chore: set max line length to 58 chars in Ruff config

- Integrate chunking into LocalAI client to avoid GPU OOM on long audio.
- Split long files into overlapping chunks; transcribe each chunk; merge segments with corrected timestamps.
- Auto-enable chunking when audio duration > LOCALAI_MAX_SINGLE_REQUEST_DURATION (default 300s).
- Add env variables:
    LOCALAI_CHUNK_DURATION (default 180)
    LOCALAI_CHUNK_OVERLAP (default 2)
    LOCALAI_MAX_SINGLE_REQUEST_DURATION (default 300)
- Add unit and integration tests for chunking logic.
- Confirmed working end-to-end with vibevoice-cpp-asr on 88-minute file.

README.md

@@ -3,7 +3,7 @@
 ScrAIbe is a transcription and summarization service that:
 
 - Sends audio to a LocalAI server running vibevoice.cpp for transcription and speaker diarization.
-- Optionally uses a second LLM to generate a detailed, structured summary.
+- Optionally uses a second LLM to generate a structured summary.
 - Provides:
   - A web GUI for uploading audio and receiving transcripts via email.
   - A CLI and Python API for direct integration.
@@ -29,12 +29,12 @@ For more information: https://apstrom.ca
   - Jobs are queued and processed in the background (Celery + Redis).
   - Emails:
     - Immediate confirmation with queue position.
-    - Final transcript (MD + JSON) when ready.
-    - Summary as MD file (if requested).
+    - Final transcript (MD + DOCX + JSON) when ready.
+    - Summary as MD + DOCX (if requested).
     - Error notification if processing fails.
 - File formats:
-  - Transcript: .md and .docx
-  - Summary (if requested): .md and .docx
+  - Transcript: .md and .docx (line-numbered, no cover page)
+  - Summary (if requested): .md and .docx (no line numbering, no cover page)
   - Full structured output: .json
 - Customizable branding:
   - Web GUI title, logo, and accent color via environment variables.
@@ -256,12 +256,13 @@ Email subject lines (customizable):
 Output files (async web GUI):
 
 When a job completes, the user receives:
+
 - Transcript:
   - .md file
-  - .docx file
+  - .docx file (line-numbered, no cover page)
 - Summary (if requested):
   - .md file
-  - .docx file
+  - .docx file (no line numbering, no cover page)
 - JSON:
   - Structured transcript with diarization and metadata
 

misc/header.html

@@ -9,46 +9,56 @@
     <link href="https://fonts.googleapis.com/css2?family=Cormorant+Garamond:wght@400;700&display=swap" rel="stylesheet">
     
     <style>
-        .header-container {{
+        .header-wrapper {{
+            display: flex;
+            flex-direction: column;
+            align-items: center;
+            justify-content: center;
+            padding: 20px 20px 0;
+            box-sizing: border-box;
+        }}
+
+        .logo-container {{
             display: flex;
             align-items: center;
-            justify-content: space-between;
-            padding: 20px 40px;
-            box-sizing: border-box;
+            justify-content: center;
+            margin-bottom: 10px;
+        }}
+
+        .logo {{
+            width: 75px;
+            height: auto;
+            display: block;
         }}
 
         .header-title {{
             font-family: 'Cormorant Garamond', serif;
-            font-size: 42px;
+            font-size: 45px;
             font-weight: bold;
             color: {accent_color};
             margin: 0;
             position: relative;
             padding: 0.4em 0;
-            flex: 1;
+            text-align: center;
+            max-width: 90%;
         }}
 
-        .header-title::before, .header-title::after {{
+        .header-title::before,
+        .header-title::after {{
             content: "";
             position: absolute;
             height: 2px;
-            width: 100%;
+            width: 80%;
             background-color: {accent_color};
-            left: 0;
+            left: 10%;
         }}
 
-        .header-title::before {{ top: 0.4em; }}
-        .header-title::after {{ bottom: 0.4em; }}
-
-        .logo-container {{
-            flex-shrink: 0;
-            margin-left: 20px;
+        .header-title::before {{
+            top: 0.4em;
         }}
 
-        .logo {{
-            height: 80px;
-            width: auto;
-            display: block;
+        .header-title::after {{
+            bottom: 0.4em;
         }}
 
         .header-description {{
@@ -71,29 +81,18 @@
         }}
 
         @media (max-width: 768px) {{
-            .header-container {{
-                flex-direction: column;
-                align-items: center;
-                padding: 15px;
-                gap: 10px;
-            }}
-
             .header-title {{
-                font-size: 28px;
-                text-align: center;
+                font-size: 31px;
             }}
 
-            .header-title::before, .header-title::after {{
+            .header-title::before,
+            .header-title::after {{
                 width: 80%;
                 left: 10%;
             }}
 
-            .logo-container {{
-                margin-left: 0;
-            }}
-
             .logo {{
-                height: 60px;
+                width: 50px;
             }}
 
             .header-description {{
@@ -103,13 +102,13 @@
     </style>
 </head>
 <body>
-    <div class="header-container">
-        <h1 class="header-title">{webui_title}</h1>
+    <div class="header-wrapper">
         <div class="logo-container">
             <a href="{header_logo_url}">
                 <img src="{header_logo_src}" alt="{webui_title}" class="logo">
             </a>
         </div>
+        <h1 class="header-title">{webui_title}</h1>
     </div>
     <div class="header-description">
         <p>

misc/upload_notification_template.html

@@ -13,7 +13,6 @@
         <h1 style="color:{accent_color};">Upload Successful</h1>
         <p>Dear user,</p>
         <p>Your file has been successfully uploaded and is now in our processing queue. This means that our system has received your file, and it is waiting to be processed. We will handle your file as soon as possible.</p>
-        <p class="success-message">Your current position in the queue is: <span style="color:{accent_color}; font-weight:bold;">{queue_position}</span>. This is the order in which your file will be processed. We appreciate your patience as we work through the queue.</p>
         <p>We will notify you once your file has been processed. If you have any urgent needs or further questions, feel free to reach out to our support team.</p>
         <div class="contact">
             <p>You can contact our support team at <a href="mailto:{contact_email}" style="color:{accent_color};">{contact_email}</a>. Please note that our support team is here to help with any questions or issues you might have.</p>

pyproject.toml

@@ -72,6 +72,9 @@ scraibe = "scraibe.cli:cli"
 [tool.poetry.extras]
 app = ["scraibe-webui"]
 
+[tool.ruff]
+line-length = 58
+
 [tool.ruff.lint.extend-per-file-ignores]
 "__init__.py" = ["E402", "F403", "F401"]
 "scraibe/misc.py" = ["E722"]

scraibe/audio.py

@@ -7,13 +7,21 @@ Simplified audio processor for ScrAIbe.
 Previously this used torch and pyannote-style processing. In the LocalAI-backed
 version, we primarily pass files to the API, but we keep a lightweight helper
 for backward compatibility.
+
+Now also includes utilities for chunking long audio into smaller segments
+to avoid GPU memory limits when using vibevoice-cpp on LocalAI.
 """
 
+import json
+import os
+import tempfile
 from subprocess import CalledProcessError, run
 import numpy as np
 
