Update app.py

app.py

@@ -1,265 +1,275 @@
-import streamlit as st
-import torch
-import os
-import sys
-import librosa
-import numpy as np
-import pandas as pd
-import joblib  # Atau pickle, sesuaikan dengan cara Anda menyimpan model
-from speechbrain.inference.speaker import SpeakerRecognition
-
-# --- KONFIGURASI APLIKASI ---
-st.set_page_config(page_title="Verifikasi Suara", layout="centered")
-st.title("🔐 Sistem Verifikasi Perintah Suara")
-st.write("Aplikasi ini hanya akan merespon perintah 'Buka' atau 'Tutup' jika diucapkan oleh pengguna yang terdaftar.")
-
-# --- PATH & PENGATURAN MODEL ---
-# Sesuaikan path ini
-APP_DIR = os.path.dirname(os.path.abspath(__file__))
-
-# Buat semua path lain berdasarkan APP_DIR
-PATH_MODEL_KWS = os.path.join(APP_DIR, "model_kws.pkl")
-PATH_LABEL_ENCODER = os.path.join(APP_DIR, "label_encoder.pkl")
-PATH_ANDA = os.path.join(APP_DIR, "enroll", "v_ilham")
-PATH_TEMAN = os.path.join(APP_DIR, "enroll", "v_danendra")
-
-# Pengaturan threshold (tetap sama)
-THRESHOLD = 0.85  # Sesuaikan ini!
-
-# --- FUNGSI BANTUAN MODEL 2 (SpeechBrain) ---
-# Fungsi-fungsi ini dari skrip kita sebelumnya
-
-def get_embedding(file_path, model_sv):
-    if not os.path.exists(file_path):
-        st.error(f"File not found: {file_path}")
-        return None
-    try:
-        embedding = model_sv.encode_file(file_path)
-        return embedding.squeeze()
-    except Exception as e:
-        st.error(f"Error processing {file_path}: {e}")
-        return None
-
-def get_similarity(emb1, emb2, model_sv):
-    emb1_batch = emb1.unsqueeze(0)
-    emb2_batch = emb2.unsqueeze(0)
-    score = model_sv.similarity(emb1_batch, emb2_batch)
-    return score.item()
-
-def create_master_voiceprint(directory_path, model_sv):
-    embeddings = []
-    if not os.path.isdir(directory_path):
-        st.warning(f"Direktori pendaftaran tidak ditemukan: {directory_path}")
-        return None
-
-    for file_name in os.listdir(directory_path):
-        if file_name.endswith(".wav"):
-            file_path = os.path.join(directory_path, file_name)
-            emb = get_embedding(file_path, model_sv)
-            if emb is not None:
-                embeddings.append(emb)
-    
-    if not embeddings:
-        st.error(f"Tidak ada file .wav ditemukan di {directory_path}")
-        return None
-        
-    master_voiceprint = torch.mean(torch.stack(embeddings), dim=0)
-    return master_voiceprint
-
-# --- LOADING MODEL (DENGAN CACHE) ---
-# Ini agar model tidak di-load ulang setiap kali ada interaksi
-
-@st.cache_resource
-def load_model_sv():
-    st.info("Memuat Model Verifikasi Suara (SpeechBrain)...")
-    try:
-        model = SpeakerRecognition.from_hparams(
-            source="speechbrain/spkrec-ecapa-tdnn",
-            savedir="pretrained_models/spkrec-ecapa-tdnn",
-            use_auth_token=False # Tambahkan ini untuk menghindari error 401
-        )
-        st.success("Model Verifikasi Suara siap.")
-        return model
-    except Exception as e:
-        st.exception(e)
-        st.error("Gagal memuat model SpeechBrain. Cek koneksi internet.")
-        return None
-
-@st.cache_resource
-def load_model_kws(path):
-    st.info("Memuat Model Pengenal Kata Kunci (KWS)...")
-    if not os.path.exists(path):
-        st.error(f"File model KWS tidak ditemukan di: {path}")
-        return None
-    try:
-        model = joblib.load(path)
-        st.success("Model KWS siap.")
