agentleFS
GoogleKnown publishergoogle-gemini/gemini-managed-agents-templates/ai-radio/skills/tts-generation/SKILL.md

tts-generation

Skill99 starsChanged 6 months agoApache-2.0View on GitHub

Convert radio show script to speech audio with telephone effect on correspondent voices using the Interactions API.

The file

Files
SKILL.md

TTS Generation

Convert the radio show script into speech audio using the Gemini TTS model via the Interactions API. Apply a telephone bandpass filter to correspondent voices so they sound like phone call-ins, while keeping the host's voice clean studio-quality.

Embedded Script

python3 skills/tts-generation/scripts/generate_tts.py --workspace ./workspace

Arguments

ArgumentDefaultDescription
--workspaceworkspaceRoot workspace directory
--workers8Max parallel TTS worker threads

What it does

  1. Reads the script from {workspace}/data/script.md.
  2. Parses it into individual (speaker, text) turns and assigns voices.
  3. Generates TTS for all turns in parallel using a thread pool (default 8 workers).
  4. Retries each failed turn up to 3 times with exponential backoff.
  5. Applies an ffmpeg telephone bandpass filter (300Hz–3.4kHz) to correspondent voices.
  6. Keeps the host (Paul) audio clean and unfiltered.
  7. Concatenates all segments in original script order into a single WAV.

Dependencies

  • google-genai (>= 2.0.0)
  • ffmpeg (system)

API Details

Uses the Interactions API with single-speaker TTS — no multi-speaker workaround needed.

Voice Assignment

Voices are assigned dynamically based on [Male] / [Female] gender tags in the script:

SpeakerVoiceAudio Treatment
Paul (host)PuckClean — no filter
Caller [Female] — 1stKoreTelephone filter
Caller [Female] — 2ndAoedeTelephone filter
Caller [Male] — 1stCharonTelephone filter
Caller [Male] — 2ndFenrirTelephone filter

Voices cycle round-robin if there are more callers than available voices. Accent tags ([Accent: Irish], etc.) are injected into the TTS prompt to influence pronunciation.

Telephone Filter

Applied via ffmpeg to correspondent audio segments:

highpass=f=300, lowpass=f=3400, acompressor, volume=1.5

This simulates the standard telephone bandwidth (300Hz–3.4kHz) and adds compression to mimic phone codec dynamics.

Output

  • Primary output: {workspace}/audio/speech/speech.wav
  • Format: WAV, 24kHz, 16-bit PCM, mono
  • Intermediate segments: {workspace}/audio/speech/segments/turn_*.wav

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.