Upload 9 files

CITATIONS.md

@@ -0,0 +1,17 @@
+# CITATIONS
+
+## Packages
+- **gradio** (Apache 2.0)
+- **transformers** (Apache 2.0) — Wolf et al. 2020
+- **torch** (BSD-style)
+- **sentencepiece** (Apache 2.0)
+- **faster-whisper** (MIT)
+- **numpy**, **tqdm** (BSD/MIT)
+
+## Models (Hugging Face)
+- `facebook/bart-large-cnn` — summarization (MIT)
+- `google/flan-t5-large` — text generation/extraction (Apache 2.0)
+- `Systran/faster-whisper-small` — transcription (MIT, multilingual)
+
+## Data
+- `data/sample_transcript_en.txt` — small synthetic example for testing.

DEMO_SCRIPT.md

@@ -0,0 +1,25 @@
+# Demo Script (≤ 5 min) — MeetingNotes AI (EN)
+
+0:00–0:20 — Hook
+- Too many meetings, not enough time.
+- MeetingNotes AI: audio/transcript → Summary + Action Items + Decisions + minutes.md
+
+0:20–1:20 — Live demo
+- Paste `data/sample_transcript_en.txt` or upload a short .mp3
+- Click **Analyze**
+- Show Summary + Actions + Decisions
+
+1:20–2:20 — minutes.md
+- Download / open the generated file
+- Show the clean structure
+
+2:20–3:30 — How it works
+- Transcription: faster-whisper (small, multilingual)
+- Summarization: BART CNN
+- Extraction: Flan-T5 with a strict JSON prompt
+
+3:30–4:30 — Value at scale
+- Saves time, clarifies responsibilities, improves follow-up
+
+4:30–5:00 — CTA
+- Open-source, easy to deploy on Hugging Face Spaces

PROMPTS.md

@@ -0,0 +1,19 @@
+# PROMPTS — MeetingNotes AI (EN)
+
+## Summarization (BART via `pipeline("summarization")`)
+- No custom prompt (default pipeline).
+
+## Action Items & Decisions (Flan-T5)
+Template used in `nlp_utils.py`:
+```
+You are a meeting note-taking assistant.
+From the transcript below, extract:
+1) a concise list of "Action Items" (who does what, use infinitive verb, include deadline if any)
+2) a list of "Decisions" (short statements)
+
+Return strict JSON with this shape:
+{"actions": ["...","..."], "decisions": ["...","..."]}
+
+Transcript:
+{TRANSCRIPT}
+```

README.md

@@ -1,12 +1,49 @@
----
-title: AIPROJECT
-emoji: 🚀
-colorFrom: pink
-colorTo: indigo
-sdk: gradio
-sdk_version: 5.49.1
-app_file: app.py
-pinned: false
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# MeetingNotes AI — Meeting Summarizer (EN)
+
+**Goal:** Upload a **meeting audio** (mp3/wav) or paste a **transcript** and get:
+- ✅ a clear **Summary**
+- 🧱 **Action Items** (who does what, by when if stated)
+- 🧩 **Decisions**
+- 🗂️ a ready-to-share **minutes.md**
+
+**Tech**
+- Transcription: `faster-whisper` (multilingual; works for English **and** French audio)
+- Summarization: `facebook/bart-large-cnn`
+- Extraction (actions/decisions): `google/flan-t5-large`
+- UI: **Gradio**
+
+## Quickstart (local)
+
+```bash
+python -m venv .venv
+source .venv/bin/activate         # Windows: .venv\Scripts\activate
+pip install -r requirements.txt
+# (Optional) install ffmpeg for audio support:
+# macOS: brew install ffmpeg
+# Ubuntu/Debian: sudo apt-get install -y ffmpeg
+python app.py
+```
+
+## Deploy on Hugging Face Spaces (recommended)
+1. Create a **Gradio** Space
+2. Upload **all files** from this folder
+3. Wait for the build to finish (it reads `requirements.txt`)
+4. Test with a `.mp3/.wav` or paste a transcript
+
+## Structure
+```
+MeetingNotes_AI_EN/
+├─ app.py                   # Gradio UI (English)
+├─ nlp_utils.py             # Transcription + summarization + action/decision extraction
+├─ requirements.txt
+├─ PROMPTS.md               # Prompts and tool-usage log
+├─ CITATIONS.md             # Packages & models used
+├─ USER_GUIDE.md            # User guide (English)
+├─ DEMO_SCRIPT.md           # ≤ 5-min demo script (English)
+├─ data/
+│  └─ sample_transcript_en.txt
+└─ outputs/                 # generated minutes.md
+```
+
+## License
+MIT — 2025