-        return model
-    except Exception as e:
-        st.exception(e)
-        return None
-
-@st.cache_resource
-def load_voiceprints(_model_sv):
-    st.info("Membuat master voiceprint...")
-    voiceprints = {}
-    
-    vp_a = create_master_voiceprint(PATH_ANDA, _model_sv)
-    if vp_a is not None:
-        voiceprints["anda"] = vp_a
-        st.success("Voiceprint 'anda' dibuat.")
-        
-    vp_b = create_master_voiceprint(PATH_TEMAN, _model_sv)
-    if vp_b is not None:
-        voiceprints["teman"] = vp_b
-        st.success("Voiceprint 'teman' dibuat.")
-        
-    if not voiceprints:
-        st.error("Gagal membuat voiceprint. Pastikan folder 'enroll' ada.")
-        return None
-    return voiceprints
-
-# --- FUNGSI PIPELINE UTAMA ---
-
-def ekstrak_fitur_kws(audio_file):
-    """
-    PENTING: Fungsi ini HARUS mengekstrak fitur yang SAMA PERSIS
-    dengan yang Anda gunakan untuk melatih Model 1 di Ppreprocessing.pdf.
-    Ini hanya contoh!
-    """
-    y, sr = librosa.load(audio_file, sr=16000)
-    
-    # Preprocessing dari PDF Anda: trim & normalisasi
-    y_trimmed, _ = librosa.effects.trim(y, top_db=20)
-    y_norm = y_trimmed / np.max(np.abs(y_trimmed))
-    
-    # Ekstrak fitur (HARUS SAMA DENGAN PDF ANDA)
-    # [cite_start]Ini hanya contoh dari PDF Anda [cite: 194-212]
-    fitur = {
-        'spectral_centroid': np.mean(librosa.feature.spectral_centroid(y=y_norm, sr=sr)),
-        'spectral_bandwidth': np.mean(librosa.feature.spectral_bandwidth(y=y_norm, sr=sr)),
-        'spectral_rolloff': np.mean(librosa.feature.spectral_rolloff(y=y_norm, sr=sr)),
-        'spectral_contrast': np.mean(librosa.feature.spectral_contrast(y=y_norm, sr=sr)),
-        'spectral_flatness': np.mean(librosa.feature.spectral_flatness(y=y_norm)),
-        'mfcc_delta2_mean': np.mean(librosa.feature.delta(librosa.feature.mfcc(y=y_norm, sr=sr, n_mfcc=13), order=2)),
-        'f0_mean': np.nanmean(librosa.pyin(y_norm, fmin=50, fmax=400, sr=sr)[0]),
-        'rms': np.mean(librosa.feature.rms(y=y_norm)),
-        'duration': librosa.get_duration(y=y_norm, sr=sr),
-        'std_amplitude': np.std(y_norm)
-    }
-    
-    # Mengisi NaN jika ada (misal dari f0)
-    df = pd.DataFrame([fitur]).fillna(0)
-    
-    # Pastikan urutan kolom sama dengan saat training!
-    # Ini hanya contoh berdasarkan 10 fitur akhir di PDF Anda
-    nama_fitur_akhir = [
-        'spectral_centroid', 'spectral_bandwidth', 'spectral_rolloff',
-        'spectral_contrast', 'spectral_flatness', 'mfcc_delta2_mean',
-        'f0_mean', 'rms', 'duration', 'std_amplitude'
-    ]
-    
-    # Filter dan urutkan
-    df = df[nama_fitur_akhir]
-    return df
-
-
-def cek_keyword(audio_file, model_kws):
-    """
-    Menjalankan Model 1 (KWS) untuk mendeteksi kata kunci.
-    """
-    try:
-        fitur = ekstrak_fitur_kws(audio_file)
-        prediksi = model_kws.predict(fitur)
-        return prediksi[0]  # Mengambil hasil prediksi (misal: "buka" atau "tutup")
-    except Exception as e:
-        st.exception(e)
-        st.error("Gagal mengekstrak fitur KWS.")
