podcast-ad-remover

KO | EN

gitlite — search

#hacktoberfest #ai #typescript #go #machine-learning #deep-learning #docker #windows #kubernetes #vue #pytorch #security #mcp #open-source

podcast-ad-remover

(★ 25)

Podcast Ad Remover (AGPAR) - an AI-powered tool that automatically downloads podcasts, transcribes them (Whisper), detects ads (Gemini/GPT/Claude), and removes them

podcast-ad-remover 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Podcast Ad Remover

Podcast Ad Remover is a self-hosted app for downloading podcast episodes, removing ad and promo segments, and publishing replacement RSS feeds that can be subscribed to from a normal podcast client.

It is built for homelab-style deployment: one Docker container, SQLite state under `/data`, local transcription with Whisper/faster-whisper, FFmpeg audio editing, and LLM-backed ad detection through Gemini, OpenAI, Anthropic, or OpenRouter.

## Current UI

### Dashboard

The compact dashboard switches between My Podcasts and the global Library without a page reload,
refreshes processing progress independently, and offers direct RSS or generic podcast-app setup for
each feed.

![Dashboard](https://raw.githubusercontent.com/jdcb4/podcast-ad-remover/master/Documentation/screenshots/dashboard.png)

### Episode Management

Episode pages expose reprocessing, manual downloads, ignore/delete actions, AI summaries, descriptions, playback, and podcast-client subscription links.

![Episode management](https://raw.githubusercontent.com/jdcb4/podcast-ad-remover/master/Documentation/screenshots/episodes.png)

### Public Subscribe Page

The optional `/subscribe` page is read-only and designed for people who only need to subscribe to existing feeds. It does not allow adding, editing, reprocessing, or deleting podcasts.

![Public subscribe page](https://raw.githubusercontent.com/jdcb4/podcast-ad-remover/master/Documentation/screenshots/public-subscribe.png)

### Operations Dashboard

The admin queue shows active jobs, queued/retry states, disk usage, next feed checks, and recently processed items.

![Operations dashboard](https://raw.githubusercontent.com/jdcb4/podcast-ad-remover/master/Documentation/screenshots/admin-queue.png)

## Features

- Podcast search, RSS subscription management, and direct YouTube channel/playlist subscriptions.
- Global podcast library with per-user My Podcasts lists and one shared copy of each podcast.
- In-place Library starring and My Podcasts/Library switching plus grid, artwork, and compact table views.
- Podcast ownership rules for per-podcast settings, with admin owner reassignment and admin-only global deletion.
- Grouped global-setting inheritance for content removal, retention, default features, and custom instructions.
- Permission-checked bulk editing for podcast settings and ownership.
- Optional ad-free badge composited onto generated podcast artwork.
- Automatic episode download and retention controls.
- Local transcription with Whisper/faster-whisper.
- LLM-based ad, promo, intro, and outro detection.
- FFmpeg-based audio cutting and rewritten RSS feed generation.
- Per-podcast feeds plus a unified feed.
- Optional AI episode summaries and spoken title intros using either local Piper TTS or Gemini TTS.
- Durable SQLite-backed processing jobs with retry and rate-limit states.
- A bounded, independently refreshed current-processing panel that does not disturb dashboard state.
- Admin queue/operations dashboard.
- Optional token-protected AI/automation REST API with scoped tokens and configurable rate limits.
- Optional management login.
- Optional feed/audio authentication with generated feed tokens.
- Optional public read-only subscribe page for frictionless podcast-client setup.
- Optional Apprise-backed admin notifications for access requests, new podcasts, completed episodes, and breaking processing errors.
- Resource controls for Whisper CPU threads, FFmpeg threads, and unloading Whisper after jobs.

## YouTube Sources

Paste a direct public YouTube channel URL or an explicit `youtube.com/playlist?list=...` URL into the
dashboard search bar. The app resolves it as one subscription, checks it on the normal feed interval,
downloads best audio-only with pinned yt-dlp tooling, and sends that audio through the same
Whisper/LLM/FFmpeg pipeline. Individual videos, Shorts, streams, private/member content, login
cookies, and YouTube search are not supported.