USER_GUIDE.md

@@ -0,0 +1,26 @@
+# User Guide — MeetingNotes AI (EN)
+
+## Run the app
+- Local: see README (venv → pip install → `python app.py`)
+- Hugging Face Spaces: upload all files and open the Space
+
+## How to use
+1. **Choose your input**:
+   - Upload a **meeting audio** (.mp3/.wav) → click **Analyze** to transcribe.
+   - OR paste a **transcript**.
+
+2. **Outputs**:
+   - **Summary** (1–2 paragraphs)
+   - **Action Items** (list)
+   - **Decisions** (list)
+   - A downloadable **minutes.md** file
+
+3. **Tips**:
+   - Prefer clean recordings for audio (less noise).
+   - Multiple speakers are fine; diarization is not enabled by default.
+   - You can edit the transcript and re-run the extraction.
+
+## Troubleshooting
+- If audio fails: ensure **ffmpeg** is available.
+- If it’s slow on CPU: use a smaller Whisper model (tiny/base) or `flan-t5-base` in `nlp_utils.py`.
+- Transcript-only flow works without ffmpeg.

app.py

@@ -0,0 +1,54 @@
+import gradio as gr, os, json
+from nlp_utils import transcribe_audio, summarize, extract_actions_decisions, make_minutes_md
+
+OUT_DIR = "outputs"
+os.makedirs(OUT_DIR, exist_ok=True)
+
+def process(audio_file, transcript_text, meeting_title):
+    text = ""
+    if audio_file is not None:
+        text = transcribe_audio(audio_file)
+    if transcript_text and transcript_text.strip():
+        extra = transcript_text.strip()
+        text = (text + "\n" + extra).strip() if text else extra
+
+    if not text or len(text) < 40:
+        return "Please upload audio OR paste a transcript (≥ 40 characters).", "", [], [], None
+
+    resum = summarize(text)
+    ed = extract_actions_decisions(text)
+    actions = ed.get("actions", [])
+    decisions = ed.get("decisions", [])
+
+    title = meeting_title or "Meeting"
+    md = make_minutes_md(title, resum, actions, decisions)
+    md_path = os.path.join(OUT_DIR, "minutes.md")
+    with open(md_path, "w", encoding="utf-8") as f:
+        f.write(md)
+
+    actions_ht = [(a, "Action") for a in actions] if actions else []
+    decisions_ht = [(d, "Decision") for d in decisions] if decisions else []
+
+    return "Done ✅", resum, actions_ht, decisions_ht, md_path
+
+with gr.Blocks(title="MeetingNotes AI — Meeting Summarizer") as demo:
+    gr.Markdown("# MeetingNotes AI — Meeting Summarizer")
+    gr.Markdown("Upload **audio** or **paste a transcript**, then click **Analyze**. Multilingual audio supported (EN/FR).")
+
+    with gr.Row():
+        with gr.Column():
+            meeting_title = gr.Textbox(label="Meeting Title", value="Product Launch — Weekly")
+            audio = gr.Audio(label="Audio (mp3/wav)", sources=["upload"], type="filepath")
+            transcript = gr.Textbox(label="Transcript (optional if audio)", lines=10, placeholder="Paste here…")
+            btn = gr.Button("Analyze")
+        with gr.Column():
+            status = gr.Textbox(label="Status")
+            resume = gr.Textbox(label="Summary", lines=8)
+            actions = gr.HighlightedText(label="Action Items", combine_adjacent=True)
+            decisions = gr.HighlightedText(label="Decisions", combine_adjacent=True)
+            files = gr.File(label="Download minutes.md")
+
+    btn.click(process, inputs=[audio, transcript, meeting_title], outputs=[status, resume, actions, decisions, files])
+
+if __name__ == "__main__":
+    demo.launch()

