"""
Speaker Identification Module for HeadMic
Resemblyzer GE2E speaker encoder — 256-dim embeddings, cosine similarity matching.
Triggered when YAMNet detects speech.
"""

import logging
import sqlite3
import time
from pathlib import Path

import numpy as np

logger = logging.getLogger("speaker_id")
logger.setLevel(logging.INFO)

SIMILARITY_THRESHOLD = 0.75


class SpeakerRecognizer:
    def __init__(self, db_path="voices.db"):
        from resemblyzer import VoiceEncoder

        self._encoder = VoiceEncoder("cpu")
        logger.info("Resemblyzer voice encoder loaded")

        self._db_path = str(db_path)
        self._init_db()
        self._cache = self._load_embeddings()
        logger.info(
            "Speaker DB ready: %d embeddings for %d speakers",
            sum(len(v) for v in self._cache.values()),
            len(self._cache),
        )

    def _init_db(self):
        with sqlite3.connect(self._db_path) as conn:
            conn.execute("""
                CREATE TABLE IF NOT EXISTS voices (
                    id INTEGER PRIMARY KEY,
                    name TEXT NOT NULL,
                    embedding BLOB NOT NULL,
                    enrolled_at REAL NOT NULL,
                    source TEXT
                )
            """)
            conn.execute(
                "CREATE INDEX IF NOT EXISTS idx_voices_name ON voices(name)"
            )

    def _load_embeddings(self):
        """Load all embeddings from DB into memory, grouped by name."""
        cache = {}
        with sqlite3.connect(self._db_path) as conn:
            rows = conn.execute("SELECT name, embedding FROM voices").fetchall()
        for name, blob in rows:
            emb = np.frombuffer(blob, dtype=np.float32).copy()
            cache.setdefault(name, []).append(emb)
        return cache

    def identify(self, audio_float32):
        """Identify speaker from float32 audio at 16kHz.

        Returns:
            (name, confidence) or (None, 0.0) if no match above threshold.
        """
        if not self._cache:
            return None, 0.0

        try:
            from resemblyzer import preprocess_wav
            wav = preprocess_wav(audio_float32, source_sr=16000)
            if len(wav) < 1600:  # too short
                return None, 0.0
            embedding = self._encoder.embed_utterance(wav)
        except Exception as e:
            logger.warning("Embedding computation failed: %s", e)
            return None, 0.0

        best_name = None
        best_score = 0.0

        for name, embeddings in self._cache.items():
            # Best score across all enrolled samples for this speaker
            scores = [np.dot(embedding, emb) for emb in embeddings]
            top = max(scores)
            if top > best_score:
                best_score = top
                best_name = name

        if best_score >= SIMILARITY_THRESHOLD:
            return best_name, round(float(best_score), 3)
        return None, 0.0

    def enroll(self, name, audio_float32, source="api"):
        """Enroll a speaker from float32 audio at 16kHz.

        Returns:
            The computed embedding (256-dim).
        """
        from resemblyzer import preprocess_wav

        wav = preprocess_wav(audio_float32, source_sr=16000)
        if len(wav) < 1600:
            raise ValueError("Audio too short for enrollment")

        embedding = self._encoder.embed_utterance(wav)
        blob = embedding.astype(np.float32).tobytes()
        now = time.time()

        with sqlite3.connect(self._db_path) as conn:
            conn.execute(
                "INSERT INTO voices (name, embedding, enrolled_at, source) VALUES (?, ?, ?, ?)",
                (name, blob, now, source),
            )

        self._cache.setdefault(name, []).append(embedding)
        logger.info("Enrolled speaker '%s' (source=%s, total=%d samples)", name, source, len(self._cache[name]))
        return embedding

    def list_speakers(self):
        """Return enrolled speaker names with sample counts."""
        return {name: len(embs) for name, embs in self._cache.items()}

    def delete_speaker(self, name):
        """Remove all embeddings for a speaker."""
        with sqlite3.connect(self._db_path) as conn:
            conn.execute("DELETE FROM voices WHERE name = ?", (name,))
        removed = self._cache.pop(name, None)
        if removed:
            logger.info("Deleted speaker '%s' (%d samples)", name, len(removed))
            return len(removed)
        return 0