 SAMPLE_RATE = 16000
 NORMALIZATION_FACTOR = 32768.0
+DEFAULT_CHUNK_DURATION = 180.0  # seconds
+DEFAULT_CHUNK_OVERLAP = 2.0    # seconds
 
 
 class AudioProcessor:
@@ -106,3 +114,109 @@ class AudioProcessor:
 
     def __repr__(self) -> str:
         return f"AudioProcessor(waveform_len={len(self.waveform)}, sr={self.sr})"
+
+
+def get_audio_duration(file_path: str) -> float:
+    """
+    Get the duration of an audio file in seconds using ffprobe.
+
+    Args:
+        file_path: Path to the audio file.
+
+    Returns:
+        Duration in seconds as a float.
+
+    Raises:
+        RuntimeError: If ffprobe fails.
+    """
+    cmd = [
+        "ffprobe",
+        "-v", "error",
+        "-show_entries", "format=duration",
+        "-of", "json",
+        file_path,
+    ]
+    try:
+        result = run(cmd, capture_output=True, text=True, check=True)
+        data = json.loads(result.stdout)
+        return float(data["format"]["duration"])
+    except (CalledProcessError, json.JSONDecodeError, KeyError) as e:
+        raise RuntimeError(f"Failed to get audio duration for {file_path}: {e}")
+
+
+def split_audio_into_chunks(
+    input_path: str,
+    max_duration: float = DEFAULT_CHUNK_DURATION,
+    overlap: float = DEFAULT_CHUNK_OVERLAP,
+    output_format: str = "wav",
+    sample_rate: int = 24000,
+) -> list:
+    """
+    Split a long audio file into overlapping chunks using ffmpeg.
+
+    Args:
+        input_path: Path to the input audio file.
+        max_duration: Maximum duration of each chunk in seconds.
+        overlap: Overlap duration in seconds between consecutive chunks.
+        output_format: Output format (e.g., 'wav').
+        sample_rate: Sample rate for output chunks.
+
+    Returns:
+        List of dicts:
+            [{"path": "chunk.wav", "start": 0.0, "end": 180.0}, ...]
+        Files must be cleaned up by the caller.
+    """
+    duration = get_audio_duration(input_path)
+
+    # If file is shorter than max_duration, no need to split
+    if duration <= max_duration:
+        return [{"path": input_path, "start": 0.0, "end": duration}]
+
+    chunks = []
+    start = 0.0
+    chunk_id = 0
+
+    while start < duration:
+        chunk_end = min(start + max_duration, duration)
+        chunk_duration = chunk_end - start
+
+        tmp = tempfile.NamedTemporaryFile(
+            delete=False,
+            suffix=f".{output_format}",
+            prefix="scraibe_chunk_",
+        )
+        chunk_path = tmp.name
+        tmp.close()
+
+        cmd = [
+            "ffmpeg",
+            "-y",
+            "-nostdin",
+            "-ss", str(start),
+            "-i", input_path,
+            "-t", str(chunk_duration),
+            "-ar", str(sample_rate),
+            "-ac", "1",
+            "-c:a", "pcm_s16le",
+            chunk_path,
+        ]
+        try:
+            run(cmd, capture_output=True, check=True)
+        except CalledProcessError as e:
+            # Clean up on error
+            if os.path.exists(chunk_path):
+                os.remove(chunk_path)
+            raise RuntimeError(
+                f"Failed to create audio chunk {chunk_id} for {input_path}: {e.stderr.decode()}"
+            )
+
+        chunks.append({
+            "path": chunk_path,
+            "start": start,
+            "end": chunk_end,
+        })
+
+        start += max_duration - overlap
+        chunk_id += 1
+
+    return chunks

scraibe/email_sender.py

@@ -8,20 +8,22 @@ Template placeholders are primarily filled via environment variables.
 """
 
 import base64
+import json
+import logging
 import os
 import re
 import smtplib
-import logging
 from email import encoders
 from email.mime.base import MIMEBase
 from email.mime.multipart import MIMEMultipart
 from email.mime.text import MIMEText
-from typing import List, Optional, Dict, Any
+from typing import Any, Dict, List, Optional
 
 from docx import Document
-from docx.shared import Inches, Pt
-from docx.oxml.ns import qn
 from docx.oxml import OxmlElement
+from docx.oxml.ns import qn
+from docx.shared import Inches, Pt
+from docx.enum.text import WD_ALIGN_PARAGRAPH
 
 logger = logging.getLogger("scraibe.email_sender")
 
@@ -214,27 +216,34 @@ def send_email(
     if not to_list:
         raise EmailError("No valid 'To' email addresses provided.")
 
-    # Build message
-    msg = MIMEMultipart("alternative")
-    msg["From"] = cfg["from_address"]
-    msg["To"] = ", ".join(to_list)
-    if cc_list:
-        msg["Cc"] = ", ".join(cc_list)
-    msg["Subject"] = subject
+    # Ensure subject is never blank
+    if not subject or not subject.strip():
+        logger.warning("Subject was blank or missing; using default subject.")
+        subject = "ScrAIbe: Your transcript is ready"
 
-    # Attach plain text
-    msg.attach(MIMEText(body, "plain"))
+    subject = subject.strip()
 
-    # Attach HTML if provided
+    has_attachments = bool(attachments)
+
+    # Build the text/HTML part (alternative)
+    alt = MIMEMultipart("alternative")
+    alt.attach(MIMEText(body, "plain"))
     if html:
-        msg.attach(MIMEText(html, "html"))
+        alt.attach(MIMEText(html, "html"))
 
-    # Attach files in a separate multipart/mixed part
-    if attachments:
-        mixed = MIMEMultipart("mixed")
-        mixed.attach(msg)
-        msg = mixed
+    if has_attachments:
+        # Outer message: multipart/mixed with headers
+        msg = MIMEMultipart("mixed")
+        msg["From"] = cfg["from_address"]
+        msg["To"] = ", ".join(to_list)
+        if cc_list:
+            msg["Cc"] = ", ".join(cc_list)
+        msg["Subject"] = subject
 
+        # Attach the alternative (text/HTML) part
+        msg.attach(alt)
+
+        # Attach files
         for file_path in attachments:
             if not os.path.isfile(file_path):
                 logger.warning("Attachment file not found, skipping: %s", file_path)
@@ -253,6 +262,14 @@ def send_email(
                     msg.attach(part)
             except Exception as e:
                 logger.warning("Failed to attach file %s: %s", file_path, e)
+    else:
+        # No attachments: use the alternative part as the root message
+        msg = alt
+        msg["From"] = cfg["from_address"]
+        msg["To"] = ", ".join(to_list)
+        if cc_list:
+            msg["Cc"] = ", ".join(cc_list)
+        msg["Subject"] = subject
 
     # Connect and send
     try:
@@ -273,9 +290,10 @@ def send_email(
         )
         server.quit()
         logger.info(
-            "Email sent to %s (CC: %s)",
+            "Email sent to %s (CC: %s) with subject: %s",
             to_list,
             cc_list or "None",
+            subject,
         )
         return True
 
@@ -284,87 +302,331 @@ def send_email(
         raise EmailError(f"Failed to send email: {e}")
 
