feat(whisper): add batch_transcribe.py and document HAL0022 transcription setup
- batch_transcribe.py: recursive multi-creator transcription runner that walks nested subdirectories, attributes creator from top-level folder name, writes batch_manifest.json with timing and per-creator results - README.md: updated with batch mode docs, HAL0022 environment details, transcript output location (C:\Users\jlightner\chrysopedia\transcripts), scheduled task usage, and transfer instructions for ub01 ingestion
This commit is contained in:
jlightner
parent
08d7d19d0e
commit
e8bc3fd9a2
2 changed files with 446 additions and 10 deletions
|
|
@ -18,12 +18,18 @@ sudo apt install ffmpeg
|
|||
|
||||
# macOS
|
||||
brew install ffmpeg
|
||||
|
||||
# Windows (via chocolatey or manual install)
|
||||
choco install ffmpeg
|
||||
```
|
||||
|
||||
### Install Python dependencies
|
||||
|
||||
```bash
|
||||
pip install -r requirements.txt
|
||||
|
||||
# For CUDA support, install torch with CUDA wheels:
|
||||
pip install torch --index-url https://download.pytorch.org/whl/cu126
|
||||
```
|
||||
|
||||
## Usage
|
||||
|
|
@ -40,7 +46,29 @@ python transcribe.py --input "path/to/video.mp4" --output-dir ./transcripts
|
|||
python transcribe.py --input ./videos/ --output-dir ./transcripts
|
||||
```
|
||||
|
||||
### Options
|
||||
### Mass batch mode (recursive, multi-creator)
|
||||
|
||||
For large content libraries with nested subdirectories per creator:
|
||||
|
||||
```bash
|
||||
python batch_transcribe.py \
|
||||
--content-root "A:\Education\Artist Streams & Content" \
|
||||
--output-dir "C:\Users\jlightner\chrysopedia\transcripts" \
|
||||
--python C:\Users\jlightner\.conda\envs\transcribe\python.exe
|
||||
|
||||
# Dry run to preview without transcribing:
|
||||
python batch_transcribe.py --content-root ... --output-dir ... --dry-run
|
||||
```
|
||||
|
||||
`batch_transcribe.py` recursively walks all subdirectories, discovers video
|
||||
files, and calls `transcribe.py` for each directory. The `creator_folder`
|
||||
field in the output JSON is set to the top-level subdirectory name (the
|
||||
artist/creator). Output directory structure mirrors the source hierarchy.
|
||||
|
||||
A `batch_manifest.json` is written to the output root on completion with
|
||||
timing, per-creator results, and error details.
|
||||
|
||||
### Options (transcribe.py)
|
||||
|
||||
| Flag | Default | Description |
|
||||
| --------------- | ----------- | ----------------------------------------------- |
|
||||
|
|
@ -51,9 +79,21 @@ python transcribe.py --input ./videos/ --output-dir ./transcripts
|
|||
| `--creator` | (inferred) | Override creator folder name in output JSON |
|
||||
| `-v, --verbose` | off | Enable debug logging |
|
||||
|
||||
### Options (batch_transcribe.py)
|
||||
|
||||
| Flag | Default | Description |
|
||||
| ----------------- | ------------ | ------------------------------------------------ |
|
||||
| `--content-root` | (required) | Root directory with creator subdirectories |
|
||||
| `--output-dir` | (required) | Root output directory for transcript JSONs |
|
||||
| `--script` | (auto) | Path to transcribe.py (default: same directory) |
|
||||
| `--python` | (auto) | Python interpreter to use |
|
||||
| `--model` | `large-v3` | Whisper model name |
|
||||
| `--device` | `cuda` | Compute device |
|
||||
| `--dry-run` | off | Preview work plan without transcribing |
|
||||
|
||||
## Output Format
|
||||
|
||||
Each video produces a JSON file matching the Chrysopedia spec:
|
||||
Each video produces a JSON file matching the Chrysopedia pipeline spec:
|
||||
|
||||
```json
|
||||
{
|
||||
|
|
@ -74,28 +114,105 @@ Each video produces a JSON file matching the Chrysopedia spec:
|
|||
}
|
||||
```
|
||||
|
||||
This format is consumed directly by the Chrysopedia pipeline stage 2
|
||||
(transcript segmentation) via the `POST /api/v1/ingest` endpoint.
