Teacher in the Loop

Teacher in the Loop

Runs small 8M-parameter text-to-speech voices distilled from Kokoro-82M entirely in your browser, with audio samples that compare the training recipes.

Made by MithrilMan on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: Teacher in the Loop is a demo and sample page for a paper on distilling the Kokoro-82M speech model into small single-voice students of 8M parameters, in English and Italian. The three released voices run entirely in your browser, and no server computes anything. A second part lets you listen to audio samples that compare the teacher, other decoders and the new recipe.

Who it suits: Speech researchers and developers who want to hear a tiny text-to-speech model run locally.

Industry | Who uses it | How they use it | Example
Speech research | Researchers in distillation | Compare the teacher with the 8M students | Listening to the same sentence from each model
Software | Developers of offline voice tools | Try an int8 voice that runs on one CPU thread | Checking how a small voice sounds in a browser
Education | Students of speech synthesis | See text turned into phonemes and then speech | Reading how the pieces fit together

If you need voices in other languages, you can skip it, because the released voices cover English and Italian.

Where to use it:

• A web browser, on the Space's page at Hugging Face
• A browser on Windows, macOS or Linux, by opening a downloaded copy of index.html

Use case | Who | What they get back
Try a released voice | Anyone | Speech generated in the browser
Listen to the samples | Researchers | Audio from the teacher and the student recipes for held-out sentences
Study the browser pipeline | Developers | A client-side phoneme and ONNX voice setup to read

How to set it up: Online, nothing needs installing. A local copy needs a browser and a little Python for the download.

• A web browser
• Python and the pip tool, only for the download step
• A network connection, because the page loads the ONNX Runtime Web script from a content delivery network

1. Install the Hugging Face hub package with pip.
2. Download the Space's files into a folder.
3. Move into that folder.
4. Open index.html in a web browser.

pip install -U huggingface_hub
hf download MithrilMan/teacher-in-the-loop --repo-type space --local-dir teacher-in-the-loop
cd teacher-in-the-loop

How to use it: Open the Space's page, or index.html from a local copy. Use the Try it part to type text and hear one of the released voices. Use the Listen part to play the samples for each recipe.

hf download MithrilMan/teacher-in-the-loop --repo-type space --local-dir teacher-in-the-loop

Example: You open the page, type a short English sentence and pick one of the released voices. The page turns the text into phonemes and plays the speech from the 8M student. You then open the samples and compare the same held-out sentence across the recipes.

For newcomers and experts: Distillation trains a small model to copy a big one. To start, type one short sentence in the Try it part and listen to the result.

Text is turned into phonemes by a JavaScript version of Misaki's English rules together with eSpeak NG, which is compiled to WebAssembly. Each voice is a compact int8 ONNX model, executed by ONNX Runtime Web using a single CPU thread. Samples are provided for four sources: the Kokoro teacher, a decoder released by Paradee, the author's re-run of the published recipe, and the new recipe.

Type text in the page
Misaki rules and eSpeak NG make phonemes
The int8 ONNX voice runs on one CPU thread
Speech plays in your browser

Good to know:

• Limits: the three released voices are single-voice students for English and Italian
• License: Apache 2.0 for the project, while eSpeak NG is GPL-3.0-or-later and ONNX Runtime Web is MIT
• Cost to run: none named
• Data it reaches: the text stays in your browser, and the page loads ONNX Runtime Web from jsDelivr, a public content delivery network
• Not for: the source names no forbidden uses, and the sample sentences come from WikiText-103 and the Italian Wikipedia under CC BY-SA

On Hugging Face, 13 November 2026

• Likes: 3
• Built with: static
• License: apache-2.0

Links

• Space: https://huggingface.co/spaces/MithrilMan/teacher-in-the-loop
• Files: https://huggingface.co/spaces/MithrilMan/teacher-in-the-loop/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.