YouTube extraction is intended for self-hosted, opt-in use with public content. Operators are
responsible for ensuring they have permission to download and process the content and that their use
complies with applicable law and platform terms.

SponsorBlock timestamp lookup is implemented but disabled by default. It makes no API request unless
`SPONSORBLOCK_ENABLED=true` is explicitly set. SponsorBlock API/database data is licensed
[CC BY-NC-SA 4.0](https://github.com/ajayyy/SponsorBlock/wiki/Database-and-API-License); enabling it
means the operator has determined that their use complies with that licence. SponsorBlock is
attributed to the [SponsorBlock project](https://github.com/ajayyy/SponsorBlock).

## Quick Start

### Docker Run

Run the published image and mount `/data` somewhere persistent:

```bash
docker run -d \
  --name podcast-ad-remover \
  -p 8000:8000 \
  -v ./data:/data \
  -e SESSION_SECRET_KEY="$(openssl rand -hex 32)" \
  -e BASE_URL="http://localhost:8000" \
  -e GEMINI_API_KEY="your_api_key" \
  jdcb4/podcast-ad-remover:latest
```

Open `http://localhost:8000`, then go to **Admin > AI Settings > Text Analysis** to confirm your provider and model settings.

Keep `SESSION_SECRET_KEY` stable once set. Changing it can invalidate browser sessions and signed feed tokens.

### Docker Compose

For local development from source:

```bash
cp env.example .env
docker compose up -d --build
```

For a production-style compose file using the published image, see `docker-compose.prod.yml` and [Documentation/Deployment.md](Documentation/Deployment.md).

### Unraid

A dedicated Unraid template is included at `Documentation/unraid/podcast-ad-remover.xml`. See [Documentation/Unraid_Deployment.md](Documentation/Unraid_Deployment.md).

## AI Providers

Gemini is the recommended default for most personal installs because the free tier is usually enough for this app's transcript-analysis workload. Gemini direct access uses Google's OpenAI-compatible endpoint through the OpenAI Python SDK.

The default Gemini cascade is:

1. `gemini-3.5-flash`
2. `gemini-3-flash`
3. `gemini-3.1-flash-lite`
4. `gemini-2.5-flash`
5. `gemini-2.5-flash-lite`

The app tries each configured model in order and falls back when a model is unavailable, fails, or hits a rate limit. OpenRouter uses the same order with `google/` model IDs.

Admins can also select **Custom OpenAI-compatible / Local** and provide an explicit API base URL plus
arbitrary model slug for Ollama, LocalAI, vLLM, or another compatible service. This is opt-in: Gemini
and the existing cloud cascades remain the defaults, and saved OpenAI cloud credentials are never
forwarded to a custom endpoint.

Current Gemini free-tier limits recorded for these defaults:

| Model | Category | RPM | TPM | RPD |
|-------|----------|-----|-----|-----|
| Gemini 2.5 Flash | Text-out models | 5 | 250K | 20 |
| Gemini 3 Flash | Text-out models | 5 | 250K | 20 |
| Gemini 2.5 Flash Lite | Text-out models | 10 | 250K | 20 |
| Gemini 3.1 Flash Lite | Text-out models | 15 | 250K | 500 |
| Gemini 3.5 Flash | Text-out models | 5 | 250K | 20 |

You can set keys in the Admin UI or with environment variables:

- `GEMINI_API_KEY`
- `OPENAI_API_KEY`
- `ANTHROPIC_API_KEY`
- `OPENROUTER_API_KEY`

Custom endpoint credentials are configured only in the Admin UI and may be left blank for keyless
local servers.

## Text-To-Speech

Piper remains the default TTS provider because it is local and does not consume API quota. Admins can optionally switch spoken title intros and audio summaries to Gemini TTS from **Admin > AI Settings > Voice and TTS**.

Gemini TTS uses the saved Gemini API keys and this default fallback order:

1. `gemini-3.1-flash-tts-preview`
2. `gemini-2.5-flash-preview-tts`

Available Gemini voices are `Orus` (default), `Enceladus`, and `Laomedeia`. Current free-tier limits recorded for each Gemini TTS model are 3 RPM, 10K TPM, and 10 RPD.

