql = f"""
        SELECT id, channel, video_id, speaker, start_time, end_time, upload_date, text, pos_tags, audio
        FROM data
        WHERE text ILIKE '%{query_safe}%'
        LIMIT 100
    """
    df = con.execute(sql).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 get_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 ""
            b64 = base64.b64encode(data).decode("utf-8")
            return f'<audio controls preload="metadata" style="height:20px;width:120px;"><source src="data:audio/mp3;base64,{b64}" type="audio/mpeg"></audio>'
        except Exception:
            return ""

    df["Audio"] = df["audio"].apply(get_audio_html)
    df.drop(columns=["audio"], inplace=True)

    # Reorder columns so "Audio" is last and limit column width for id
    display_cols = ["id", "channel", "video_id", "speaker", "start_time", "end_time", "upload_date", "text", "pos_tags", "Audio"]
    df = df[display_cols]

    # Show table with HTML rendering for audio
    st.markdown("### Results Table (Sortable with Audio Column)")
    st.write("(Scroll right to view audio controls)")
    st.write(df.to_html(escape=False, index=False), unsafe_allow_html=True)