 
+# ------------ DOCX helpers ------------
+
+# Namespaces
+W_NS = "http://schemas.openxmlformats.org/wordprocessingml/2006/main"
+
+
+def _set_element_attr(elem, attr, value):
+    elem.set(f"{{{W_NS}}}{attr}", str(value))
+
+
+def _create_transcript_section_properties(section):
+    """
+    Configure the section properties for transcript DOCX:
+    - Margins: 1 inch all sides
+    - Single column layout
+    - No built-in line numbering (we embed line numbers as text for portability)
+    - Remove document grid to avoid off-by-one line numbering
+    """
+    sectPr = section._sectPr
+
+    # Margins: 1 inch = 1440 twips
+    pgMar = sectPr.find(f"{{{W_NS}}}pgMar")
+    if pgMar is None:
+        pgMar = OxmlElement("w:pgMar")
+        sectPr.append(pgMar)
+    _set_element_attr(pgMar, "top", "1440")
+    _set_element_attr(pgMar, "right", "1440")
+    _set_element_attr(pgMar, "bottom", "1440")
+    _set_element_attr(pgMar, "left", "1440")
+    _set_element_attr(pgMar, "header", "720")
+    _set_element_attr(pgMar, "footer", "720")
+    _set_element_attr(pgMar, "gutter", "0")
+
+    # Ensure single column (no multi-column layout)
+    cols = sectPr.find(f"{{{W_NS}}}cols")
+    if cols is not None:
+        _set_element_attr(cols, "num", "1")
+        _set_element_attr(cols, "space", "720")
+
+    # Remove document grid entirely
+    for docGrid in sectPr.findall(f"{{{W_NS}}}docGrid"):
+        sectPr.remove(docGrid)
+
+    # Remove any built-in line numbering; we will use text-based line numbers
+    for lnNumType in sectPr.findall(f"{{{W_NS}}}lnNumType"):
+        sectPr.remove(lnNumType)
+
+
+def _add_transcript_paragraph(doc, line_text, line_number):
+    """
+    Add a single transcript line as a paragraph with an embedded line number.
+    Uses a left tab stop so the line number appears in the left margin area,
+    independent of built-in line numbering, ensuring consistent behavior
+    across Word, LibreOffice, Google Docs, etc.
+    """
+    line_text = line_text.strip()
+    if not line_text:
+        return
+
+    p = doc.add_paragraph()
+
+    # Set up paragraph formatting:
+    # - No left indent; we control spacing via tab stop
+    # - Single line spacing, no extra before/after
+    pPr = p._p.get_or_add_pPr()
+
+    # Remove any default indent
+    pPr.find(f"{{{W_NS}}}ind") and pPr.remove(pPr.find(f"{{{W_NS}}}ind"))
+
+    # Define a left tab stop for line numbers (e.g. 360 twips ≈ 0.25")
+    tabs = OxmlElement("w:tabs")
+    tab = OxmlElement("w:tab")
+    tab.set("{http://schemas.openxmlformats.org/wordprocessingml/2006/main}val", "left")
+    tab.set("{http://schemas.openxmlformats.org/wordprocessingml/2006/main}pos", "360")
+    tabs.append(tab)
+    pPr.append(tabs)
+
+    spacing = OxmlElement("w:spacing")
+    _set_element_attr(spacing, "before", "0")
+    _set_element_attr(spacing, "after", "0")
+    _set_element_attr(spacing, "line", "360")       # 1.5 line spacing (12pt * 1.5 = 18pt → 360 twips)
+    _set_element_attr(spacing, "lineRule", "auto")
+    pPr.append(spacing)
+
+    # Try to match: [00:00] SPEAKER 1: content
+    m = re.match(r"\[(\d+:\d+(?::\d+)?)\]\s*(.+?):\s*(.*)", line_text)
+
+    # Line number run (no underline)
+    run_ln = p.add_run(str(line_number))
+    run_ln.font.name = "Courier"
+    run_ln.font.size = Pt(12)
+    run_ln.underline = False
+
+    # Tab + spaces between line number and content
+    # - 2 base spaces + 7 more for first line of speaker turn
+    # - 2 base spaces + 3 more for continuation lines
+    if m:
+        extra_spaces = "       "  # 7 spaces for speaker lines
+    else:
+        extra_spaces = "   "  # 3 spaces for continuation lines
+
+    run_tab = p.add_run("\t  " + extra_spaces)
+    run_tab.font.name = "Courier"
+    run_tab.font.size = Pt(12)
+    run_tab.underline = False
+
+    if m:
+        ts, speaker, content = m.groups()
+        label_text = f"[{ts}] {speaker.upper()}:"
+
+        # Label run (underline)
+        run_label = p.add_run(label_text)
+        run_label.underline = True
+        run_label.font.name = "Courier"
+        run_label.font.size = Pt(12)
+
+        # Space run (no underline)
+        run_space = p.add_run(" ")
+        run_space.underline = False
+        run_space.font.name = "Courier"
+        run_space.font.size = Pt(12)
+
+        # Content run (no underline)
+        run_txt = p.add_run(content.strip())
+        run_txt.underline = False
+        run_txt.font.name = "Courier"
+        run_txt.font.size = Pt(12)
+    else:
+        # Non-standard line: plain text
+        run = p.add_run(line_text)
+        run.underline = False
+        run.font.name = "Courier"
+        run.font.size = Pt(12)
+
+
+# ------------ Public DOCX functions ------------
+
 def create_transcript_docx(text: str, filename: str):
     """
-    Create a .docx transcript with:
-      - 1.5" left margin, 1" right margin
-      - 12pt Courier
-      - Continuous line numbering on the left
-      - Speaker names capitalized and indented; spoken text further indented
-      - No section headings; use bold/underline only.
+    Create a transcript DOCX with:
+    - 1" margins on all sides
+    - 12pt Courier font
+    - Embedded line numbers starting at 1 on the first page
+      (portable across Word, LibreOffice, Google Docs)
+    - Line numbers reflect visual lines on the page, not speaker turns.
+    - Proper formatting for timestamps and speaker labels
     """
     doc = Document()
 
-    # Set margins via section properties
-    section = doc.sections[0]
-    section.left_margin = Inches(1.5)
-    section.right_margin = Inches(1.0)
-    section.top_margin = Inches(1.0)
-    section.bottom_margin = Inches(1.0)
-
-    # Enable continuous line numbering on the left
-    sectPr = section._sectPr
-    lnNumType = sectPr.find(qn("w:lnNumType"))
-    if lnNumType is None:
-        lnNumType = OxmlElement("w:lnNumType")
-        sectPr.append(lnNumType)
-    lnNumType.set(qn("w:start"), "continuous")
-    lnNumType.set(qn("w:countBy"), "1")
-
-    # Default font
+    # Set base font (Normal style)
     style = doc.styles["Normal"]
-    font = style.font
-    font.name = "Courier"
-    font.size = Pt(12)
+    style.font.name = "Courier"
+    style.font.size = Pt(12)
 
-    # Parse lines
-    lines = text.strip().split("\n")
-    for line in lines:
+    # Remove any default paragraphs (ensure no phantom first line)
+    body = doc.element.body
+    for p in list(body.findall(f"{{{W_NS}}}p")):
+        body.remove(p)
+
+    # Configure section properties (margins, no built-in line numbering)
+    _create_transcript_section_properties(doc.sections[0])
+
+    # Max characters per visual line (content only; total line including line number and spaces <= 60)
+    max_chars = 54
+
+    # Lines per page before restarting numbering
+    lines_per_page = 29
+
+    # Current line counter for visual lines
+    line_number = 0
+
+    # Split transcript into logical lines
+    logical_lines = text.strip().splitlines()
+
+    def ensure_new_page_if_needed():
+        nonlocal line_number
+        if line_number >= lines_per_page:
+            # Insert a page break paragraph (no line number, no text)
+            p_break = doc.add_paragraph()
+            pPr = p_break._p.get_or_add_pPr()
+
+            # Clear any inherited formatting
+            for child in list(pPr):
+                tag = child.tag.split("}")[-1] if "}" in child.tag else child.tag
+                if tag in ("tabs", "spacing", "ind"):
+                    pPr.remove(child)
+
+            # Standard page break via paragraph property
+            page_break = OxmlElement("w:pageBreak")
+            page_break.set("{http://schemas.openxmlformats.org/wordprocessingml/2006/main}val", "1")
+            pPr.append(page_break)
+
+            # Reset line counter for new page
+            line_number = 0
+
+    for line in logical_lines:
         line = line.strip()
         if not line:
             continue
 
-        # Try to parse: [00:00] SPEAKER: text
+        # Try to match: [00:00] SPEAKER 1: content
         m = re.match(r"\[(\d+:\d+(?::\d+)?)\]\s*(.+?):\s*(.*)", line)
         if m:
             ts, speaker, content = m.groups()
-            # Speaker line: bold, underlined, indented
-            p_spk = doc.add_paragraph()
-            p_spk.paragraph_format.left_indent = Inches(0.25)
-            run_spk = p_spk.add_run(f"[{ts}] {speaker.upper()}")
-            run_spk.bold = True
-            run_spk.underline = True
-            run_spk.font.name = "Courier"
-            run_spk.font.size = Pt(12)
-
-            # Spoken text line: further indented
-            p_txt = doc.add_paragraph()
-            p_txt.paragraph_format.left_indent = Inches(0.5)
-            run_txt = p_txt.add_run(content.strip())
-            run_txt.font.name = "Courier"
-            run_txt.font.size = Pt(12)
+            label_text = f"[{ts}] {speaker.upper()}:"
+            content = content.strip()
         else:
-            # Fallback for non-standard lines
-            p = doc.add_paragraph()
-            run = p.add_run(line)
-            run.font.name = "Courier"
-            run.font.size = Pt(12)
+            label_text = ""
+            content = line.strip()
 