|
||||
|
||||
## Resumability
|
||||
|
||||
The script automatically skips videos whose output JSON already exists. To
|
||||
Both scripts automatically skip videos whose output JSON already exists. To
|
||||
re-transcribe a file, delete its output JSON first.
|
||||
|
||||
## Current Transcription Environment
|
||||
|
||||
### Machine: HAL0022 (10.0.0.131)
|
||||
|
||||
- **GPU:** NVIDIA GeForce RTX 4090 (24GB VRAM)
|
||||
- **OS:** Windows 11
|
||||
- **Python:** Conda env `transcribe` at `C:\Users\jlightner\.conda\envs\transcribe\python.exe`
|
||||
- **CUDA:** PyTorch with cu126 wheels
|
||||
|
||||
### Content Source
|
||||
|
||||
```
|
||||
A:\Education\Artist Streams & Content\
|
||||
├── au5/ (334 videos)
|
||||
├── Keota/ (193 videos)
|
||||
├── DJ Shortee/ (83 videos)
|
||||
├── KOAN Sound/ (68 videos)
|
||||
├── Teddy Killerz/ (62 videos)
|
||||
├── ... (42 creators, 1197 videos total across 146 directories)
|
||||
```
|
||||
|
||||
### Transcript Output Location
|
||||
|
||||
```
|
||||
C:\Users\jlightner\chrysopedia\transcripts\
|
||||
```
|
||||
|
||||
Directory structure mirrors the source hierarchy. Each video produces a
|
||||
`<filename>.json` transcript file.
|
||||
|
||||
**Transfer to ub01:** Transcripts need to be copied to
|
||||
`/vmPool/r/services/chrysopedia_data/transcripts/` on ub01 for pipeline
|
||||
ingestion. This can be done via SMB (`\\ub01\vmPool\services\chrysopedia_data\transcripts`)
|
||||
or via `scp`/`rsync` from a Linux machine with access to both.
|
||||
|
||||
### Running the Batch Job
|
||||
|
||||
The batch transcription runs as a Windows Scheduled Task to survive SSH
|
||||
disconnections:
|
||||
|
||||
```powershell
|
||||
# The task is already created. To re-run:
|
||||
schtasks /run /tn "ChrysopediaTranscribe"
|
||||
|
||||
# Check status:
|
||||
schtasks /query /tn "ChrysopediaTranscribe" /v /fo list | findstr /i "status result"
|
||||
|
||||
# Monitor log:
|
||||
Get-Content 'C:\Users\jlightner\chrysopedia\transcription.log' -Tail 30
|
||||
|
||||
# Or follow live:
|
||||
Get-Content 'C:\Users\jlightner\chrysopedia\transcription.log' -Tail 20 -Wait
|
||||
```
|
||||
|
||||
### Scripts on HAL0022
|
||||
|
||||
```
|
||||
C:\Users\jlightner\chrysopedia\
|
||||
├── transcribe.py # Single-file/directory transcription
|
||||
├── batch_transcribe.py # Recursive multi-creator batch runner
|
||||
├── run_transcription.bat # Batch file invoked by scheduled task
|
||||
├── launch_transcription.py # Alternative launcher (subprocess)
|
||||
├── transcription.log # Current batch run log output
|
||||
└── transcripts/ # Output directory
|
||||
├── batch_manifest.json
|
||||
├── au5/
|
||||
├── Break/
|
||||
└── ...
|
||||
```
|
||||
|
||||
## Performance
|
||||
|
||||
Whisper large-v3 on an RTX 4090 processes audio at roughly 10–20× real-time.
|
||||
A 2-hour video takes ~6–12 minutes. For 300 videos averaging 1.5 hours each,
|
||||
the initial transcription pass takes roughly 15–40 hours of GPU time.
|
||||
A 2-hour video takes ~6–12 minutes. For the full 1,197-video library, expect
|
||||
roughly 20–60 hours of GPU time depending on average video length.