-        return "error"
-
-def verifikasi_suara(audio_file, model_sv, voiceprints, threshold):
-    """
-    Menjalankan Model 2 (SV) untuk verifikasi pembicara.
-    """
-    try:
-        test_embedding = get_embedding(audio_file, model_sv)
-        if test_embedding is None:
-            return False, 0.0, "Gagal buat embedding"
-
-        best_score = -1.0
-        best_match = "None"
-        
-        for name, master_vp in voiceprints.items():
-            score = get_similarity(test_embedding, master_vp, model_sv)
-            if score > best_score:
-                best_score = score
-                best_match = name
-        
-        if best_score >= threshold:
-            return True, best_score, best_match
-        else:
-            return False, best_score, "None"
-            
-    except Exception as e:
-        st.exception(e)
-        st.error("Gagal saat verifikasi suara.")
-        return False, 0.0, "Error"
-
-# --- MAIN APP ---
-# Memuat semua model saat aplikasi dimulai
-model_sv = load_model_sv()
-model_kws = load_model_kws(PATH_MODEL_KWS)
-voiceprints = load_voiceprints(model_sv)
-
-# Cek jika model gagal di-load
-if not all([model_sv, model_kws, voiceprints]):
-    st.error("Gagal memuat semua model atau voiceprint. Aplikasi tidak bisa berjalan.")
-else:
-    st.header("Upload Audio Perintah (.wav)")
-    uploaded_file = st.file_uploader("Pilih file audio...", type=["wav"])
-    
-    # Buat file audio sementara
-    temp_audio_path = None
-    if uploaded_file is not None:
-        # Simpan file yang di-upload ke disk sementara
-        # karena librosa & speechbrain butuh path file
-        with open("temp_audio.wav", "wb") as f:
-            f.write(uploaded_file.getbuffer())
-        temp_audio_path = "temp_audio.wav"
-        
-        st.audio(temp_audio_path)
-
-    if st.button("Proses Perintah", disabled=(temp_audio_path is None)):
-        if temp_audio_path:
-            with st.spinner("Menganalisis audio..."):
-                
-                # --- LANGKAH 1: Cek Kata Kunci ---
-                st.subheader("Hasil Model 1: Pengenalan Kata Kunci")
-                kata_kunci = cek_keyword(temp_audio_path, model_kws)
-                
-                if kata_kunci in ["buka", "tutup"]:
-                    st.info(f"Kata kunci terdeteksi: **{kata_kunci.upper()}**")
-                    
-                    # --- LANGKAH 2: Verifikasi Suara ---
-                    st.subheader("Hasil Model 2: Verifikasi Suara")
-                    terverifikasi, skor, nama = verifikasi_suara(
-                        temp_audio_path, model_sv, voiceprints, THRESHOLD
-                    )
-                    
-                    st.info(f"Skor kemiripan tertinggi: **{skor:.2%}** (dengan '{nama}')")
-                    
-                    # --- KEPUTUSAN AKHIR ---
-                    st.header("Keputusan Akhir")
-                    if terverifikasi:
-                        st.success(f"✅ DITERIMA. Suara terverifikasi sebagai '{nama}'. Perintah **{kata_kunci.upper()}** dijalankan.")
-                    else:
-                        st.error(f"❌ DITOLAK. Suara tidak dikenal. Perintah **{kata_kunci.upper()}** dibatalkan.")
-                        
-                elif kata_kunci == "error":
-                    st.error("Terjadi error saat memproses kata kunci.")
-                else:
-                    st.header("Keputusan Akhir")
-                    st.warning(f"❌ DITOLAK. Perintah tidak dikenal (terdeteksi sebagai: '{kata_kunci}').")