+        # Split content into visual lines at word boundaries
+        content_lines = []
+        words = content.split()
+        current = ""
+        for w in words:
+            if len(current) == 0:
+                current = w
+            elif len(current) + 1 + len(w) <= max_chars:
+                current += " " + w
+            else:
+                content_lines.append(current)
+                current = w
+        if current:
+            content_lines.append(current)
+
+        # First visual line: include label if present, ensuring total <= max_chars
+        if content_lines:
+            ensure_new_page_if_needed()
+
+            first_content = content_lines.pop(0)
+            if label_text:
+                prefix = label_text + " "
+                # If too long, trim first_content at word boundary
+                if len(prefix) + len(first_content) > max_chars:
+                    allowed = max_chars - len(prefix)
+                    if allowed < 1:
+                        allowed = 1
+                    # Truncate at word boundary
+                    candidate = first_content[:allowed]
+                    last_space = candidate.rfind(" ")
+                    if last_space > 0:
+                        candidate = candidate[:last_space]
+                    first_content = candidate
+                first_line_text = prefix + first_content
+            else:
+                first_line_text = first_content
+
+            line_number += 1
+            _add_transcript_paragraph(doc, first_line_text, line_number=line_number)
+
+        # Subsequent visual lines: no label, just content
+        for cl in content_lines:
+            ensure_new_page_if_needed()
+            line_number += 1
+            _add_transcript_paragraph(doc, cl, line_number=line_number)
+
+    # Add page numbers to footer: "X of Y" (bottom left)
+    section = doc.sections[0]
+    footer = section.footer
+    footer.is_linked_to_previous = False
+    footer_para = footer.paragraphs[0] if footer.paragraphs else footer.add_paragraph()
+    footer_para.alignment = WD_ALIGN_PARAGRAPH.CENTER
+
+    # Clear any existing content
+    for r in footer_para.runs:
+        r.text = ""
+
+    def add_field(run, code):
+        fldChar = OxmlElement("w:fldChar")
+        fldChar.set(qn("w:fldCharType"), "begin")
+        run._r.append(fldChar)
+
+        instrText = OxmlElement("w:instrText")
+        instrText.set(qn("xml:space"), "preserve")
+        instrText.text = code
+        run._r.append(instrText)
+
+        fldCharEnd = OxmlElement("w:fldChar")
+        fldCharEnd.set(qn("w:fldCharType"), "end")
+        run._r.append(fldCharEnd)
+
+    run_page = footer_para.add_run()
+    add_field(run_page, " PAGE ")
+
+    run_of = footer_para.add_run(" of ")
+
+    run_total = footer_para.add_run()
+    add_field(run_total, " NUMPAGES ")
+
+    # Save
     doc.save(filename)
 
 
 def create_summary_docx(text: str, filename: str):
     """
-    Create a .docx summary with consistent font.
-    No section headings; use bold/underline only.
+    Create a summary DOCX with:
+    - 1" margins on all sides
+    - 12pt Courier font
+    - No line numbering
     """
     doc = Document()
-    style = doc.styles["Normal"]
-    font = style.font
-    font.name = "Courier"
-    font.size = Pt(12)
 
-    for line in text.splitlines():
+    # Base font
+    style = doc.styles["Normal"]
+    style.font.name = "Courier"
+    style.font.size = Pt(12)
+
+    # Margins: 1 inch all sides
+    for section in doc.sections:
+        section.left_margin = Inches(1.0)
+        section.right_margin = Inches(1.0)
+        section.top_margin = Inches(1.0)
+        section.bottom_margin = Inches(1.0)
+
+    # Remove default paragraph
+    body = doc.element.body
+    for p in list(body.findall(f"{{{W_NS}}}p")):
+        body.remove(p)
+
+    # Add summary content
+    lines = text.strip().splitlines()
+    for line in lines:
+        line = line.strip()
+        if not line:
+            continue
         p = doc.add_paragraph(line)
         p.paragraph_format.space_after = Pt(4)
 

scraibe/localai_client.py

@@ -9,11 +9,21 @@ It replaces the previous local Whisper + Pyannote pipeline by sending
 audio files to the /v1/audio/diarization endpoint and mapping the
 response into the same Transcript format used by the UI.
 
+For long audio files, it can chunk the input to avoid GPU OOM errors.
+
 Environment Variables:
     LOCALAI_API_URL: (required) Base URL of the LocalAI server
                      (e.g., http://localhost:8080)
     LOCALAI_API_KEY: (optional) API key, if configured
     LOCALAI_MODEL:   (optional) Model name to use (default: vibevoice-diarize)
+
+    Chunking / long audio (all optional):
+    LOCALAI_CHUNK_DURATION:               Max duration of each chunk in seconds
+                                           (default: 180.0)
+    LOCALAI_CHUNK_OVERLAP:                Overlap between consecutive chunks in seconds
+                                           (default: 2.0)
+    LOCALAI_MAX_SINGLE_REQUEST_DURATION:  If audio duration exceeds this, chunking
+                                           is enabled automatically (default: 300.0)
 """
 
 import os
@@ -24,6 +34,8 @@ from typing import Dict, List, Any, Optional
 
 import httpx
 
+from .audio import get_audio_duration, split_audio_into_chunks
+
 logger = logging.getLogger("scraibe.localai_client")
 
 
@@ -41,14 +53,20 @@ class LocalAIClient:
       - Upload audio file as multipart/form-data.
       - Parse diarization + transcription response (verbose_json).
       - Map response into the same structure expected by Scraibe's Transcript.
+      - For long audio: chunk, transcribe each chunk, merge results.
     """
 
+    # Default thresholds for chunking long audio to avoid GPU OOM.
+    # These can be overridden via environment or at call time.
+    DEFAULT_CHUNK_DURATION = 180.0  # seconds
+    DEFAULT_CHUNK_OVERLAP = 2.0    # seconds
+
     def __init__(
         self,
         api_url: Optional[str] = None,
         api_key: Optional[str] = None,
         model: Optional[str] = None,
-        timeout: float = 600.0,
+        timeout: float = 3600.0,
     ):
         """
         Args:
@@ -82,6 +100,55 @@ class LocalAIClient:
             follow_redirects=True,
         )
 
+    @staticmethod
+    def _env_float(var: str, default: float) -> float:
+        """
+        Read a float from environment with a fallback default.
+        """
+        val = (os.getenv(var) or "").strip()
+        if val == "":
+            return default
+        try:
+            return float(val)
+        except ValueError:
+            logger.warning(
+                "Invalid value for %s: %s; using default %s", var, val, default
+            )
+            return default
+
+    def _effective_chunk_duration(self, provided: Optional[float]) -> float:
+        """
+        Resolve chunk_duration using this precedence:
+          1) provided argument
+          2) LOCALAI_CHUNK_DURATION env
+          3) class default
+        """
+        if provided is not None:
+            return provided
+        return self._env_float("LOCALAI_CHUNK_DURATION", self.DEFAULT_CHUNK_DURATION)
+
+    def _effective_chunk_overlap(self, provided: Optional[float]) -> float:
+        """
+        Resolve chunk_overlap:
+          1) provided argument
+          2) LOCALAI_CHUNK_OVERLAP env
+          3) class default
+        """
+        if provided is not None:
+            return provided
+        return self._env_float("LOCALAI_CHUNK_OVERLAP", self.DEFAULT_CHUNK_OVERLAP)
+
+    def _effective_max_single_request_duration(self, provided: Optional[float]) -> float:
+        """
+        Resolve max_single_request_duration:
+          1) provided argument
+          2) LOCALAI_MAX_SINGLE_REQUEST_DURATION env
+          3) default 300.0
+        """
+        if provided is not None:
+            return provided
+        return self._env_float("LOCALAI_MAX_SINGLE_REQUEST_DURATION", 300.0)
+
     def close(self):
         """Close the underlying HTTP client."""
         self._client.close()
@@ -107,6 +174,10 @@ class LocalAIClient:
         include_text: Optional[bool] = None,
         verbose: bool = False,
         return_raw: bool = False,
+        use_chunking: Optional[bool] = None,
+        chunk_duration: Optional[float] = None,
+        chunk_overlap: Optional[float] = None,
+        max_single_request_duration: Optional[float] = None,
         **_ignored,
     ) -> Dict[str, Any]:
         """
@@ -114,6 +185,8 @@ class LocalAIClient:
         - A normalized dict with segments, speakers, transcripts.
         - Optionally, the raw verbose_json response (for JSON export).
 
