Update src/streamlit_app.py

src/streamlit_app.py

@@ -1,40 +1,190 @@
-import altair as alt
+# streamlit_app.py
+# ------------------------------------------------------------
+# Voice Guard (Streamlit) - env-only config (no st.secrets required)
+# - Tries app/ then src/ for the Detector
+# - Accepts mic (best-effort) or upload
+# - Shows probabilities, decision details, and CAM heatmap
+# - If MODEL_WEIGHTS_URL is set, downloads weights on boot when missing
+# ------------------------------------------------------------
+
+import os
+import io
+import pathlib
+import urllib.request
 import numpy as np
-import pandas as pd
 import streamlit as st
+from PIL import Image
+from matplotlib import cm
+
+# --------------------------- Import Detector ---------------------------
+Detector = None
+_last_err = None
+for mod in [
+    "app.inference_wav2vec",
+    "app.inference",
+    "src.inference_wav2vec",
+    "src.inference",
+]:
+    try:
+        Detector = __import__(mod, fromlist=["Detector"]).Detector
+        break
+    except Exception as e:
+        _last_err = e
+
+if Detector is None:
+    st.error(
+        "Could not import Detector from app/ or src/. "
+        "Please include app/inference_wav2vec.py (preferred) or app/inference.py. "
+        f"Last import error: {_last_err}"
+    )
+    st.stop()
+
+# ----------------------- Weights: ensure on disk -----------------------
+def cfg(name: str, default: str = "") -> str:
+    """Read from environment only (HF Variables & Secrets are env)."""
+    val = os.getenv(name)
+    return val if val not in (None, "") else default
+
+def ensure_weights() -> str:
+    """
+    Ensure model weights exist at MODEL_WEIGHTS_PATH.
+    If missing and MODEL_WEIGHTS_URL is set, download them.
+    """
+    default_path = "app/models/weights/wav2vec2_classifier.pth"
+    wp = cfg("MODEL_WEIGHTS_PATH", default_path)
+    url = cfg("MODEL_WEIGHTS_URL", "")
+
+    dest = pathlib.Path(wp)
+    if not dest.exists():
+        if url:
+            dest.parent.mkdir(parents=True, exist_ok=True)
+            with st.spinner(f"Downloading model weights to {dest} …"):
+                urllib.request.urlretrieve(url, str(dest))
+                st.toast("Weights downloaded", icon="✅")
+        else:
+            st.warning(
+                f"Model weights not found at '{wp}'. "
+                "Upload the .pth file to that path in the repo OR set MODEL_WEIGHTS_URL in "
+                "Settings → Variables & secrets so the app can download them."
+            )
+    return str(dest)
+
+@st.cache_resource(show_spinner=True)
+def load_detector() -> "Detector":
+    weights_path = ensure_weights()
+    det = Detector(weights_path=weights_path)
+    return det
+
+det = load_detector()
+
+# ----------------------------- Utilities -------------------------------
+def cam_to_png_bytes(cam: np.ndarray) -> bytes:
+    """Map [H,W] float array (0..1) to magma RGB PNG bytes."""
+    cam = np.asarray(cam, dtype=np.float32)
+    cam = np.nan_to_num(cam, nan=0.0)
+    cam = np.clip(cam, 0.0, 1.0)
+    rgb = (cm.magma(cam)[..., :3] * 255).astype(np.uint8)
+    img = Image.fromarray(rgb)
+    bio = io.BytesIO()
+    img.save(bio, format="PNG")
+    return bio.getvalue()
+
+def analyze(wav_bytes: bytes, source_hint: str):
+    """Call detector predict + explain; returns (proba_dict, explain_dict)."""
+    proba = det.predict_proba(wav_bytes, source_hint=source_hint)
+    exp = det.explain(wav_bytes, source_hint=source_hint)
+    return proba, exp
+
+# ------------------------------- UI -----------------------------------
+st.set_page_config(page_title="Voice Guard", page_icon="🛡️", layout="wide")
+st.title("🛡️ Voice Guard — Human vs AI Speech")
+
+left, right = st.columns([1, 2], gap="large")
+
+with left:
+    st.subheader("Input")
+    tabs = st.tabs(["🎙️ Microphone", "📁 Upload"])
+
+    wav_bytes = None
+    source_hint = None
+
+    # Microphone tab (best effort; if not supported, use Upload)
+    with tabs[0]:
+        st.caption("Record ~3–7 seconds. If mic fails in your browser, use Upload.")
+        try:
+            from audio_recorder_streamlit import audio_recorder
+            audio = audio_recorder(
+                text="Record",
+                recording_color="#ff6a00",
+                neutral_color="#2b2b2b",
+                icon_size="2x",
+            )
+            if audio:
+                wav_bytes = audio  # component returns WAV bytes
+                source_hint = "microphone"
+                st.audio(wav_bytes, format="audio/wav")
+        except Exception:
+            st.info("Recorder component not available here—please use the Upload tab.")
+
+    # Upload tab (most reliable across platforms)
+    with tabs[1]:
+        f = st.file_uploader(
+            "Upload an audio file (wav/mp3/m4a/aac)",
+            type=["wav", "mp3", "m4a", "aac"],
+        )
+        if f is not None:
+            wav_bytes = f.read()
+            source_hint = "upload"
+            st.audio(wav_bytes)
+
+    st.markdown("---")
+    run = st.button(
+        "🔍 Analyze", type="primary", use_container_width=True, disabled=wav_bytes is None
+    )
+
+with right:
+    st.subheader("Results")
+
+    if run and wav_bytes:
+        try:
+            with st.spinner("Analyzing…"):
+                proba, exp = analyze(wav_bytes, source_hint or "auto")
+
+            ph = float(proba.get("human", 0.0))
+            pa = float(proba.get("ai", 0.0))
+            label = (proba.get("label", "human") or "human").upper()
+            thr = float(proba.get("threshold", 0.5))
+            rule = proba.get("decision", "threshold")
+            thr_src = proba.get("threshold_source", "—")
+            rscore = proba.get("replay_score", None)
+
+            c1, c2, c3 = st.columns(3)
+            with c1:
+                st.metric("Human", f"{ph*100:.1f}%")
+            with c2:
+                st.metric("AI", f"{pa*100:.1f}%")
+            with c3:
+                color = "#22c55e" if label == "HUMAN" else "#fb7185"
+                st.markdown(
+                    f"**Final Label:** <span style='color:{color}'>{label}</span>",
+                    unsafe_allow_html=True,
+                )
+                st.caption(
+                    f"thr({thr_src})={thr:.2f} • rule={rule} • replay={'—' if rscore is None else f'{float(rscore):.2f}'}"
+                )
+
+            st.markdown("##### Explanation Heatmap")
+            cam = np.asarray(exp.get("cam"), dtype=np.float32)
+            st.image(
+                cam_to_png_bytes(cam),
+                caption="Spectrogram importance",
+                use_column_width=True,
+            )
+
+            with st.expander("Raw JSON (debug)"):
+                st.json({"proba": proba, "explain": {"cam_shape": list(cam.shape)}})
+
+        except Exception as e:
+            st.error(f"Analyze failed: {e}")
 
