kokoro-tts
Skill2 starsChanged 35 days agoMITView on GitHub
Turn text into natural spoken audio (WAV/MP3) on your own computer with Kokoro, a small open text-to-speech model. No API key, no online service, runs on the CPU faster than real time. Use for a voice-over or narration for a video, reading a script or an article aloud, an audio version of a document, a spoken message, or whenever the person asks for a voice, "say this", TTS or text to speech. Gives per-line timings so a video's cuts and captions can follow the voice. Voices in American and British English, Spanish, French, Italian, Portuguese, Hindi, Japanese and Chinese.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- gabrielkoerich/skillstheir own version, first seen Nov 1, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- gabrielkoerich/skills★ 8same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository10 other files it gives its agents.
- chrome-devtoolsSkill
- computer-use-linuxSkill
- create-skillSkill
- docxSkill
- find-skillsSkill
- kinetic-reelSkill
- mcp-builderSkill
- painted-animationSkill
- pptxSkill
- xlsxSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 74
Write the script first, speak it, and only then lay out the video: each shot starts at its line's start from voice.json, and the video is …
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Kokoro text to speech
Kokoro-82M (Apache 2.0) through kokoro-onnx. About four seconds of speech per second of work on a plain CPU, so
a five-minute narration takes a minute or two. It needs about 1 GB of memory, which fits the everyday limit: no
computer_memory request is needed for speech alone.
1. Set up (once per computer)
bash <skill directory>/scripts/setup.shThe skill directory is the line at the very top of this skill. The script makes a Python environment and downloads
the model (about 350 MB) into /shared/kokoro, where every agent on this account can use it. When it is already
there the script returns at once, so run it every time.
2. Speak
/shared/kokoro/venv/bin/python <skill directory>/scripts/speak.py script.txt --out out/voice.wav --voice af_heart
/shared/kokoro/venv/bin/python <skill directory>/scripts/speak.py --text "One short line." --out out/line.wavscript.txt: one sentence or beat per line. A blank line is a longer pause. Each line is spoken on its own, so a line is also the unit you can time a cut or a caption to.--voice(list below),--speed 0.8..1.3(1.0 default),--gap 0.25seconds between lines,--pause 0.7seconds for a blank line,--langonly when the voice's own language is not the text's.- It prints the total length and writes
out/voice.jsonbeside the audio:[{ "text", "start", "end" }, ...]in seconds. Build the video's timeline from that file instead of guessing how long a line takes. - MP3 or AAC:
ffmpeg -y -i out/voice.wav -af loudnorm=I=-16:TP=-1.5 -b:a 160k out/voice.mp3.
3. Voices
The first letter is the language, the second is f female or m male.
| Language | --lang | Good first choices | Also |
| American English | en-us | af_heart, af_bella, am_michael, am_fenrir | af_nicole (soft), af_sarah, af_sky, am_adam, am_puck, am_onyx |
| British English | en-gb | bf_emma, bm_george | bf_isabella, bf_lily, bm_fable, bm_daniel, bm_lewis |
| Spanish | es | ef_dora, em_alex | |
| French | fr-fr | ff_siwis | |
| Italian | it | if_sara, im_nicola | |
| Portuguese (Brazil) | pt-br | pf_dora, pm_alex | |
| Hindi | hi | hf_alpha, hm_omega | hf_beta, hm_psi |
| Japanese | ja | jf_alpha, jm_kumo | |
| Chinese | cmn | zf_xiaoxiao, zm_yunxi |
speak.py --voices prints every voice. There is no voice for other languages (Persian, Arabic, Turkish, German,
Russian and so on): say so plainly instead of reading the text with an English voice, which comes out as nonsense.
Writing for the ear
- Short sentences, one idea each. Write numbers and symbols the way they are said: "twelve percent", "version two point five", "kuru dot app". Spell out an abbreviation the first time unless it is said as a word.
- Fix a mispronounced name by respelling it the way it sounds ("Kooroo"), then listen again.
- A comma is a short breath and a full stop a longer one. Use them to pace the read; do not write "(pause)".
- One voice for the whole piece. Change the voice only for a second speaker.
Check it before you use it
You cannot hear the file, so measure it:
ffprobe -v error -show_entries format=duration -of csv=p=0 out/voice.wav # the length you expect?
ffmpeg -i out/voice.wav -af volumedetect -f null - 2>&1 | grep -E "mean_volume|max_volume" # not silent, not clippedvoice.json should have one entry per spoken line, and no line should be far longer than its text suggests. Say
in the report that the audio was checked by measurement, not by ear.
With a video
Write the script first, speak it, and only then lay out the video: each shot starts at its line's start from
voice.json, and the video is as long as the voice plus a beat at each end. Mix the voice over music by ducking
the music, not by raising the voice:
ffmpeg -y -i out/voice.wav -i music.m4a -filter_complex "[1:a]volume=0.25[m];[0:a][m]amix=inputs=2:duration=first:normalize=0,loudnorm=I=-14:TP=-1.2[a]" -map "[a]" -c:a aac -b:a 192k out/mix.m4aDoes 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
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.