|
||||
|
||||
## Directory Convention
|
||||
|
||||
The script infers the `creator_folder` field from the parent directory of each
|
||||
video file. Organize videos like:
|
||||
video file (or the top-level creator folder in batch mode). Organize videos like:
|
||||
|
||||
```
|
||||
videos/
|
||||
content-root/
|
||||
├── Skope/
|
||||
│ ├── Sound Design Masterclass pt1.mp4
|
||||
│ └── Sound Design Masterclass pt2.mp4
|
||||
│ ├── Youtube/
|
||||
│ │ ├── Sound Design Masterclass pt1.mp4
|
||||
│ │ └── Sound Design Masterclass pt2.mp4
|
||||
│ └── Patreon/
|
||||
│ └── Advanced Wavetables.mp4
|
||||
├── Mr Bill/
|
||||
│ └── Youtube/
|
||||
│ └── Glitch Techniques.mp4
|
||||
```
|
||||
|
||||
|
|
|
|||
319
whisper/batch_transcribe.py
Normal file
319
whisper/batch_transcribe.py
Normal file
|
|
@ -0,0 +1,319 @@
|
|||
#!/usr/bin/env python3
|
||||
"""
|
||||
Chrysopedia — Batch Transcription Runner
|
||||
|
||||
Recursively iterates all creator subdirectories under a content root and runs
|
||||
transcribe.py against each leaf directory containing videos. Outputs are
|
||||
written to a parallel directory structure under --output-dir that mirrors
|
||||
the source hierarchy:
|
||||
|
||||
output-dir/
|
||||
├── au5/
|
||||
│ ├── Patreon/
|
||||
│ │ ├── video1.mp4.json
|
||||
│ │ └── video2.mp4.json
|
||||
│ └── Youtube/
|
||||
│ └── ...
|
||||
├── Skope/
|
||||
│ └── ...
|
||||
|
||||
The --creator flag passed to transcribe.py is always the top-level folder
|
||||
name (the artist), regardless of subdirectory nesting depth.
|
||||
|
||||
Resumable: transcribe.py already skips videos whose output JSON exists.
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import argparse
|
||||
import json
|
||||
import logging
|
||||
import os
|
||||
import subprocess
|
||||
import sys
|
||||
import time
|
||||
from datetime import datetime, timezone
|
||||
from pathlib import Path
|
||||
|
||||
LOG_FORMAT = "%(asctime)s [%(levelname)s] %(message)s"
|
||||
logging.basicConfig(format=LOG_FORMAT, level=logging.INFO)
|
||||
logger = logging.getLogger("chrysopedia.batch_transcribe")
|
||||
|
||||
SUPPORTED_EXTENSIONS = {".mp4", ".mkv", ".avi", ".mov", ".webm", ".flv", ".wmv"}
|
||||
|
||||
|
||||
def find_video_dirs(content_root: Path) -> list[dict]:
|
||||
"""
|
||||
Recursively find all directories containing video files.
|
||||
Returns list of dicts with folder path, creator name, and video count.
|
||||
"""
|
||||
results = []
|
||||
for dirpath, dirnames, filenames in os.walk(content_root):
|
||||
dirpath = Path(dirpath)
|
||||
videos = [
|
||||
f for f in filenames
|
||||
if Path(f).suffix.lower() in SUPPORTED_EXTENSIONS
|
||||
]
|
||||
if not videos:
|
||||
continue
|
||||
|
||||
# Creator is always the top-level subdirectory under content_root
|
||||
rel = dirpath.relative_to(content_root)
|
||||
creator = rel.parts[0]
|
||||
|
||||
results.append({
|
||||
"folder": dirpath,
|
||||
"creator": creator,
|
||||
"rel_path": rel,
|
||||
"video_count": len(videos),
|
||||
})
|
||||
|
||||
return sorted(results, key=lambda x: (x["creator"], str(x["rel_path"])))
|
||||
|
||||
|
||||
def count_existing_transcripts(output_dir: Path) -> int:
|
||||
"""Count existing transcript JSONs in an output folder."""