-"""
-# Welcome to Streamlit!
-
-Edit `/streamlit_app.py` to customize this app to your heart's desire :heart:.
-If you have any questions, checkout our [documentation](https://docs.streamlit.io) and [community
-forums](https://discuss.streamlit.io).
-
-In the meantime, below is an example of what you can do with just a few lines of code:
-"""
-
-num_points = st.slider("Number of points in spiral", 1, 10000, 1100)
-num_turns = st.slider("Number of turns in spiral", 1, 300, 31)
-
-indices = np.linspace(0, 1, num_points)
-theta = 2 * np.pi * num_turns * indices
-radius = indices
-
-x = radius * np.cos(theta)
-y = radius * np.sin(theta)
-
-df = pd.DataFrame({
-    "x": x,
-    "y": y,
-    "idx": indices,
-    "rand": np.random.randn(num_points),
-})
-
-st.altair_chart(alt.Chart(df, height=700, width=700)
-    .mark_point(filled=True)
-    .encode(
-        x=alt.X("x", axis=None),
-        y=alt.Y("y", axis=None),
-        color=alt.Color("idx", legend=None, scale=alt.Scale()),
-        size=alt.Size("rand", legend=None, scale=alt.Scale(range=[1, 150])),
-    ))
+st.caption("Tip: Uploading a short 3–7s clip is the most reliable across browsers.")