radio-susan/analyse_tracks.py

#!/usr/bin/env python3
"""Analyse audio tracks with Essentia. Run daily via cron.
Use venv: /var/lib/radio/venv/bin/python3 /var/lib/radio/analyse_tracks.py
"""

import math
import os
import sqlite3
import subprocess
import sys
import tempfile
import time
from pathlib import Path

FEATURES_DB = "/var/lib/radio/audio_features.db"
MUSIC_ROOT = "/disks/Plex/Music/"
SUPPORTED = {".flac", ".mp3", ".ogg", ".opus", ".m4a", ".wav"}

# Energy percentile breakpoints from bulk analysis of ~6500 tracks.
# Maps raw log-energy values to even 0-1 distribution.
# Index 0 = 0th percentile, index 20 = 100th percentile (every 5%).
ENERGY_BREAKPOINTS = [
    0.0, 0.7148, 0.7581, 0.7801, 0.7999, 0.8163, 0.8289, 0.8401,
    0.8493, 0.8587, 0.8667, 0.8744, 0.8811, 0.8877, 0.8949, 0.9018,
    0.9089, 0.9169, 0.9252, 0.9377, 1.0,
]


def normalize_energy(raw_log_energy: float) -> float:
    """Map a raw log-energy value to 0-1 using the library's percentile curve."""
    bp = ENERGY_BREAKPOINTS
    if raw_log_energy <= bp[0]:
        return 0.0
    if raw_log_energy >= bp[-1]:
        return 1.0
    # Find which bucket it falls in and interpolate
    for i in range(1, len(bp)):
        if raw_log_energy <= bp[i]:
            frac = (raw_log_energy - bp[i - 1]) / (bp[i] - bp[i - 1]) if bp[i] != bp[i - 1] else 0.5
            return round((i - 1 + frac) / (len(bp) - 1), 4)
    return 1.0


def init_db():
    """Create features DB and table if needed."""
    conn = sqlite3.connect(FEATURES_DB)
    conn.execute("PRAGMA journal_mode=WAL")
    conn.execute("""
        CREATE TABLE IF NOT EXISTS features (
            path TEXT PRIMARY KEY,
            energy REAL,
            bpm REAL,
            loudness REAL,
            danceability REAL,
            analysed_at REAL
        )
    """)
    conn.commit()
    return conn


def get_analysed_paths(conn):
    """Return set of already-analysed file paths."""
    cur = conn.execute("SELECT path FROM features")
    return {row[0] for row in cur}


def find_audio_files():
    """Walk music root for all supported audio files."""
    root = Path(MUSIC_ROOT)
    files = []
    for p in root.rglob("*"):
        if p.suffix.lower() in SUPPORTED and p.is_file():
            files.append(str(p))
    return files


def analyse_track(filepath):
    """Extract audio features using Essentia. Returns dict or None on error."""
    try:
        import essentia
        import essentia.standard as es
    except ImportError:
        print("Error: essentia not installed. Install with: pip install essentia", file=sys.stderr)
        sys.exit(1)

    # Essentia's bundled decoder segfaults on opus and struggles with some
    # m4a/ogg files. Pre-decode these to a temp WAV via ffmpeg.
    NEEDS_PREDECODE = {".opus", ".ogg", ".m4a"}
    tmp_wav = None
    load_path = filepath

    if Path(filepath).suffix.lower() in NEEDS_PREDECODE:
        try:
            tmp_wav = tempfile.NamedTemporaryFile(suffix=".wav", delete=False)
            tmp_wav.close()
            subprocess.run(
                ["ffmpeg", "-y", "-i", filepath, "-ac", "1", "-ar", "22050", "-sample_fmt", "s16", "-f", "wav", tmp_wav.name],
                capture_output=True, timeout=15,
            )
            load_path = tmp_wav.name
        except Exception as e:
            print(f"  Skipping (ffmpeg decode error): {e}", file=sys.stderr)
            if tmp_wav and os.path.exists(tmp_wav.name):
                os.unlink(tmp_wav.name)
            return None

    try:
        sr = 22050 if tmp_wav else 44100
        audio = es.MonoLoader(filename=load_path, sampleRate=sr)()
    except Exception as e:
        print(f"  Skipping (load error): {e}", file=sys.stderr)
        if tmp_wav and os.path.exists(tmp_wav.name):
            os.unlink(tmp_wav.name)
        return None
    finally:
        if tmp_wav and os.path.exists(tmp_wav.name):
            os.unlink(tmp_wav.name)

    # Energy — raw log-energy mapped through percentile curve
    try:
        raw_energy = es.Energy()(audio)
        raw_log = min(1.0, math.log1p(raw_energy) / 15.0)
        energy_norm = normalize_energy(raw_log)
        energy_raw = raw_log
    except Exception:
        energy_norm = 0.5
        energy_raw = 0.5

    # BPM
    try:
        bpm, _, _, _, _ = es.RhythmExtractor2013()(audio)
        if bpm < 30 or bpm > 300:
            bpm = 0.0
    except Exception:
        bpm = 0.0

    # Loudness
    try:
        loudness = es.Loudness()(audio)
    except Exception:
        loudness = 0.0

    # Danceability
    try:
        danceability, _ = es.Danceability()(audio)
    except Exception:
        danceability = 0.0

    return {
        "energy": round(energy_norm, 4),
        "energy_raw": round(energy_raw, 4),
        "bpm": round(bpm, 2),
        "loudness": round(loudness, 4),
        "danceability": round(danceability, 4),
    }


def main():
    print(f"Essentia track analyser — {time.strftime('%Y-%m-%d %H:%M:%S')}")

    conn = init_db()
    analysed = get_analysed_paths(conn)
    print(f"Already analysed: {len(analysed)} tracks")

    all_files = find_audio_files()
    print(f"Found on disk: {len(all_files)} audio files")

    pending = [f for f in all_files if f not in analysed]
    print(f"To analyse: {len(pending)} new tracks")

    if not pending:
        print("Nothing to do.")
        conn.close()
        return

    done = 0
    errors = 0
    start = time.time()

    for i, filepath in enumerate(pending):
        if (i + 1) % 100 == 0 or i == 0:
            elapsed = time.time() - start
            rate = (done / elapsed) if elapsed > 0 and done > 0 else 0
            eta = ((len(pending) - i) / rate / 60) if rate > 0 else 0
            print(f"  [{i+1}/{len(pending)}] {rate:.1f} tracks/sec, ETA {eta:.0f} min")

        features = analyse_track(filepath)
        if features is None:
            errors += 1
            continue

        try:
            conn.execute(
                "INSERT OR REPLACE INTO features (path, energy, energy_raw, bpm, loudness, danceability, analysed_at) VALUES (?, ?, ?, ?, ?, ?, ?)",
                (filepath, features["energy"], features["energy_raw"], features["bpm"], features["loudness"], features["danceability"], time.time()),
            )
            if done % 50 == 0:
                conn.commit()
            done += 1
        except Exception as e:
            print(f"  DB error for {filepath}: {e}", file=sys.stderr)
            errors += 1

    conn.commit()
    conn.close()

    elapsed = time.time() - start
    print(f"Done: {done} analysed, {errors} errors, {elapsed:.1f}s total")


if __name__ == "__main__":
    main()