nlp_utils.py

@@ -0,0 +1,101 @@
+import os, json, re, datetime
+from typing import Dict, List
+from transformers import pipeline
+from faster_whisper import WhisperModel
+
+# --------- Lazy singletons ---------
+_SUMMARIZER = None
+_EXTRACTOR = None
+_WHISPER = None
+
+def get_summarizer():
+    global _SUMMARIZER
+    if _SUMMARIZER is None:
+        _SUMMARIZER = pipeline("summarization", model="facebook/bart-large-cnn")
+    return _SUMMARIZER
+
+def get_extractor():
+    """Flan-T5 used for JSON-style action/decision extraction via text2text pipeline."""
+    global _EXTRACTOR
+    if _EXTRACTOR is None:
+        _EXTRACTOR = pipeline("text2text-generation", model="google/flan-t5-large", max_new_tokens=256)
+    return _EXTRACTOR
+
+def get_whisper(device: str = "auto"):
+    global _WHISPER
+    if _WHISPER is None:
+        # Small multilingual model: works for English + French audio
+        _WHISPER = WhisperModel("Systran/faster-whisper-small", device=device, compute_type="int8")
+    return _WHISPER
+
+# --------- Core ---------
+def transcribe_audio(audio_path: str) -> str:
+    model = get_whisper()
+    segments, info = model.transcribe(audio_path, beam_size=1)
+    text = " ".join(seg.text.strip() for seg in segments)
+    return text.strip()
+
+def _chunk(text: str, max_chars: int) -> List[str]:
+    parts, buf, size = [], [], 0
+    import re as _re
+    for sent in _re.split(r'(?<=[\.!\?])\s+', text):
+        if size + len(sent) > max_chars and buf:
+            parts.append(" ".join(buf)); buf, size = [], 0
+        buf.append(sent); size += len(sent) + 1
+    if buf: parts.append(" ".join(buf))
+    return parts
+
+def summarize(text: str) -> str:
+    summarizer = get_summarizer()
+    chunks = _chunk(text, 2200)
+    partials = [summarizer(ch, do_sample=False)[0]["summary_text"] for ch in chunks]
+    merged = " ".join(partials)
+    final = summarizer(merged, do_sample=False, max_length=200, min_length=60)[0]["summary_text"]
+    return final
+
+def extract_actions_decisions(text: str) -> Dict[str, List[str]]:
+    prompt = f"""You are a meeting note-taking assistant.
+From the transcript below, extract:
+1) a concise list of "Action Items" (who does what, use infinitive verb, include deadline if mentioned)
+2) a list of "Decisions" (short statements)
+
+Return strict JSON with this shape:
+{{"actions": ["...","..."], "decisions": ["...","..."]}}
+
+Transcript:
+{text[:7000]}
+"""
+    gen = get_extractor()
+    out = gen(prompt)[0]["generated_text"]
+    try:
+        data = json.loads(out)
+        actions = [s.strip() for s in data.get("actions", []) if s.strip()]
+        decisions = [s.strip() for s in data.get("decisions", []) if s.strip()]
+        return {"actions": actions, "decisions": decisions}
+    except Exception:
+        # Fallback heuristic if JSON parsing fails
+        actions, decisions = [], []
+        for line in text.splitlines():
+            if re.search(r"(?i)\b(action|todo|to do):", line):
+                actions.append(re.sub(r"(?i)^.*?:\s*", "", line).strip())
+            if re.search(r"(?i)\b(decision|decisions):", line):
+                decisions.append(re.sub(r"(?i)^.*?:\s*", "", line).strip())
+        return {"actions": actions, "decisions": decisions}
+
+def make_minutes_md(title: str, summary: str, actions: List[str], decisions: List[str]) -> str:
+    now = datetime.datetime.now().strftime("%Y-%m-%d %H:%M")
+    lines = [
+        f"# {title} — Minutes",
+        f"_Generated on {now}_",
+        "",
+        "## Summary",
+        summary.strip() if summary else "—",
+        "",
+        "## Action Items",
+        *[f"- [ ] {a}" for a in (actions or ["—"])],
+        "",
+        "## Decisions",
+        *[f"- {d}" for d in (decisions or ["—"])],
+        "",
+    ]
+    return "\n".join(lines)

requirements.txt

@@ -0,0 +1,7 @@
+gradio>=4.44.0
+transformers>=4.44.0
+torch>=2.2.0
+sentencepiece>=0.1.99
+faster-whisper>=1.0.0
+numpy>=1.26.4
+tqdm>=4.66.4

sample_transcript_en.txt

@@ -0,0 +1,7 @@
+[00:00] Alice: Welcome everyone. Goal: finalize the launch plan.
+[00:15] Bob: We still need visuals for the campaign.
+[00:30] Chloe: Design team will share a first draft on Wednesday.
+[00:45] Alice: Decision: we keep the budget at $20k.
+[01:00] Bob: Action: I'll contact the media agency today.
+[01:15] Chloe: Action: I'll prepare a checklist for the product page.
+[01:30] Alice: Next meeting Friday 10am. End.