Speech Generation

Speech Generation (Text-to-Speech)

Step 0 — plan first. Before generating, run the visual-planning skill (the pre-generation gate for all media, audio included): clarify the ask (voice, language, format, where it's used), pin the tone/pacing, confirm the exact script and pronunciations, then generate and verify.

Use the mcp__ElevenLabs__text_to_speech tool to convert text to natural-sounding speech via ElevenLabs.

Quick Reference

When to Use

• User asks to "read this aloud", "create a voiceover", "generate narration"
• User wants audio versions of text content
• User needs podcast-style audio, announcements, or voice prompts
• User is creating audio for a video or presentation

Tool Reference

text_to_speech

Key parameters:

Only one of voice_name or voice_id can be provided. If neither is given, the default voice is used.

Models

For detailed model comparison and selection guidance, see voice-settings.md.

Voice Selection

Use mcp__ElevenLabs__search_voices to find voices already in the user's library, or mcp__ElevenLabs__search_voice_library to browse the full ElevenLabs voice library.

Matching voices to content

Voice design

Use mcp__ElevenLabs__text_to_voice to generate a custom voice from a text description (e.g. "A warm female voice with a slight British accent"). This creates 3 preview variations. Save the best one with mcp__ElevenLabs__create_voice_from_preview.

Voice cloning

Use mcp__ElevenLabs__voice_clone to create an instant voice clone from audio files.

Voice Controls

Controls are summarized below. For parameter interaction details, named presets, and advanced tuning, see voice-settings.md.

Output Formats

For platform-specific format recommendations, see announcement.md.

Common Patterns

Documentation narration

See documentation.md for text cleaning, code block handling, and segmentation.

text: [cleaned documentation content]
voice_name: "Rachel"
model_id: "eleven_v3"
stability: 0.55
similarity_boost: 0.80
style: 0.10
speed: 0.90

Video voiceover

See voiceover.md for timing tables, text prep, ffmpeg mixing, the timed voiceover pipeline, and golden rules.

text: "Introducing our new dashboard. With real-time analytics,
      you can track performance at a glance."
voice_name: "Adam"
model_id: "eleven_v3"
stability: 0.50
similarity_boost: 0.80
style: 0.25
speed: 0.95

Announcement / notification

See announcement.md for brevity rules, templates, and platform formats.

text: "Deployment complete. All 12 services are running."
voice_name: "Bella"
model_id: "eleven_v3"
stability: 0.75
similarity_boost: 0.85
style: 0.05
speed: 1.05

Story / narrative

See narration.md for character voices, long-form strategy, and pacing.

text: [story content]
voice_name: "Rachel"
model_id: "eleven_v3"
stability: 0.25
similarity_boost: 0.80
style: 0.50
speed: 0.90

Multi-language

See multilingual.md for language tiers, pronunciation hints, and batch workflows.

Set language to the ISO 639-1 code and write text in the target language. Use eleven_v3 (70+ languages) or eleven_multilingual_v2 (29 languages).

Podcast / multi-speaker

See podcast.md for multi-speaker workflow, voice pairing, and episode structure.

Generate each speaker's lines separately with different voices, then concatenate with natural gaps.

Additional ElevenLabs Tools

Combining with Other Media Skills

Voiceover + Video

1. Generate the video with generate_video
2. Generate the voiceover with mcp__ElevenLabs__text_to_speech
3. Combine with ffmpeg: ffmpeg -i video.mp4 -i voiceover.mp3 -c:v copy -c:a aac output.mp4

Narration + Background Music

1. Generate narration with mcp__ElevenLabs__text_to_speech
2. Generate background music with generate_music
3. Mix audio: ffmpeg -i narration.mp3 -i music.wav -filter_complex "[1:a]volume=0.2[bg];[0:a][bg]amix=inputs=2:duration=longest" output.mp3

Podcast Production

1. Generate each speaker's lines separately with different voices
2. Generate intro/outro jingle with generate_music
3. Concatenate: ffmpeg -i intro.wav -i part1.mp3 -i part2.mp3 -i outro.wav -filter_complex "concat=n=4:v=0:a=1" podcast.mp3

See podcast.md for the complete step-by-step workflow.

Tips

• Write text as you want it spoken — use punctuation for natural pauses
• Use "..." for longer pauses: "And the winner is... congratulations!"
• Spell out abbreviations if you want them read as words: "API" vs "A.P.I."
• For numbers, write them as words if pronunciation matters: "twenty-three" vs "23"
• Test with a short sentence first to verify the voice before generating long content
• Use output_directory to control where files are saved (defaults to ~/Desktop)
• For advanced parameter tuning, consult voice-settings.md
• For video voiceover, follow the Golden Rules and Timed Voiceover Pipeline
• Check voice tiers before selecting a voice — library/cloned voices need a paid plan
• Use the Voice Selection Guide to match voice type to content — always test the voice before writing the full script

Reference Files