🎤 TYSA: The Yapping Spotify Announcer

TYSA is your digital radio DJ. A Python script that detects your current Spotify track and announces it with smooth, AI-generated commentary. It's like having a late-night FM host living in your terminal.

Yes, I know this completely outs me as someone who listens to classical music. And yes, that's exactly why GPT simplification exists - because nobody wants to hear "Symphony No. 9 in D Minor, Op. 125 'Choral': IV. Presto - Allegro assai" when you can just say "Ode to Joy." 🎻

(Note: This is for macOS with Spotify Desktop only (4now))

🎙️ The Origin Story

Spotify Wrapped told me I'm in the top 1% of listeners with 100,000 minutes streamed. Impressive, right? Plot twist: play me my top song and I can barely tell you the name or artist. Turns out you can listen to thousands of hours of music and learn absolutely nothing if you're not paying attention.

So I figured, what if I learned unconsciously? If someone announced the song and artist every single time a track played, maybe, just maybe - the information would eventually stick. TYSA is that someone. Now my terminal is a radio DJ that won't shut up, and honestly, it's working.

Features

• GPT-4o-mini simplifies complex classical titles and detects song language
• Multilingual support with automatic language detection
• ElevenLabs TTS for natural voice announcements (v3 or flash_v2_5)
• Localized broadcasts: seamlessly switch between languages mid-announcement
• Caches everything - replays cost $0

Cost Estimate

Characters per song:

• Basic/Smart: ~45 chars (no brackets)
• Wizard: ~90 chars (with language switching brackets)

ElevenLabs free tier (monthly):

• Flash (basic/smart): 20,000 chars = ~444 unique songs
• v3 (wizard): 10,000 chars = ~110 unique songs

Paid tier ($5/month):

• Flash: 60,000 chars = 1,333 songs
• v3: 30,000 chars = 333 songs

Replays: $0 - Everything is cached!

How Long Can I Listen?

10 hours/day = ~171 songs (assuming 3.5 min/song)

Setup

git clone <your-repo-url>
cd tysa
cp .env.example .env

Add your API keys to .env:

• ElevenLabs: https://elevenlabs.io/ (required)
• OpenAI: https://platform.openai.com/api-keys (only for smart/wizard mode)

Run it:

./run.sh

Stop with Ctrl+C or pkill -f tysa.py or use your own terminal multiplexer

Configuration

Variable	Default	What It Does
ELEVENLABS_API_KEY	required	Your ElevenLabs API key
ELEVENLABS_VOICE_ID	RexqLjNzkCjWogguKyff	Voice ID (default: Bradley). Pick a multilingual voice!
OPENAI_API_KEY	smart/wizard only	Your OpenAI API key (not needed for basic mode)
OPENAI_MODEL	gpt-4o-mini	Model for title simplification
MODE	smart	Announcement mode: basic, smart, or wizard
LANGUAGE_CODE	en	Base language (en, sv, de, fr, etc.). Match to your voice!
NOW_PLAYING_TEXT	Now playing	Custom "Now playing" text (basic mode only)
BY_TEXT	by	Custom "by" text (basic mode only)
PLAYBACK_VOLUME	0.5	Volume (0.0=silent, 1.0=max)
POLL_INTERVAL_SECONDS	1	How often to check for track changes
DEBUG	false	Log to terminal (true) or file only (false)
OUTPUT_DIR	output	Where to save generated MP3s
RUN_MODE	continuous	continuous or once

Modes 🎚️

Basic Mode (MODE=basic):

• No GPT, no OpenAI API key needed
• Fastest & cheapest
• User provides announcement strings
• Example: Set NOW_PLAYING_TEXT=Nu spelas and BY_TEXT=av → "Nu spelas: Bohemian Rhapsody - av - Queen"

Smart Mode (MODE=smart):

• GPT translates & simplifies titles
• Uses eleven_flash_v2_5
• Entire announcement in base language
• Example (LANGUAGE_CODE=sv): "Nu spelas: Bohemian Rhapsody - av - Queen"

Wizard Mode 🧙‍♂️ (MODE=wizard):

• GPT + multilingual inline switching
• Uses eleven_v3 with brackets
• Songs pronounced in native language
• Example (LANGUAGE_CODE=sv): "Nu spelas: [read in en]Gaia [read in sv] - av - [read in en]Oliver Ólafsson"

How It Works

1. AppleScript polls Spotify for track changes
2. Announcement generated based on mode:
   • Basic: Direct string substitution (Cheapest and fastest)
   • Smart: GPT simplifies & translates (For my classical friends)
   • Wizard: GPT + multilingual formatting (Reads songs and artists in their native language)
3. ElevenLabs converts to speech (flash_v2_5 or v3)
4. Audio plays via afplay
5. Everything caches for instant replays

Generated files go to output/ like Beethoven_Moonlight_Sonata.mp3

Troubleshooting

"Nothing is currently playing" - Spotify must be running and playing

"Missing required environment variables" - Check .env has ELEVENLABS_API_KEY (and OPENAI_API_KEY if using smart/wizard mode)

Permission errors - System Preferences → Privacy & Security → Automation, grant terminal access to Spotify

Volume issues - Adjust PLAYBACK_VOLUME in .env (0.0 to 1.0)

Advanced

Run once:

RUN_MODE=once ./run.sh

Manual Python:

python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
python tysa.py

Requirements

macOS, Python 3.8+, Spotify Desktop, ElevenLabs API key, OpenAI API key (smart/wizard mode only)

License

MIT