## Authentication And Feed Access

The app supports separate choices for management access and podcast feed access:

- Management pages can require login.
- The public `/subscribe` page can remain available without login.
- Feeds and audio can remain unauthenticated for the smoothest podcast-client setup.
- Feed/audio authentication can be enabled when you want clients to use credentials or generated feed-token URLs.

Protected feed URLs containing `token=` are bearer secrets. Anyone with the full URL can read that feed and download audio until the token is revoked.

## AI API

Admins can enable an optional REST API under `/api/v1` from **Admin > System Settings**. API clients use separate `par_...` bearer tokens with scoped access (`read`, `write`, `process`, `admin`) and SQLite-backed per-token/IP rate limits. See [Documentation/API.md](Documentation/API.md).

## Admin Notifications

Notifications are off by default. Admins can enable them from **Admin > Notifications** by adding one or more Apprise URLs and selecting which events should send alerts.

Initial supported events are:

- user access requested
- new podcast added to the global library
- episode processed and available in feeds
- breaking processing errors, including max-retry failures and top-level worker errors

Apprise supports many targets, including ntfy, Gotify, Pushover, Discord, Telegram, Slack, email/SMTP, webhooks, and SMS providers. Treat notification URLs as secrets because many contain tokens or webhook credentials.

## Persistent Data

Mount `/data` in Docker. It contains:

- `/data/db/podcasts.db`
- `/data/podcasts/`
- `/data/feeds/`
- `/data/artwork/`
- `/data/models/`
- `/data/backups/`
- logs

Database migrations are designed to preserve existing installs. Formal migrations create backups under `/data/backups/` before applying schema changes.

New podcasts inherit all four global setting groups. Upgraded podcasts keep their existing explicit
content-removal, retention, and feature values; only podcasts whose custom instructions were blank or
NULL begin inheriting global custom instructions. The stored podcast values are retained while a group
inherits, so disabling inheritance restores the podcast's previous overrides.

## Verification

Before completing code changes:

```bash
npm run verify
```

For Docker/release work:

```bash
npm run verify:docker
```

Normal development integrates into `dev`. A clean committed Dev revision can be built or published with rolling and commit-specific tags:

```bash
npm run docker:dev
npm run docker:dev:publish
```

These produce `jdcb4/podcast-ad-remover:dev` and `jdcb4/podcast-ad-remover:dev-<git-sha>` without changing production tags.

Experimental branch images can be published without touching `latest`:

```bash
npm run docker:experimental -- --push --tag experimental
```

Experimental Apple Silicon / ARM64 images can be built without Piper TTS:

```bash
npm run docker:experimental:arm64 -- --push
```

`linux/amd64` remains the primary release target. The ARM64 experimental image skips Piper because its phonemizer dependency is not currently available as a simple Linux arm64 wheel. Podcast download, local transcription, ad detection, cutting, feeds, and the web UI remain the target feature set. Spoken summaries and title intros can still be tested on no-Piper images by selecting Gemini TTS and configuring a Gemini API key.

Production promotion from `dev` to `master` requires explicit approval. Release publishing then runs from a clean `master` checkout and tags both the version and `latest`:

```bash
npm run docker:publish
```

## Documentation

- [Project Index](Documentation/PROJECT_INDEX.md)
- [AI API](Documentation/API.md)
- [Architecture](Documentation/Architecture.md)
- [Deployment](Documentation/Deployment.md)
- [Environment Variables](Documentation/Environment_Variables.md)
- [Verification](Documentation/VERIFICATION.md)
- [Versioning](Documentation/VERSIONING.md)
- [Changelog](Documentation/CHANGELOG.md)
- [Decisions](Documentation/DECISIONS.md)
- [Resource Audit](Documentation/RESOURCE_AUDIT.md)
- [Local LLM Evaluation](Documentation/LOCAL_LLM_EVALUATION.md)
- [Local LLM HTML Comparison Report](Documentation/LOCAL_LLM_EVALUATION_REPORT.html)
- [Roadmap](Documentation/ROADMAP.md)
- [Naming](Documentation/NAMING.md)
- [Security](SECURITY.md)

## License

MIT License

podcast-ad-remover

podcast-ad-remover 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

podcast-ad-remover

podcast-ad-remover 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README