Video/Audio Processor

Process video and audio recordings into text transcripts and visual frame captures for analysis. Auto-detects your hardware and picks the optimal transcription model, device, and settings.

Phase 0: Detect Compute Environment

Run this at the start of every invocation. It determines everything downstream.

python3 -c "
import platform, shutil, subprocess, json

info = {'os': platform.system(), 'arch': platform.machine(), 'ram_gb': 0, 'gpu': 'none', 'vram_gb': 0, 'device': 'cpu', 'dtype': 'float32'}

# RAM
try:
    if platform.system() == 'Darwin':
        import os; info['ram_gb'] = round(os.sysconf('SC_PAGE_SIZE') * os.sysconf('SC_PHYS_PAGES') / (1024**3))
    elif platform.system() == 'Linux':
        with open('/proc/meminfo') as f:
            for line in f:
                if line.startswith('MemTotal'):
                    info['ram_gb'] = round(int(line.split()[1]) / (1024**2))
                    break
    else:
        import ctypes
        mem = ctypes.c_ulonglong(0)
        ctypes.windll.kernel32.GetPhysicallyInstalledMemory(ctypes.byref(mem))
        info['ram_gb'] = round(mem.value / (1024**2))
except: pass

# GPU detection
try:
    import torch
    if torch.cuda.is_available():
        info['gpu'] = torch.cuda.get_device_name(0)
        info['vram_gb'] = round(torch.cuda.get_device_properties(0).total_mem / (1024**3))
        info['device'] = 'cuda'
        info['dtype'] = 'float16'
    elif hasattr(torch.backends, 'mps') and torch.backends.mps.is_available():
        info['gpu'] = 'Apple Silicon (MPS)'
        info['vram_gb'] = info['ram_gb']  # unified memory
        info['device'] = 'mps'
        info['dtype'] = 'float16'
    elif hasattr(torch, 'hip') or 'AMD' in str(getattr(torch, '_C', '')):
        info['gpu'] = 'AMD (ROCm)'
        info['device'] = 'cuda'  # ROCm uses cuda API
        info['dtype'] = 'float16'
except ImportError:
    pass

# ffmpeg detection
info['ffmpeg'] = shutil.which('ffmpeg') is not None
info['ffprobe'] = shutil.which('ffprobe') is not None

print(json.dumps(info))
"

Parse the JSON output and store it internally as COMPUTE. Present the results to the user:

Detected hardware:

• OS: {os} ({arch})
• RAM: {ram_gb} GB
• GPU: {gpu} ({vram_gb} GB VRAM)
• Compute device: {device}
• ffmpeg: {installed/missing}

Whisper Engine Selection

Based on hardware, choose the transcription engine:

Model Selection Matrix

Based on hardware and recording length, recommend a model:

Present the recommendation and let the user choose:

AskUserQuestion: "Which transcription model should I use?"
Options:
- [Recommended model] (Recommended) - [reason based on their hardware]
- large-v3 - Best accuracy, needs 8+ GB VRAM or lots of RAM
- medium - High accuracy, balanced speed
- small - Good accuracy, faster processing
- base - Adequate for most meetings, fastest

Store the chosen model as WHISPER_MODEL and engine as WHISPER_ENGINE.

Phase 1: Identify the Recording

Find the file. Zoom recordings typically follow the pattern:

• GMT{date}-{time}_Recording_{resolution}.mp4 (video)
• GMT{date}-{time}_Recording.m4a (audio only)
• GMT{date}-{time}_RecordingnewChat.txt (chat log)

Check for all three artifacts. Read the chat file first (it's small and gives immediate context).

Phase 2: Install Dependencies

Check and install what's needed based on platform.

ffmpeg

# Check if ffmpeg is available
ffmpeg -version 2>&1 | head -1
ffprobe -version 2>&1 | head -1

If ffmpeg is missing, install based on platform:

Python + Whisper

python3 -c "import torch; print(torch.__version__)" 2>&1

If torch is missing, install based on platform:

Then install the transcription engine:

# For CUDA systems (faster-whisper with CTranslate2 acceleration)
pip3 install faster-whisper

# For MPS / CPU systems (standard OpenAI Whisper)
pip3 install openai-whisper

Phase 3: Get Duration

Use ffprobe (cross-platform) to get the file duration:

ffprobe -v quiet -show_entries format=duration -of csv=p=0 "/path/to/file"

This returns duration in seconds. Convert to human-readable format and display to user. Use this to estimate transcription time based on the selected model and hardware.

Present the plan:

Processing plan:

• File: {filename} ({duration})
• Engine: {WHISPER_ENGINE} with {WHISPER_MODEL} model on {device}
• Estimated transcription time: {estimate}
• Frame extraction: 1 per minute ({frame_count} frames)

Want me to adjust anything before starting?

Time Estimates

Phase 4: Extract Visual Frames

Extract one frame per minute for visual context. Use a cross-platform temp directory:

# Create temp directory (cross-platform)
python3 -c "import tempfile, os; d = os.path.join(tempfile.gettempdir(), 'av-frames'); os.makedirs(d, exist_ok=True); print(d)"

Then extract frames:

ffmpeg -i "/path/to/video.mp4" \
  -vf "fps=1/60,scale=1280:-1" -q:v 3 {TEMP_DIR}/frame_%03d.jpg

Use the Read tool on each frame to view it. Frames show:

• Screen shares (code, slides, UIs)
• Participant names (bottom-left of each video tile)
• Timestamps (if visible in screen share)

Phase 5: Transcribe Audio

Run the transcription engine. Choose the correct code based on WHISPER_ENGINE:

faster-whisper (CUDA systems)

from faster_whisper import WhisperModel
import json

model = WhisperModel("{WHISPER_MODEL}", device="cuda", compute_type="float16")
segments, info = model.transcribe("/path/to/audio", beam_size=5, language="en")

results = []
for segment in segments:
    results.append({
        "start": segment.start,
        "end": segment.end,
        "text": segm