-
-            # Hapus file sementara
-            os.remove(temp_audio_path)
+import streamlit as st
+import torch
+import os
+import sys
+import librosa
+import numpy as np
+import pandas as pd
+import joblib  # Memuat .pkl
+from speechbrain.inference.speaker import SpeakerRecognition
+
+# --- KONFIGURASI APLIKASI ---
+st.set_page_config(page_title="Verifikasi Suara", layout="centered")
+st.title("🔐 Sistem Verifikasi Perintah Suara")
+st.write("Aplikasi ini hanya akan merespon perintah 'Buka' atau 'Tutup' jika diucapkan oleh pengguna yang terdaftar.")
+
+# --- PATH & PENGATURAN MODEL ---
+APP_DIR = os.path.dirname(os.path.abspath(__file__))
+PATH_MODEL_KWS = os.path.join(APP_DIR, "model_kws.pkl")
+PATH_LABEL_ENCODER = os.path.join(APP_DIR, "label_encoder.pkl") # DIPERLUKAN
+PATH_ILHAM = os.path.join(APP_DIR, "enroll", "v_ilham")  # Disesuaikan
+PATH_DANENDRA = os.path.join(APP_DIR, "enroll", "v_danendra") # Disesuaikan
+
+THRESHOLD = 0.85  # Sesuaikan ini!
+
+# --- FUNGSI BANTUAN MODEL 2 (SpeechBrain) ---
+
+def get_embedding(file_path, model_sv):
+    if not os.path.exists(file_path):
+        st.error(f"File not found: {file_path}")
+        return None
+    try:
+        embedding = model_sv.encode_file(file_path)
+        return embedding.squeeze()
+    except Exception as e:
+        st.error(f"Error processing {file_path}: {e}")
+        return None
+
+def get_similarity(emb1, emb2, model_sv):
+    emb1_batch = emb1.unsqueeze(0)
+    emb2_batch = emb2.unsqueeze(0)
+    score = model_sv.similarity(emb1_batch, emb2_batch)
+    return score.item()
+
+def create_master_voiceprint(directory_path, model_sv):
+    embeddings = []
+    if not os.path.isdir(directory_path):
+        st.warning(f"Direktori pendaftaran tidak ditemukan: {directory_path}")
+        return None
+
+    for file_name in os.listdir(directory_path):
+        if file_name.endswith(".wav"):
+            file_path = os.path.join(directory_path, file_name)
+            emb = get_embedding(file_path, model_sv)
+            if emb is not None:
+                embeddings.append(emb)
+    
+    if not embeddings:
+        st.error(f"Tidak ada file .wav ditemukan di {directory_path}")
+        return None
+        
+    master_voiceprint = torch.mean(torch.stack(embeddings), dim=0)
+    return master_voiceprint
+
+# --- LOADING MODEL (DENGAN CACHE) ---
+
+@st.cache_resource
+def load_model_sv():
+    st.info("Memuat Model Verifikasi Suara (SpeechBrain)...")
+    try:
+        model = SpeakerRecognition.from_hparams(
+            source="speechbrain/spkrec-ecapa-tdnn",
+            savedir="pretrained_models/spkrec-ecapa-tdnn",
+            use_auth_token=False
+        )
+        st.success("Model Verifikasi Suara siap.")
+        return model
+    except Exception as e:
+        st.exception(e)
+        st.error("Gagal memuat model SpeechBrain. Cek koneksi internet.")
+        return None
+
+@st.cache_resource
+def load_model_kws_and_le(path_model, path_le):
+    st.info("Memuat Model Pengenal Kata Kunci (KWS) & Label Encoder...")
+    model_kws = None
+    le = None
+    
+    if not os.path.exists(path_model):
+        st.error(f"File model KWS tidak ditemukan di: {path_model}")
+    else:
+        try:
+            model_kws = joblib.load(path_model)
+            st.success("Model KWS siap.")
+        except Exception as e:
+            st.exception(e)
+            st.error("Gagal memuat model KWS.")
+
+    if not os.path.exists(path_le):
+        st.error(f"File Label Encoder tidak ditemukan di: {path_le}")
+    else:
+        try:
+            le = joblib.load(path_le)
+            st.success(f"Label Encoder siap (Kelas: {le.classes_}).")
