Dual-Host Daily Podcast Generator

Automated daily podcast with two AI hosts. Generates text brief + dual-voice audio, publishes to RSS, delivers via messaging.

Concept

• Format: Two hosts — one explains/analyzes, the other asks and transitions
• Duration: Configurable, default ~7 minutes
• Style: Casual, opinionated, conversational — like two friends chatting about the news
• Topics: Customizable (default: AI/Tech, Stocks, Macro, Crypto)

Architecture

Fetch News → Text Brief → Dual-Voice Script → TTS Audio → S3 Upload → RSS Update → Deliver

Configuration

Set these in your environment:

Variable	Description
S3_BUCKET	S3 bucket name
PODCAST_DOMAIN	Custom domain or S3 URL
FISH_API_KEY	Fish Audio API key (https://fish.audio)
FISH_VOICE_A	Fish Audio voice ID for Host A
FISH_VOICE_B	Fish Audio voice ID for Host B

Step 1: Fetch News

Use web_fetch to scrape sources in parallel. Default sources:

1. https://news.ycombinator.com/ — Tech
2. https://www.coindesk.com/ — Crypto
3. https://techcrunch.com/category/artificial-intelligence/ — AI
4. https://finance.yahoo.com/ — Markets

Customize sources to match your podcast topic.

Step 2: Generate Text Brief

Organize news into sections with emoji headers:

☀️ Daily Brief | Mar 3, 2026

━━━━━━━━━━━━━━━━━━

🤖 Tech / AI

① Headline
→ One-line take

━━━━━━━━━━━━━━━━━━

📈 Markets

① Headline
→ One-line take

━━━━━━━━━━━━━━━━━━

🎯 Key Takeaway
Summary paragraph

Step 3: Generate Dual-Voice Script

Rewrite the brief as a dialogue. Prefix each line with speaker tag:

HostA: Welcome to today's episode...
HostB: Some big stories today...
HostA: Right, let's start with...

Guidelines: - Host A: Explains and analyzes, knowledgeable but casual - Host B: Asks, transitions, reacts - Substantial turns, not one-liners - Include analysis and discussion, not just headlines - End with a lighter topic + sign-off

Step 4: Generate Audio

Fish Audio (recommended — natural, multi-voice):

python3 scripts/fish_dual_tts.py <script.txt> <output.mp3>

Parses speaker tags, sends each segment to Fish Audio, concatenates into final MP3.

Edge TTS (free fallback, single voice):

edge-tts --voice en-US-GuyNeural --rate "+5%" --file script.txt --write-media output.mp3

Step 5: Publish

bash scripts/generate_episode.sh <date> <EP-number> <title> <description> <mp3-file>

What it does: 1. Upload MP3 to S3 2. Get actual duration via ffprobe 3. Insert <item> into RSS feed (newest first) 4. Update <lastBuildDate>

Step 6: Deliver

Send text brief + audio via your preferred channel (Telegram, Discord, Slack, etc.)

RSS Feed

See references/rss-format.md for XML template.

Key rules: - <itunes:duration> = actual duration from ffprobe (never hardcode) - <enclosure length> = actual file size in bytes - <itunes:owner> with email for Apple/Spotify verification - Cover: 3000x3000 JPEG minimum

Hosting Options

Option	Notes
S3 + Cloudflare Worker	Free HTTPS, recommended
S3 + CloudFront	Native AWS
Any static host	Just serve MP3 + feed.xml

Cron (OpenClaw)

openclaw cron add --task "Generate daily podcast" --cron "0 8 * * *" --tz "Your/Timezone"

Dependencies

• python3 + requests — Fish Audio TTS
• ffmpeg / ffprobe — Audio processing
• aws CLI — S3 upload
• edge-tts (optional) — Free fallback TTS