|
||||
if not output_dir.exists():
|
||||
return 0
|
||||
return sum(1 for p in output_dir.iterdir() if p.suffix == ".json")
|
||||
|
||||
|
||||
def main() -> int:
|
||||
parser = argparse.ArgumentParser(
|
||||
description="Batch transcription runner for Chrysopedia"
|
||||
)
|
||||
parser.add_argument(
|
||||
"--content-root",
|
||||
required=True,
|
||||
help="Root directory containing creator subdirectories with videos",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--output-dir",
|
||||
required=True,
|
||||
help="Root output directory for transcript JSONs",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--script",
|
||||
default=None,
|
||||
help="Path to transcribe.py (default: same directory as this script)",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--python",
|
||||
default=sys.executable,
|
||||
help="Python interpreter to use (default: current interpreter)",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--model",
|
||||
default="large-v3",
|
||||
help="Whisper model name (default: large-v3)",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--device",
|
||||
default="cuda",
|
||||
help="Compute device (default: cuda)",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--dry-run",
|
||||
action="store_true",
|
||||
help="List what would be transcribed without running",
|
||||
)
|
||||
args = parser.parse_args()
|
||||
|
||||
content_root = Path(args.content_root)
|
||||
output_root = Path(args.output_dir)
|
||||
script_path = Path(args.script) if args.script else Path(__file__).parent / "transcribe.py"
|
||||
|
||||
if not content_root.is_dir():
|
||||
logger.error("Content root does not exist: %s", content_root)
|
||||
return 1
|
||||
|
||||
if not script_path.is_file():
|
||||
logger.error("transcribe.py not found at: %s", script_path)
|
||||
return 1
|
||||
|
||||
# Discover all directories with videos (recursive)
|
||||
video_dirs = find_video_dirs(content_root)
|
||||
|
||||
if not video_dirs:
|
||||
logger.warning("No video files found anywhere under %s", content_root)
|
||||
return 0
|
||||
|
||||
# Build work plan with existing transcript counts
|
||||
total_videos = 0
|
||||
total_existing = 0
|
||||
work_plan = []
|
||||
|
||||
for item in video_dirs:
|
||||
# Output mirrors the relative source structure
|
||||
out_dir = output_root / item["rel_path"]
|
||||
n_existing = count_existing_transcripts(out_dir)
|
||||
n_remaining = item["video_count"] - n_existing
|
||||
|
||||
total_videos += item["video_count"]
|
||||
total_existing += n_existing
|
||||
|
||||
work_plan.append({
|
||||
**item,
|
||||
"output": out_dir,
|
||||
"existing": n_existing,
|
||||
"remaining": n_remaining,
|
||||
})
|
||||
|
||||
# Aggregate by creator for summary
|
||||
creator_stats = {}
|
||||
for item in work_plan:
|
||||
c = item["creator"]
|
||||
if c not in creator_stats:
|
||||
creator_stats[c] = {"videos": 0, "existing": 0, "remaining": 0, "dirs": 0}
|
||||
creator_stats[c]["videos"] += item["video_count"]
|
||||
creator_stats[c]["existing"] += item["existing"]
|
||||
creator_stats[c]["remaining"] += item["remaining"]
|
||||
creator_stats[c]["dirs"] += 1
|
||||
|
||||
logger.info("=" * 70)
|
||||
logger.info("BATCH TRANSCRIPTION PLAN")
|
||||
logger.info("=" * 70)
|
||||
logger.info("Content root: %s", content_root)
|
||||
logger.info("Output root: %s", output_root)
|
||||
logger.info("Creators: %d", len(creator_stats))
|
||||
logger.info("Directories: %d", len(work_plan))
|
||||
logger.info("Total videos: %d", total_videos)
|
||||
logger.info("Already done: %d", total_existing)
|
||||
logger.info("Remaining: %d", total_videos - total_existing)
|
||||
logger.info("=" * 70)
|
||||
|
||||
for name in sorted(creator_stats.keys()):
|
||||
s = creator_stats[name]
|
||||
status = "DONE" if s["remaining"] == 0 else f"{s['remaining']} to do"
|
||||
logger.info(
|
||||
" %-35s %4d videos (%d dirs), %4d done [%s]",
|
||||
name, s["videos"], s["dirs"], s["existing"], status,
|
||||
)
|
||||
|
||||
logger.info("=" * 70)
|
||||
|
||||
if args.dry_run:
|
||||
logger.info("DRY RUN — exiting without transcribing.")
