add replicate service

app.py

@@ -3,9 +3,9 @@ from pathlib import Path
 import tempfile, subprocess, threading, queue
 import textwrap
 
-st.set_page_config(page_title="Lec2Note2 – Lecture-to-Notes", layout="wide")
+st.set_page_config(page_title="Lec2Note – Lecture-to-Notes", layout="wide")
 
-st.title("📝 Lec2Note2 – Automatic Lecture Notes Generator")
+st.title("📝 Lec2Note – Automatic Lecture Notes Generator")
 
 st.markdown(
     textwrap.dedent(

lec2note/ingestion/whisper_runner.py

@@ -10,8 +10,9 @@ from typing import List, Dict, Optional, Any
 
 import torch, json, os
 from whisper import load_model  # type: ignore
-from openai import OpenAI
-
+import replicate
+import httpx
+import time
 __all__ = ["WhisperRunner"]
 
 
@@ -40,22 +41,43 @@ class WhisperRunner:  # noqa: D101
 
             logger.info("[Whisper] transcribing %s (local)", audio_path.name)
             result = model.transcribe(str(audio_path), language=lang)
+            segments = result.get("segments", [])
         else:
             # remote API mode
-            api_base = os.getenv("AIHUB_API_BASE")
-            api_key = os.getenv("AIHUB_API_KEY")
+            api_key = os.getenv("REPLICATE_API_TOKEN")
             if not api_key:
-                raise EnvironmentError("AIHUB_API_KEY not set")
+                raise EnvironmentError("REPLICATE_API_TOKEN not set")
+            logger.info("[Whisper] uploading %s to Replicate (whisper-large-v3)", audio_path.name)
+
+            client = replicate.Client(
+                api_token=api_key,
+                timeout=httpx.Timeout(60.0, connect=60.0, read=60.0),
+            )
+
+            version = "8099696689d249cf8b122d833c36ac3f75505c666a395ca40ef26f68e7d3d16e"  # whisper-large-v3
+            prediction = client.predictions.create(
+                version=version,
+                input={
+                    "audio": open(audio_path, "rb"),
+                    "language": lang,
+                },
+            )
+
+            poll_interval = float(os.getenv("REPLICATE_POLL_INTERVAL", "5"))
+            while prediction.status not in ("succeeded", "failed", "canceled"):
+                time.sleep(poll_interval)
+                prediction.reload()
+                logger.info("[Whisper] Replicate prediction status: %s", prediction.status)
+
+            if prediction.status != "succeeded":
+                raise RuntimeError(f"Replicate failed: {prediction.error}")
 
-            client = OpenAI(api_key=api_key, base_url=api_base)
-            logger.info("[Whisper] uploading %s to API (whisper-large-v3)", audio_path.name)
-            with audio_path.open("rb") as f:
-                resp = client.audio.transcriptions.create(model="whisper-large-v3", file=f, language=lang)
-            # resp.text contains full text, but we need segments; assume API returns segments list if 'json' format
-            segments = resp.segments if hasattr(resp, "segments") else [{"start": 0.0, "end": 0.0, "text": resp.text}]
-            result = {"segments": segments}
+            resp = prediction.output
+            segments = resp["segments"] if isinstance(resp, dict) and "segments" in resp else [
+                {"start": 0.0, "end": 0.0, "text": resp.get("text", "")}
+            ]
 
-        # convert to our schema
+        # convert to our schema (works for both modes)
         logger.info("[Whisper] got %d segments", len(segments))
         results = [
             {

lec2note/utils/logging_config.py

@@ -1,4 +1,4 @@
-"""Global logging configuration for Lec2Note2.
+"""Global logging configuration for Lec2Note.
 
 Call ``setup_logging()`` once at program start to enable consistent log format.
 """

requirements.txt

@@ -19,4 +19,5 @@ scikit-image==0.25.1
 imagehash==4.3.1
 tenacity==8.2.3
 streamlit>=1.35
+replicate
 tqdm