+        For long audio, it can automatically chunk the file to avoid GPU OOM.
+
         Args:
             audio_path: Path to the audio file.
             language: Language hint, forwarded if set.
@@ -129,6 +202,93 @@ class LocalAIClient:
                           Defaults to True.
             verbose: If True, prints progress messages.
             return_raw: If True, also return the raw API response in 'raw_result'.
+            use_chunking: Whether to enable chunking for long audio.
+                          If None, enabled automatically based on duration.
+            chunk_duration: Max duration per chunk in seconds.
+                            Falls back to LOCALAI_CHUNK_DURATION env, then 180.0.
+            chunk_overlap: Overlap between chunks in seconds.
+                           Falls back to LOCALAI_CHUNK_OVERLAP env, then 2.0.
+            max_single_request_duration: If audio duration exceeds this, chunking
+                                         is enabled (unless explicitly disabled).
+                                         Falls back to LOCALAI_MAX_SINGLE_REQUEST_DURATION
+                                         env, then 300.0.
+        """
+        if verbose:
+            print("Starting diarization and transcription via LocalAI.")
+
+        logger.info("diarize_and_transcribe requested for: %s", audio_path)
+
+        # Resolve chunking parameters with environment support
+        chunk_duration = self._effective_chunk_duration(chunk_duration)
+        chunk_overlap = self._effective_chunk_overlap(chunk_overlap)
+        max_single = self._effective_max_single_request_duration(max_single_request_duration)
+
+        if use_chunking is None:
+            try:
+                duration = get_audio_duration(audio_path)
+            except RuntimeError:
+                duration = None
+
+            use_chunking = (duration is not None and duration > max_single)
+            logger.info(
+                "Auto-chunking decision: duration=%s, threshold=%s, use_chunking=%s",
+                duration,
+                max_single,
+                use_chunking,
+            )
+
+        if use_chunking:
+            return self._diarize_and_transcribe_chunked(
+                audio_path=audio_path,
+                language=language,
+                num_speakers=num_speakers,
+                min_speakers=min_speakers,
+                max_speakers=max_speakers,
+                clustering_threshold=clustering_threshold,
+                min_duration_on=min_duration_on,
+                min_duration_off=min_duration_off,
+                response_format=response_format,
+                include_text=include_text,
+                verbose=verbose,
+                return_raw=return_raw,
+                chunk_duration=chunk_duration,
+                chunk_overlap=chunk_overlap,
+            )
+
+        # Single-request path (existing behavior)
+        return self._diarize_and_transcribe_single(
+            audio_path=audio_path,
+            language=language,
+            num_speakers=num_speakers,
+            min_speakers=min_speakers,
+            max_speakers=max_speakers,
+            clustering_threshold=clustering_threshold,
+            min_duration_on=min_duration_on,
+            min_duration_off=min_duration_off,
+            response_format=response_format,
+            include_text=include_text,
+            verbose=verbose,
+            return_raw=return_raw,
+        )
+
+    def _diarize_and_transcribe_single(
+        self,
+        audio_path: str,
+        *,
+        language: Optional[str] = None,
+        num_speakers: Optional[int] = None,
+        min_speakers: Optional[int] = None,
+        max_speakers: Optional[int] = None,
+        clustering_threshold: Optional[float] = None,
+        min_duration_on: Optional[float] = None,
+        min_duration_off: Optional[float] = None,
+        response_format: Optional[str] = None,
+        include_text: Optional[bool] = None,
+        verbose: bool = False,
+        return_raw: bool = False,
+    ) -> Dict[str, Any]:
+        """
+        Internal: single-request diarization and transcription.
         """
         if verbose:
             print("Starting diarization and transcription via LocalAI.")
@@ -214,6 +374,153 @@ class LocalAIClient:
 
         return parsed
 
+    def _diarize_and_transcribe_chunked(
+        self,
+        audio_path: str,
+        *,
+        language: Optional[str] = None,
+        num_speakers: Optional[int] = None,
+        min_speakers: Optional[int] = None,
+        max_speakers: Optional[int] = None,
+        clustering_threshold: Optional[float] = None,
+        min_duration_on: Optional[float] = None,
+        min_duration_off: Optional[float] = None,
+        response_format: Optional[str] = None,
+        include_text: Optional[bool] = None,
+        verbose: bool = False,
+        return_raw: bool = False,
+        chunk_duration: float = DEFAULT_CHUNK_DURATION,
+        chunk_overlap: float = DEFAULT_CHUNK_OVERLAP,
+    ) -> Dict[str, Any]:
+        """
+        Internal: chunked diarization and transcription for long audio.
+
+        - Splits audio into overlapping chunks.
+        - Transcribes each chunk via /v1/audio/diarization.
+        - Merges segments with adjusted timestamps.
+        """
+        if verbose:
+            print("Audio is long; splitting into chunks to avoid GPU memory issues.")
+
+        logger.info(
+            "Chunked transcription: chunk_duration=%s, overlap=%s",
+            chunk_duration,
+            chunk_overlap,
+        )
+
+        chunks = split_audio_into_chunks(
+            input_path=audio_path,
+            max_duration=chunk_duration,
+            overlap=chunk_overlap,
+        )
+
+        if len(chunks) == 1:
+            # No actual split needed; fall back to single-request path
+            return self._diarize_and_transcribe_single(
+                audio_path=chunks[0]["path"],
+                language=language,
+                num_speakers=num_speakers,
+                min_speakers=min_speakers,
+                max_speakers=max_speakers,
+                clustering_threshold=clustering_threshold,
+                min_duration_on=min_duration_on,
+                min_duration_off=min_duration_off,
+                response_format=response_format,
+                include_text=include_text,
+                verbose=verbose,
+                return_raw=return_raw,
+            )
+
+        all_segments: List[List[float]] = []
+        all_speakers: List[str] = []
+        all_transcripts: List[str] = []
+        raw_results: List[Dict[str, Any]] = []
+        temp_files = [c["path"] for c in chunks]
+
+        try:
+            for i, chunk_info in enumerate(chunks):
+                chunk_path = chunk_info["path"]
+                chunk_start = chunk_info["start"]
+
+                if verbose:
+                    print(
+                        f"Transcribing chunk {i+1}/{len(chunks)} "
+                        f"(start={chunk_start:.1f}s)"
+                    )
+
+                logger.info(
+                    "Transcribing chunk %d/%d, start=%.1f", i + 1, len(chunks), chunk_start
+                )
+
+                # Use single-request logic for each chunk
+                chunk_result = self._diarize_and_transcribe_single(
+                    audio_path=chunk_path,
+                    language=language,
+                    num_speakers=num_speakers,
+                    min_speakers=min_speakers,
+                    max_speakers=max_speakers,
+                    clustering_threshold=clustering_threshold,
+                    min_duration_on=min_duration_on,
+                    min_duration_off=min_duration_off,
+                    response_format=response_format,
+                    include_text=include_text,
+                    verbose=False,
+                    return_raw=return_raw,
+                )
+
+                segs = chunk_result.get("segments", [])
+                spks = chunk_result.get("speakers", [])
+                txts = chunk_result.get("transcripts", [])
+                raw = chunk_result.get("raw_result")
+
+                # Adjust timestamps to global timeline
+                adjusted_segs = []
+                for seg, sp, txt in zip(segs, spks, txts):
+                    start = float(seg[0]) + chunk_start
+                    end = float(seg[1]) + chunk_start
+                    adjusted_segs.append([start, end])
+                    all_speakers.append(sp)
+                    all_transcripts.append(txt)
+                all_segments.extend(adjusted_segs)
+
+                if return_raw and raw is not None:
+                    raw_results.append(raw)
+
+        finally:
+            # Clean up temporary chunk files
+            for path in temp_files:
+                if path and os.path.exists(path) and path != audio_path:
+                    try:
+                        os.remove(path)
+                    except Exception as e:
+                        logger.warning("Failed to remove chunk file %s: %s", path, e)
+
+        # Sort segments by start time
+        combined = list(zip(all_segments, all_speakers, all_transcripts))
+        combined.sort(key=lambda x: x[0][0])
+        all_segments = [x[0] for x in combined]
+        all_speakers = [x[1] for x in combined]
+        all_transcripts = [x[2] for x in combined]
+
+        if verbose:
+            print(
+                f"Chunked transcription complete. Total segments: {len(all_segments)}"
+            )
+
+        result = {
+            "segments": all_segments,
+            "speakers": all_speakers,
+            "transcripts": all_transcripts,
+        }
+
+        if return_raw and raw_results:
+            result["raw_result"] = {
+                "chunked": True,
+                "chunks": raw_results,
+            }
+
+        return result
+
     def _parse_diarization_response(self, result: Dict[str, Any]) -> Dict[str, Any]:
         """
         Convert LocalAI verbose_json response into the internal format used by Scraibe:

scraibe/summarizer.py

@@ -43,7 +43,7 @@ class SummarizerClient:
         api_url: Optional[str] = None,
         api_key: Optional[str] = None,
         model: Optional[str] = None,
-        timeout: float = 600.0,
+        timeout: float = 3600.0,
     ):
         self.api_url = (api_url or os.getenv("SUMMARIZER_API_URL")).strip().rstrip("/")
         self.api_key = api_key or os.getenv("SUMMARIZER_API_KEY") or None

scraibe/tasks.py

@@ -70,20 +70,37 @@ def _get_subject(env_var: str, default: str) -> str:
 def get_queue_position(task_id: str) -> int:
     """
     Estimate the job's position in the queue.
+    Returns:
+        - A positive int if we can estimate (1 = first in line).
+        - 0 if we cannot reliably determine position.
     """
     try:
         inspect = celery_app.control.inspect()
-        ready = inspect.active() or {}
-        reserved = inspect.reserved() or {}
-        count = 0
-        for _, tasks in list(ready.values()) + list(reserved.values()):
+        reserved = inspect.reserved() or {}  # queued but not yet running
+        active = inspect.active() or {}      # currently running
+
+        # Count tasks ahead of this one in the reserved (waiting) queue
+        ahead = 0
+        found = False
+        for _, tasks in list(reserved.values()):
             for t in tasks:
-                if t.get("id") == task_id:
+                tid = t.get("id")
+                if tid == task_id:
+                    found = True
                     break
-                count += 1
-        return max(count + 1, 1)
+                ahead += 1
+            if found:
+                break
+
+        # If not found in reserved, it may already be active or not yet visible.
+        # In that case, treat it as position 1.
+        if found:
+            return max(ahead + 1, 1)
+        else:
+            return 1
     except Exception:
-        return -1
+        # If inspection fails, don't guess; caller should use a safe message.
+        return 0
 
 
 def send_initial_email(to: str, queue_pos: int):
@@ -103,8 +120,12 @@ def send_initial_email(to: str, queue_pos: int):
 
     if queue_pos > 0:
         body += f"Your request is currently number {queue_pos} in the queue.\n"
+        queue_position_display = (
+            f'<span style="color:{_accent_color()}; font-weight:bold;">{queue_pos}</span>'
+        )
     else:
         body += "Your request has been queued for processing.\n"
+        queue_position_display = "the queue"
 
     body += (
         "\n"
@@ -119,7 +140,7 @@ def send_initial_email(to: str, queue_pos: int):
     try:
         html = load_template(
             "upload_notification_template.html",
-            queue_position=str(max(queue_pos, 1)),
+            queue_position_text=queue_position_display,
         )
     except EmailError as e:
         logger.warning("Failed to render upload notification template: %s", e)
@@ -141,6 +162,7 @@ def send_success_email(
     """
     Send final email with transcript and attachments.
     Subject is customizable via EMAIL_SUBJECT_SUCCESS.
+    Falls back to a safe default if the env var is missing or blank.
     """
     subject = _get_subject(
         "EMAIL_SUBJECT_SUCCESS",
@@ -183,7 +205,7 @@ def send_success_email(
             html=html,
             attachments=attachments,
         )
-        logger.info("Success email sent to %s for job %s", to, task_id)
+        logger.info("Success email sent to %s for job %s with subject: %s", to, task_id, subject)
     except EmailError as e:
         logger.error("Failed to send success email to %s for job %s: %s", to, task_id, e)
 
@@ -229,6 +251,8 @@ def send_error_email(to: str, error_message: str, task_id: str):
     name="scraibe.tasks.process_transcription_task",
     bind=True,
     max_retries=1,
+    task_time_limit=14400,  # 4 hours
+    task_soft_time_limit=13500,  # warn at 3h45m
 )
 def process_transcription_task(
     self,
@@ -306,14 +330,18 @@ def process_transcription_task(
 
                 prompt = (
                     "Below is a transcript with speaker labels like 'SPEAKER 1', 'SPEAKER 2', etc. "
-                    "Based on how they speak and the context, suggest realistic names for each speaker. "
-                    "Do not add extra commentary. Output ONLY a mapping in this exact format, one per line:
-SPEAKER 1: Suggested Name
-SPEAKER 2: Suggested Name
-SPEAKER 3: Suggested Name
-
-Transcript:
-" + transcript_text
+                    "Based on the context and how each speaker talks, identify each speaker as:\n"
+                    "- Their real name, if it is clearly mentioned or strongly implied, OR\n"
+                    "- A concise role/position (e.g., Judge, Doctor, Manager, Interviewer, Client, Witness), "
+                    "if their identity is not clear.\n"
+                    "Do not invent random personal names. "
+                    "Do not add extra commentary. Output ONLY a mapping in this exact format, one per line:\n"
+                    "SPEAKER 1: Name or Role\n"
+                    "SPEAKER 2: Name or Role\n"
+                    "SPEAKER 3: Name or Role\n"
+                    "\n"
+                    "Transcript:\n"
+                    + transcript_text
                 )
 
                 response = summarizer._chat_completion(
@@ -331,7 +359,7 @@ Transcript:
                     re.IGNORECASE,
                 ):
                     spk = f"SPEAKER {m.group(1).strip()}"
-                    name = m.group(2).strip().rstrip(".")
+                    name = m.group(2).strip().rstrip(".").upper()
                     if name:
                         speaker_map[spk] = name
 
@@ -389,9 +417,12 @@ Transcript:
             f.write(transcript_text)
         temp_files.append(md_transcript_path)
 
-        # Transcript .docx
+        # Transcript .docx (standalone, no cover page)
         docx_transcript_path = _safe_filename("TRANSCRIPT", local, date_tag, ".docx")
-        create_transcript_docx(transcript_text, docx_transcript_path)
+        create_transcript_docx(
+            transcript_text,
+            docx_transcript_path,
+        )
         temp_files.append(docx_transcript_path)
 
         # JSON as SOURCE
@@ -415,26 +446,39 @@ Transcript:
         temp_files.append(json_path)
 
         # Summary files (if present)
+        md_summary_path = None
+        docx_summary_path = None
+
         if summary_text:
+            # Summary .md
             md_summary_path = _safe_filename("SUMMARY", local, date_tag, ".md")
             with open(md_summary_path, "w", encoding="utf-8") as f:
                 f.write("# Summary\n\n")
                 f.write(summary_text)
             temp_files.append(md_summary_path)
 
+            # Summary .docx (standalone, no cover page)
             docx_summary_path = _safe_filename("SUMMARY", local, date_tag, ".docx")
-            create_summary_docx(summary_text, docx_summary_path)
+            create_summary_docx(
+                summary_text,
+                docx_summary_path,
+            )
             temp_files.append(docx_summary_path)
 
