text-to-speech · benchflow-ai — Claude Code skill for TTS audio mastering
The text-to-speech skill guides TTS cleanup, loudness matching, and segment timing for video audio exports.
Apache-2.0 4 installs ★ 1,933 updated 2026-10-24
npx skills add benchflow-ai/skillsbench --skill text-to-speech -a claude-codeWhat text-to-speech does
The skill covers preparing speech clips for delivery: filtering low rumble, optionally softening harsh high frequencies, and adding short fades at clip edges. It uses FFmpeg ebur128 to measure loudness and recommends loudnorm after cleanup and timing edits.
When assembling a track, match each clip to its target time window. Add silence to a short clip; gently adjust or carefully trim a long one, then apply fades at the boundaries. If duration changes after loudness normalization, run normalization again. Otherwise, the final level may no longer match the intended target.
The appropriate target depends on the task. The skill gives -23 LUFS and about -1.5 dBTP as common reference values. Check the generated audio’s native sample rate before resampling. It also outlines choosing between offline, cloud, and formant TTS engines.
What it helps with
- Remove rumble and clicks from TTS clips
- Normalize speech loudness before export
- Fit spoken segments to target time windows
- Assemble TTS clips into a video track
Example prompts for the agent
- Clean these TTS clips and match their loudness for a video export.
- Assemble the speech segments and fit each one to its target window.
- Recheck loudness after changing the duration of the audio.
How to install text-to-speech
1Skills CLI (Vercel)
npx skills add benchflow-ai/skillsbench --skill text-to-speech -a claude-code2Manually: put the files in place
~/.claude/skills/text-to-speech/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/text-to-speech/
1Skills CLI (Vercel)
npx skills add benchflow-ai/skillsbench --skill text-to-speech -a codex2Manually: put the files in place
.agents/skills/text-to-speech/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add benchflow-ai/skillsbench --skill text-to-speech -a cursor2Manually: put the files in place
.cursor/skills/text-to-speech/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install benchflow-ai/skillsbench tasks/multilingual-video-dubbing/environment/skills/text-to-speech2Manually: put the files in place
.github/skills/text-to-speech/SKILL.md1Skills CLI (Vercel)
npx skills add benchflow-ai/skillsbench --skill text-to-speech -a gemini-cli2Platform CLI
gemini skills install https://github.com/benchflow-ai/skillsbench --path tasks/multilingual-video-dubbing/environment/skills/text-to-speechManually: put the files in place
.devin/skills/text-to-speech/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add benchflow-ai/skillsbench --skill text-to-speech -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about text-to-speech
How do I install text-to-speech in Claude Code?
Run npx skills add benchflow-ai/skillsbench --skill text-to-speech -a claude-code. The skill file is at ~/.claude/skills/text-to-speech/SKILL.md.
What does text-to-speech do?
It provides steps for cleaning TTS audio, matching loudness, and aligning speech segments before video delivery.
How should I handle a segment that misses its target duration?
Pad a short clip with silence. For a long one, make a gentle speed adjustment or trim it carefully, then add fades at the boundaries.
Source text
SKILL.md · Apache-2.0
--- name: text-to-speech description: "Practical mastering steps for TTS audio: cleanup, loudness normalization, alignment, and delivery specs." --- # SKILL: TTS Audio Mastering This skill focuses on producing clean, consistent, and delivery-ready TTS audio for video tasks. It covers speech cleanup, loudness normalization, segment boundaries, and export specs. ## 1. TTS Engine & Output Basics Choose a TTS engine based on deployment constraints and quality needs: * **Neural offline** (e.g., Kokoro): stable, high quality, no network dependency. * **Cloud TTS** (e.g., Edge-TTS / OpenAI TTS): convenient, higher naturalness but network-dependent. * **Formant TTS** (e.g., espeak-ng): for prototyping only; often less natural. **Key rule:** Always confirm the **native sample rate** of the generated audio before resampling for video delivery. --- ## 2. Speech Cleanup (Per Segment) Apply lightweight processing to avoid common artifacts: * **Rumble/DC removal:** high-pass filter around **20 Hz** * **Harshness control:** optional low-pass around **16 kHz** (helps remove digital fizz) * **Click/pop prevention:** short fades at boundaries (e.g., **50 ms** fade-in and fade-out) Recommended FFmpeg pattern (example): * Add filters in a single chain, and keep them consistent across segments. --- ## 3. Loudness Normalization Target loudness depends on the benchmark/task spec. A common target is ITU-R BS.1770 loudness measurement: * **Integrated loudness:** **-23 LUFS** * **True peak:** around **-1.5 dBTP** * **LRA:** around **11** (optional) Recommended workflow: 1. **Measure loudness** using FFmpeg `ebur128` (or equivalent meter). 2. **Apply normalization** (e.g., `loudnorm`) as the final step after cleanup and timing edits. 3. If you adjust tempo/duration after normalization, re-normalize again. --- ## 4. Timing & Segment Boundary Handling When stitching segment-level TTS into a full track: * Match each segment to its target window as closely as possible. * If a segment is shorter than its window, pad with silence. * If a segment is longer, use gentle duration control (small speed change) or truncate carefully. * Always apply boundary fades after padding/trimming to avoid clicks. **Sync guideline:** keep end-to-end drift small (e.g., **<= 0.2s**) unless the task states otherwise.
© benchflow-ai · Apache-2.0 · benchflow-ai/skillsbench
Similar skills
All skills for AI and ML →text-to-speech by other authors comes first.