+        except Exception as e:
+            st.exception(e)
+            st.error("Gagal memuat Label Encoder.")
+            
+    return model_kws, le
+
+@st.cache_resource
+def load_voiceprints(_model_sv):
+    st.info("Membuat master voiceprint...")
+    voiceprints = {}
+    
+    vp_a = create_master_voiceprint(PATH_ILHAM, _model_sv)
+    if vp_a is not None:
+        voiceprints["v_ilham"] = vp_a  # Perbaikan: Gunakan nama yang benar
+        st.success("Voiceprint 'v_ilham' dibuat.")
+        
+    vp_b = create_master_voiceprint(PATH_DANENDRA, _model_sv)
+    if vp_b is not None:
+        voiceprints["v_danendra"] = vp_b # Perbaikan: Gunakan nama yang benar
+        st.success("Voiceprint 'v_danendra' dibuat.")
+        
+    if not voiceprints:
+        st.error("Gagal membuat voiceprint. Pastikan folder 'enroll' ada.")
+        return None
+    return voiceprints
+
+# --- FUNGSI PIPELINE UTAMA ---
+
+def ekstrak_fitur_kws(audio_file):
+    """
+    Fungsi ini mengekstrak fitur yang SAMA PERSIS
+    dengan yang Anda gunakan untuk melatih Model 1.
+    """
+    try:
+        y, sr = librosa.load(audio_file, sr=16000)
+        
+        y_trimmed, _ = librosa.effects.trim(y, top_db=20)
+        
+        # Perbaikan: Pencegahan error jika audio hening total
+        if len(y_trimmed) == 0:
+            st.warning("Audio terdeteksi hening. Fitur tidak bisa diekstrak.")
+            return None # Kembalikan None jika hening
+        
+        # Perbaikan: Pencegahan error pembagian dengan nol
+        y_norm = y_trimmed / (np.max(np.abs(y_trimmed)) + 1e-6)
+        
+        fitur = {
+            'spectral_centroid': np.mean(librosa.feature.spectral_centroid(y=y_norm, sr=sr)),
+            'spectral_bandwidth': np.mean(librosa.feature.spectral_bandwidth(y=y_norm, sr=sr)),
+            'spectral_rolloff': np.mean(librosa.feature.spectral_rolloff(y=y_norm, sr=sr)),
+            'spectral_contrast': np.mean(librosa.feature.spectral_contrast(y=y_norm, sr=sr)),
+            'spectral_flatness': np.mean(librosa.feature.spectral_flatness(y=y_norm)),
+            'mfcc_delta2_mean': np.mean(librosa.feature.delta(librosa.feature.mfcc(y=y_norm, sr=sr, n_mfcc=13), order=2)),
+            'f0_mean': np.nanmean(librosa.pyin(y_norm, fmin=librosa.note_to_hz('C2'), fmax=librosa.note_to_hz('C7'), sr=sr)[0]),
+            'rms': np.mean(librosa.feature.rms(y=y_norm)),
+            'duration': librosa.get_duration(y=y_norm, sr=sr),
+            'std_amplitude': np.std(y_norm)
+        }
+        
+        df = pd.DataFrame([fitur]).fillna(0)
+        
+        nama_fitur_akhir = [
+            'spectral_centroid', 'spectral_bandwidth', 'spectral_rolloff',
+            'spectral_contrast', 'spectral_flatness', 'mfcc_delta2_mean',
+            'f0_mean', 'rms', 'duration', 'std_amplitude'
+        ]
+        
+        df = df[nama_fitur_akhir]
+        return df
+        
+    except Exception as e:
+        st.exception(e)
+        st.error("Gagal mengekstrak fitur KWS.")
+        return None
+
+
+def cek_keyword(audio_file, model_kws, le): # Perbaikan: Tambahkan 'le'
+    """
+    Menjalankan Model 1 (KWS) untuk mendeteksi kata kunci.