+        # 5) Build attachments list
+
+        # Always: JSON, transcript MD, transcript DOCX
         attachments = [
             md_transcript_path,
             docx_transcript_path,
             json_path,
         ]
+
+        # If summary is present, add summary MD and DOCX
         if summary_text:
             attachments += [md_summary_path, docx_summary_path]
 
-        # 5) Send success email
+        # 6) Send success email
         send_success_email(
             to=email_to,
             transcript_text=transcript_text,
@@ -454,7 +498,7 @@ Transcript:
         )
         raise e
     finally:
-        # 6) Cleanup
+        # 7) Cleanup
         for path in temp_files:
             _remove_file(path)
         if audio_path:

scraibe/webui.py

@@ -144,7 +144,7 @@ def create_app():
                 task_choice = gr.Radio(
                     choices=[
                         ("Transcribe", "transcribe"),
-                        ("Transcript & Summarize", "transcript_and_summarize"),
+                        ("Transcribe & summarize", "transcript_and_summarize"),
                     ],
                     value="transcribe",
                     label="Task",
@@ -153,7 +153,7 @@ def create_app():
 
                 identify_speakers = gr.Checkbox(
                     label="Identify speakers (best effort using AI)",
-                    value=False,
+                    value=True,
                     info="If enabled, AI will attempt to infer real names for speakers and replace Speaker 1/2/etc. in the transcript.",
                 )
 
@@ -307,6 +307,20 @@ def create_app():
         body {{
             font-family: Arial, sans-serif;
         }}
+        /* Increase main title font size */
+        h1,
+        .webui-title,
+        .header-title {{
+            font-size: 60px !important;
+        }}
+        /* Hide Gradio's "Use via API" link/button */
+        #share-btn,
+        a[href*="/api"],
+        a[href*="#/api"],
+        a[href*="#api"],
+        .gradio-container a[href*="api"] {{
+            display: none !important;
+        }}
         /* Mobile-friendly adjustments */
         @media (max-width: 700px) {{
             .gradio-container {{

tests/test_audio_chunking.py

@@ -0,0 +1,86 @@
+import os
+import subprocess
+import tempfile
+import pytest
+
+from scraibe.audio import (
+    get_audio_duration,
+    split_audio_into_chunks,
+)
+
+TEST_AUDIO_1 = "tests/audio_test_1.mp4"
+TEST_AUDIO_2 = "tests/audio_test_2.mp4"
+
+
+@pytest.fixture(params=[TEST_AUDIO_1, TEST_AUDIO_2])
+def test_audio_path(request):
+    return request.param
+
+
+def test_get_audio_duration(test_audio_path):
+    dur = get_audio_duration(test_audio_path)
+    assert isinstance(dur, float)
+    assert dur > 0
+
+
+def test_split_audio_into_chunks_no_split_short(test_audio_path):
+    # For short files, should return the same file with no extra chunks
+    chunks = split_audio_into_chunks(
+        input_path=test_audio_path,
+        max_duration=600.0,  # larger than both test files
+        overlap=2.0,
+    )
+    assert len(chunks) == 1
+    assert chunks[0]["path"] == test_audio_path
+    assert chunks[0]["start"] == 0.0
+    dur = get_audio_duration(test_audio_path)
+    assert abs(chunks[0]["end"] - dur) < 0.05
+
+
+def test_split_audio_into_chunks_creates_chunks(tmp_path):
+    # Use a small chunk duration to force splitting
+    chunks = split_audio_into_chunks(
+        input_path=TEST_AUDIO_1,
+        max_duration=2.0,
+        overlap=0.5,
+    )
+    assert len(chunks) > 1
+
+    # Check that each chunk file exists and is non-empty
+    for c in chunks:
+        assert os.path.exists(c["path"])
+        assert os.path.getsize(c["path"]) > 0
+
+    # Check time ordering and overlap
+    for i in range(1, len(chunks)):
+        prev = chunks[i - 1]
+        curr = chunks[i]
+        assert curr["start"] >= prev["start"]
+        assert curr["start"] < prev["end"]  # overlap
+
+    # Cleanup
+    for c in chunks:
+        if os.path.exists(c["path"]):
+            os.remove(c["path"])
+
+
+def test_split_audio_into_chunks_total_coverage(test_audio_path):
+    dur = get_audio_duration(test_audio_path)
+
+    # Use small chunks to ensure coverage
+    chunks = split_audio_into_chunks(
+        input_path=test_audio_path,
+        max_duration=2.0,
+        overlap=0.5,
+    )
+
+    # First chunk starts at 0
+    assert chunks[0]["start"] == 0.0
+
+    # Last chunk end should cover the duration
+    assert chunks[-1]["end"] >= dur - 0.05
+
+    # Cleanup
+    for c in chunks:
+        if os.path.exists(c["path"]):
+            os.remove(c["path"])

tests/test_docx_generation.py

@@ -0,0 +1,96 @@
+"""
+Local test for transcript/summary/combined .docx generation.
+Checks:
+- Line numbering only on transcript pages.
+- Page numbering (X of Y) in footer.
+- Cover pages present and centered.
+- Combined document structure.
+"""
+
+import sys
+import os
+import tempfile
+
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+from scraibe.email_sender import (
+    create_transcript_docx,
+    create_summary_docx,
+    create_combined_docx,
+)
+
+TRANSCRIPT_TEXT = """[00:00] Speaker 1: Good morning, everyone. Thank you for joining today's meeting.
+[00:12] Speaker 2: Good morning. I'm looking forward to discussing the new requirements.
+[00:25] Speaker 1: Let's start with the timeline. We need to finalize the scope by Friday.
+[00:38] Speaker 2: Agreed. I'll send a summary of the key points after this call.
+[00:45] Speaker 1: Perfect. If there are no other items, we can wrap up here."""
+
+SUMMARY_TEXT = """# Meeting Overview
+## Key Discussion Points
+### Timeline and Scope
+#### Next Steps"""
+
+COVER_DATE = "June 14, 2026"
+TRANSCRIPT_DESC = "Transcript of a project planning meeting discussing timelines and scope."
+SUMMARY_DESC = "Summary of a project planning meeting covering key decisions and next steps."
+
+
+def main():
+    with tempfile.TemporaryDirectory() as tmpdir:
+        print("Using temp directory:", tmpdir)
+
+        # 1) Transcript-only
+        transcript_path = os.path.join(tmpdir, "TRANSCRIPT_TEST.docx")
+        print("Creating transcript-only docx:", transcript_path)
+        create_transcript_docx(
+            text=TRANSCRIPT_TEXT,
+            filename=transcript_path,
+            include_cover=True,
+            cover_date=COVER_DATE,
+            cover_desc=TRANSCRIPT_DESC,
+        )
+        print("OK: transcript-only created.")
+
+        # 2) Summary-only
+        summary_path = os.path.join(tmpdir, "SUMMARY_TEST.docx")
+        print("Creating summary-only docx:", summary_path)
+        create_summary_docx(
+            text=SUMMARY_TEXT,
+            filename=summary_path,
+            include_cover=True,
+            cover_date=COVER_DATE,
+            cover_desc=SUMMARY_DESC,
+        )
+        print("OK: summary-only created.")
+
+        # 3) Combined
+        combined_path = os.path.join(tmpdir, "COMBINED_TEST.docx")
+        print("Creating combined docx:", combined_path)
+        create_combined_docx(
+            transcript_text=TRANSCRIPT_TEXT,
+            summary_text=SUMMARY_TEXT,
+            filename=combined_path,
+            transcript_cover_date=COVER_DATE,
+            transcript_cover_desc=TRANSCRIPT_DESC,
+            summary_cover_date=COVER_DATE,
+            summary_cover_desc=SUMMARY_DESC,
+        )
+        print("OK: combined created.")
+
+        # Basic size sanity checks
+        for path in [transcript_path, summary_path, combined_path]:
+            size = os.path.getsize(path)
+            print(f"File: {os.path.basename(path)} - size: {size} bytes")
+            if size < 10000:
+                print("WARNING: File seems unusually small:", path)
+
+        print("\nAll .docx files generated successfully.")
+        print("Please open them in Word to verify:")
+        print("- Only transcript pages have line numbers.")
+        print("- Footer shows 'X of Y' on all pages.")
+        print("- Cover pages are centered and use the correct date format.")
+        print("- Combined doc order: cover, page break, summary, page break, transcript.")
+
+
+if __name__ == "__main__":
+    main()