|
||||
return 0
|
||||
|
||||
# Execute
|
||||
manifest = {
|
||||
"started_at": datetime.now(timezone.utc).isoformat(),
|
||||
"content_root": str(content_root),
|
||||
"output_root": str(output_root),
|
||||
"model": args.model,
|
||||
"device": args.device,
|
||||
"results": [],
|
||||
}
|
||||
|
||||
total_processed = 0
|
||||
total_skipped = 0
|
||||
total_failed_dirs = 0
|
||||
batch_start = time.time()
|
||||
|
||||
for i, item in enumerate(work_plan, 1):
|
||||
if item["remaining"] == 0:
|
||||
logger.info("[%d/%d] SKIP %s (all %d videos already transcribed)",
|
||||
i, len(work_plan), item["rel_path"], item["video_count"])
|
||||
total_skipped += item["video_count"]
|
||||
continue
|
||||
|
||||
logger.info("=" * 70)
|
||||
logger.info("[%d/%d] TRANSCRIBING: %s (creator: %s)",
|
||||
i, len(work_plan), item["rel_path"], item["creator"])
|
||||
logger.info(" %d videos, %d remaining",
|
||||
item["video_count"], item["remaining"])
|
||||
logger.info("=" * 70)
|
||||
|
||||
dir_start = time.time()
|
||||
|
||||
cmd = [
|
||||
args.python,
|
||||
str(script_path),
|
||||
"--input", str(item["folder"]),
|
||||
"--output-dir", str(item["output"]),
|
||||
"--model", args.model,
|
||||
"--device", args.device,
|
||||
"--creator", item["creator"],
|
||||
]
|
||||
|
||||
try:
|
||||
result = subprocess.run(
|
||||
cmd,
|
||||
capture_output=False,
|
||||
text=True,
|
||||
timeout=14400, # 4-hour timeout per directory
|
||||
)
|
||||
dir_elapsed = time.time() - dir_start
|
||||
|
||||
n_after = count_existing_transcripts(item["output"])
|
||||
n_new = n_after - item["existing"]
|
||||
|
||||
manifest["results"].append({
|
||||
"creator": item["creator"],
|
||||
"rel_path": str(item["rel_path"]),
|
||||
"videos": item["video_count"],
|
||||
"new_transcripts": n_new,
|
||||
"total_transcripts": n_after,
|
||||
"exit_code": result.returncode,
|
||||
"elapsed_seconds": round(dir_elapsed, 1),
|
||||
})
|
||||
|
||||
if result.returncode == 0:
|
||||
total_processed += n_new
|
||||
total_skipped += item["existing"]
|
||||
logger.info("Completed %s: %d new transcripts in %.1f s",
|
||||
item["rel_path"], n_new, dir_elapsed)
|
||||
else:
|
||||
total_failed_dirs += 1
|
||||
logger.error("FAILED %s (exit code %d) after %.1f s",
|
||||
item["rel_path"], result.returncode, dir_elapsed)
|
||||
|
||||
except subprocess.TimeoutExpired:
|
||||
total_failed_dirs += 1
|
||||
logger.error("TIMEOUT: %s exceeded 4-hour limit", item["rel_path"])
|
||||
manifest["results"].append({
|
||||
"creator": item["creator"],
|
||||
"rel_path": str(item["rel_path"]),
|
||||
"videos": item["video_count"],
|
||||
"error": "timeout",
|
||||
})
|
||||
except Exception as exc:
|
||||
total_failed_dirs += 1
|
||||
logger.exception("ERROR processing %s: %s", item["rel_path"], exc)
|
||||
manifest["results"].append({
|
||||
"creator": item["creator"],
|
||||
"rel_path": str(item["rel_path"]),
|
||||
"videos": item["video_count"],
|
||||
"error": str(exc),
|
||||
})
|
||||
|
||||
batch_elapsed = time.time() - batch_start
|
||||
manifest["completed_at"] = datetime.now(timezone.utc).isoformat()
|
||||
manifest["elapsed_seconds"] = round(batch_elapsed, 1)
|
||||
manifest["summary"] = {
|
||||
"processed": total_processed,
|
||||
"skipped": total_skipped,
|
||||
"failed_dirs": total_failed_dirs,
|
||||
}
|
||||
|
||||
# Write manifest
|
||||
manifest_path = output_root / "batch_manifest.json"
|
||||
output_root.mkdir(parents=True, exist_ok=True)
|
||||
with open(manifest_path, "w", encoding="utf-8") as f:
|
||||
json.dump(manifest, f, indent=2, ensure_ascii=False)
|
||||
|
||||
logger.info("=" * 70)
|
||||
logger.info("BATCH COMPLETE")
|
||||
logger.info(" Processed: %d new transcripts", total_processed)
|
||||
logger.info(" Skipped: %d (already existed)", total_skipped)
|
||||
logger.info(" Failed dirs: %d", total_failed_dirs)
|
||||
logger.info(" Elapsed: %.1f s (%.1f hours)", batch_elapsed, batch_elapsed / 3600)
|
||||
logger.info(" Manifest: %s", manifest_path)
|
||||
logger.info("=" * 70)
|
||||
|
||||
return 1 if total_failed_dirs > 0 else 0
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
sys.exit(main())
|
||||
Loading…
Add table
Reference in a new issue