+    """
+    fitur = ekstrak_fitur_kws(audio_file)
+    if fitur is None:
+        return "error"
+        
+    prediksi_angka = model_kws.predict(fitur)
+    prediksi_label = le.inverse_transform(prediksi_angka) # Perbaikan: Ubah angka jadi teks
+    return prediksi_label[0]
+
+def verifikasi_suara(audio_file, model_sv, voiceprints, threshold):
+    """
+    Menjalankan Model 2 (SV) untuk verifikasi pembicara.
+    """
+    try:
+        test_embedding = get_embedding(audio_file, model_sv)
+        if test_embedding is None:
+            return False, 0.0, "Gagal buat embedding"
+
+        best_score = -1.0
+        best_match = "None"
+        
+        for name, master_vp in voiceprints.items():
+            score = get_similarity(test_embedding, master_vp, model_sv)
+            if score > best_score:
+                best_score = score
+                best_match = name
+        
+        if best_score >= threshold:
+            return True, best_score, best_match
+        else:
+            return False, best_score, "None"
+            
+    except Exception as e:
+        st.exception(e)
+        st.error("Gagal saat verifikasi suara.")
+        return False, 0.0, "Error"
+
+# --- MAIN APP ---
+model_sv = load_model_sv()
+# Perbaikan: Muat model dan label encoder
+model_kws, le = load_model_kws_and_le(PATH_MODEL_KWS, PATH_LABEL_ENCODER)
+voiceprints = load_voiceprints(model_sv)
+
+# Cek jika model gagal di-load
+if not all([model_sv, model_kws, le, voiceprints]):
+    st.error("Gagal memuat semua model/voiceprint. Aplikasi tidak bisa berjalan. Cek error di atas.")
+else:
+    st.header("Upload Audio Perintah (.wav)")
+    uploaded_file = st.file_uploader("Pilih file audio...", type=["wav"])
+    
+    temp_audio_path = None
+    if uploaded_file is not None:
+        with open("temp_audio.wav", "wb") as f:
+            f.write(uploaded_file.getbuffer())
+        temp_audio_path = "temp_audio.wav"
+        st.audio(temp_audio_path)
+
+    if st.button("Proses Perintah", disabled=(temp_audio_path is None)):
+        if temp_audio_path:
+            with st.spinner("Menganalisis audio..."):
+                
+                # --- LANGKAH 1: Cek Kata Kunci ---
+                st.subheader("Hasil Model 1: Pengenalan Kata Kunci")
+                kata_kunci = cek_keyword(temp_audio_path, model_kws, le) # Perbaikan: kirim 'le'
+                
+                # Perbaikan: Cek dengan le.classes_
+                if kata_kunci in le.classes_: 
+                    st.info(f"Kata kunci terdeteksi: **{kata_kunci.upper()}**")
+                    
+                    # --- LANGKAH 2: Verifikasi Suara ---
+                    st.subheader("Hasil Model 2: Verifikasi Suara")
+                    terverifikasi, skor, nama = verifikasi_suara(
+                        temp_audio_path, model_sv, voiceprints, THRESHOLD
+                    )
+                    
+                    st.info(f"Skor kemiripan tertinggi: **{skor:.2%}** (dengan '{nama}')")
+                    
+                    # --- KEPUTUSAN AKHIR ---
+                    st.header("Keputusan Akhir")
+                    if terverifikasi:
+                        st.success(f"✅ DITERIMA. Suara terverifikasi sebagai '{nama}'. Perintah **{kata_kunci.upper()}** dijalankan.")
+                    else:
+                        st.error(f"❌ DITOLAK. Suara tidak dikenal. Perintah **{kata_kunci.upper()}** dibatalkan.")
+                        
+                elif kata_kunci == "error":
+                    st.error("Terjadi error saat memproses kata kunci.")
+                else:
+                    st.header("Keputusan Akhir")
+                    st.warning(f"❌ DITOLAK. Perintah tidak dikenal (terdeteksi sebagai: '{kata_kunci}').")
+
+            # Hapus file sementara
+            if os.path.exists("temp_audio.wav"):
+                os.remove("temp_audio.wav")