tests/test_localai_chunking.py

@@ -0,0 +1,230 @@
+import os
+import json
+import tempfile
+from unittest.mock import patch, MagicMock
+
+import pytest
+
+from scraibe.localai_client import LocalAIClient, LocalAIError
+from scraibe.audio import get_audio_duration, split_audio_into_chunks
+
+
+TEST_AUDIO_1 = "tests/audio_test_1.mp4"
+
+
+def make_fake_segments(start=0.0, count=3):
+    segments = []
+    for i in range(count):
+        s = start + i * 2.0
+        e = s + 2.0
+        segments.append({
+            "start": s,
+            "end": e,
+            "speaker": "SPEAKER_00",
+            "text": f"Segment text {i}",
+        })
+    return segments
+
+
+def fake_localai_response(segments):
+    return {
+        "segments": segments,
+        "text": " ".join(seg["text"] for seg in segments),
+    }
+
+
+@pytest.fixture
+def client():
+    with patch.object(LocalAIClient, "__init__", lambda self, **kw: None):
+        c = LocalAIClient()
+        c.api_url = "http://localhost:8080"
+        c.model = "vibevoice-diarize"
+        c.api_key = None
+        c._client = MagicMock()
+        return c
+
+
+def test_parse_diarization_response(client):
+    segs = make_fake_segments()
+    raw = fake_localai_response(segs)
+
+    out = client._parse_diarization_response(raw)
+
+    assert "segments" in out
+    assert "speakers" in out
+    assert "transcripts" in out
+    assert len(out["segments"]) == len(segs)
+    for i, s in enumerate(segs):
+        assert out["segments"][i][0] == s["start"]
+        assert out["segments"][i][1] == s["end"]
+        assert out["speakers"][i] == s["speaker"]
+        assert out["transcripts"][i] == s["text"]
+
+
+def test_parse_diarization_empty(client):
+    out = client._parse_diarization_response({"segments": []})
+    assert out["segments"] == []
+    assert out["speakers"] == []
+    assert out["transcripts"] == []
+
+
+def test_diarize_and_transcribe_single_happy(client):
+    with patch.object(client, "_client") as mock_client:
+        mock_resp = MagicMock()
+        mock_resp.status_code = 200
+        mock_resp.json.return_value = fake_localai_response(make_fake_segments())
+        mock_client.post.return_value = mock_resp
+
+        result = client.diarize_and_transcribe(
+            audio_path=TEST_AUDIO_1,
+            verbose=False,
+            return_raw=True,
+        )
+
+        assert "segments" in result
+        assert "raw_result" in result
+        assert len(result["segments"]) > 0
+
+
+def test_chunking_triggered_for_long_audio(client):
+    # Simulate long audio by patching get_audio_duration
+    with patch("scraibe.localai_client.get_audio_duration") as mock_dur, \
+         patch.object(client, "_diarize_and_transcribe_chunked") as mock_chunked:
+
+        mock_dur.return_value = 600.0  # 10 minutes
+        mock_chunked.return_value = {
+            "segments": [],
+            "speakers": [],
+            "transcripts": [],
+        }
+
+        client.diarize_and_transcribe(
+            audio_path=TEST_AUDIO_1,
+            verbose=False,
+            use_chunking=None,
+            max_single_request_duration=300.0,
+        )
+
+        mock_chunked.assert_called_once()
+
+
+def test_chunking_not_triggered_for_short_audio(client):
+    with patch("scraibe.localai_client.get_audio_duration") as mock_dur, \
+         patch.object(client, "_diarize_and_transcribe_chunked") as mock_chunked, \
+         patch.object(client, "_diarize_and_transcribe_single") as mock_single:
+
+        mock_dur.return_value = 120.0
+        mock_single.return_value = {
+            "segments": [],
+            "speakers": [],
+            "transcripts": [],
+        }
+
+        client.diarize_and_transcribe(
+            audio_path=TEST_AUDIO_1,
+            verbose=False,
+            use_chunking=None,
+            max_single_request_duration=300.0,
+        )
+
+        mock_chunked.assert_not_called()
+        mock_single.assert_called_once()
+
+
+def test_chunked_transcription_adjusts_timestamps(client):
+    # Mock split_audio_into_chunks to return two chunks
+    chunk1_path = TEST_AUDIO_1
+    chunk2_path = TEST_AUDIO_1  # reusing same file; in real usage different
+
+    chunks = [
+        {"path": chunk1_path, "start": 0.0, "end": 10.0},
+        {"path": chunk2_path, "start": 10.0, "end": 20.0},
+    ]
+
+    with patch("scraibe.localai_client.split_audio_into_chunks") as mock_split, \
+         patch.object(client, "_diarize_and_transcribe_single") as mock_single, \
+         patch("os.remove"):
+
+        mock_split.return_value = chunks
+
+        # First chunk: segments 0–4
+        # Second chunk: segments 0–4 (local times)
+        def side_effect(audio_path, **kw):
+            if audio_path == chunk1_path:
+                segs = make_fake_segments(start=0.0, count=2)
+            else:
+                segs = make_fake_segments(start=0.0, count=2)
+            return client._parse_diarization_response(fake_localai_response(segs))
+
+        mock_single.side_effect = side_effect
+
+        result = client._diarize_and_transcribe_chunked(
+            audio_path=TEST_AUDIO_1,
+            verbose=False,
+            return_raw=False,
+            chunk_duration=10.0,
+            chunk_overlap=2.0,
+        )
+
+        # Check we got 4 segments total
+        assert len(result["segments"]) == 4
+
+        # First two segments should be in [0, 4]
+        assert result["segments"][0][0] == 0.0
+        assert result["segments"][1][0] == 2.0
+
+        # Next two segments should be shifted by 10
+        assert result["segments"][2][0] == 10.0
+        assert result["segments"][3][0] == 12.0
+
+
+@pytest.mark.integration
+def test_integration_chunked_transcription_with_localai():
+    """
+    Integration test: run chunked transcription against a live LocalAI instance.
+    Only runs if LOCALAI_API_URL is set and an audio file is provided.
+    This test is skipped by default unless run with:
+        pytest -m integration
+    """
+    api_url = os.getenv("LOCALAI_API_URL")
+    if not api_url:
+        pytest.skip("LOCALAI_API_URL not set; skipping integration test")
+
+    # Use one of the bundled test audio files
+    audio_path = TEST_AUDIO_1
+    if not os.path.exists(audio_path):
+        pytest.skip(f"Test audio not found: {audio_path}")
+
+    # Force chunking with a very small max_single_request_duration
+    # Use environment-configured model or a sensible default
+    model = os.getenv("LOCALAI_MODEL") or "vibevoice-cpp-asr"
+
+    client = LocalAIClient(api_url=api_url, model=model)
+    try:
+        result = client.diarize_and_transcribe(
+            audio_path=audio_path,
+            verbose=True,
+            return_raw=True,
+            use_chunking=True,
+            chunk_duration=3.0,
+            chunk_overlap=0.5,
+            max_single_request_duration=1.0,
+        )
+
+        assert "segments" in result
+        assert len(result["segments"]) > 0
+
+        # Basic sanity: segments are time-ordered
+        for i in range(1, len(result["segments"])):
+            prev_end = result["segments"][i - 1][1]
+            curr_start = result["segments"][i][0]
+            assert curr_start >= result["segments"][i - 1][0]
+
+        # If raw_result indicates chunked, ensure structure is sensible
+        raw = result.get("raw_result")
+        if raw and raw.get("chunked"):
+            assert "chunks" in raw
+            assert len(raw["chunks"]) > 1
+
+    finally:
+        client.close()