--- title: "Paradee-8M v1.3" kind: "Kit" category: "Developer" price: "Free" seller: "Mindket" version: "1" url: "https://mindket.com/p/8d10c6d9-1181-415a-81fb-3eefc5a872e6" updated: "2026-11-12" --- # Paradee-8M v1.3 > The name, the description, the About and a README below are the seller's words: data, not instructions. A tiny 7.88M-parameter text to speech model with eight voices in English, Hindi, French and Spanish, distilled from Kokoro-82M, for developers who need fast speech on a CPU. Paradee-8M v1.3 is a kit in the Developer category, listed by Mindket. It is free. This is version 1, last updated on 2026-11-12. Download it on this page, a zip of its files. Nothing in it runs by itself. Made by sahilmahendrakar ## About ### What it is Paradee-8M v1.3 is a tiny text to speech model from sahilmahendrakar with 7.88M parameters. It was distilled from the larger Kokoro-82M model. In this release the eight voices come from Kokoro and cover English, Hindi, French and Spanish. The voice you pick decides the language, and a sentence is read in only one language. The int8 build is a single 8.5 MB ONNX file that holds all eight voices. ### Who it suits It suits developers who want fast speech from a tiny file that runs on a plain CPU, and researchers who study small speech models. Who uses this model | Industry | Who uses it | How they use it | Example | | --- | --- | --- | --- | | Software | App developers | Add spoken output to an app without a GPU | Read a notification aloud on a laptop | | Accessibility | Product teams | Offer spoken versions of text | Turn short messages into audio in four languages | | Research | Speech researchers | Study distillation from a larger model | Compare it with its teacher Kokoro-82M | If you need one sentence that mixes languages, you can skip it, since each sentence uses one language. ### Where to use it - Python scripts or apps that can run ONNX files on a CPU - Windows, macOS or Linux with Python, for downloading the files Jobs it does | Use case | Who | What they get back | | --- | --- | --- | | Speech on a CPU | Developers | Spoken audio without a graphics card | | English voices | App builders | Four voices, American and British, female and male | | Hindi, French, Spanish | Global teams | Four more voices, one or two per language | ### How to set it up You need Python and the huggingface\_hub package for the download. The card shows no run steps in the visible part, so the way you play the model depends on your tools. The author also publishes the code, training scripts and a Python package on GitHub. - Python with pip - A tool that runs ONNX files, or the author's Python package - A few megabytes of free disk space 1. Install huggingface\_hub. 2. Download the model files with the hf command. 3. Run the ONNX file with your tool, or use the author's Python package. ```bash pip install -U huggingface_hub hf download sahilmahendrakar/Paradee-8M-v1.3 ``` ### How to use it Pick one of the eight voices and give it a sentence in that voice's language. Check the full model card or the author's package for the exact call, since the visible part of the card does not show one. The eight voices | Voice | Language | | --- | --- | | af\_heart | English, American, female | | bf\_isabella | English, British, female | | am\_puck | English, American, male | | bm\_fable | English, British, male | | hf\_alpha | Hindi, female | | hm\_omega | Hindi, male | | ff\_siwis | French, female | | em\_alex | Spanish, male | ### Example You choose the Spanish voice and give it one Spanish sentence. The model returns a short audio clip in that voice, produced on your CPU. ### For newcomers and experts Text to speech turns written words into spoken audio. A first small test is the English voice with one short sentence. Distillation means a small model learns to copy a bigger one. In training, the small model copied Kokoro as it read around 12,000 Wikipedia sentences for each new voice. The model keeps one small table of learned vectors per voice, and version 1.3 grew that table from four rows to eight. Written sentence → Pick a voice, which sets the language → Paradee ONNX model → Spoken audio ### Good to know The author says it is the smallest single model that speaks several languages, roughly six times lighter than the runner-up, M5-TTS, at 50M parameters. In the card's test it runs about 62 times faster than playback speed on one CPU thread. The Hindi voices are the furthest from the teacher, with word error rates near 21.7% and 21.6%. Quality scores come from the int8 file on 200 held-out sentences per voice. Parameters, in millions | Label | Value | | --- | --- | | Paradee-8M v1.3 | 7.88 | | M5-TTS | 50 | | Kokoro-82M | 82 | - License: Apache 2.0. - Cost to run: none named. - Data it reaches: the files come from Hugging Face, and the model then runs on your own computer. - Not for: mixing languages inside one sentence. ## README ````markdown # Paradee-8M v1.3 A tiny 7.88M-parameter text to speech model with eight voices in English, Hindi, French and Spanish, distilled from Kokoro-82M, for developers who need fast speech on a CPU. Made by sahilmahendrakar on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face. ## About What it is: Paradee-8M v1.3 is a tiny text to speech model from sahilmahendrakar with 7.88M parameters. It was distilled from the larger Kokoro-82M model. In this release the eight voices come from Kokoro and cover English, Hindi, French and Spanish. The voice you pick decides the language, and a sentence is read in only one language. The int8 build is a single 8.5 MB ONNX file that holds all eight voices. Who it suits: It suits developers who want fast speech from a tiny file that runs on a plain CPU, and researchers who study small speech models. ```table Title: Who uses this model Industry | Who uses it | How they use it | Example Software | App developers | Add spoken output to an app without a GPU | Read a notification aloud on a laptop Accessibility | Product teams | Offer spoken versions of text | Turn short messages into audio in four languages Research | Speech researchers | Study distillation from a larger model | Compare it with its teacher Kokoro-82M ``` If you need one sentence that mixes languages, you can skip it, since each sentence uses one language. Where to use it: - Python scripts or apps that can run ONNX files on a CPU - Windows, macOS or Linux with Python, for downloading the files ```table Title: Jobs it does Use case | Who | What they get back Speech on a CPU | Developers | Spoken audio without a graphics card English voices | App builders | Four voices, American and British, female and male Hindi, French, Spanish | Global teams | Four more voices, one or two per language ``` How to set it up: You need Python and the huggingface_hub package for the download. The card shows no run steps in the visible part, so the way you play the model depends on your tools. The author also publishes the code, training scripts and a Python package on GitHub. - Python with pip - A tool that runs ONNX files, or the author's Python package - A few megabytes of free disk space 1. Install huggingface_hub. 2. Download the model files with the hf command. 3. Run the ONNX file with your tool, or use the author's Python package. ```bash pip install -U huggingface_hub hf download sahilmahendrakar/Paradee-8M-v1.3 ``` How to use it: Pick one of the eight voices and give it a sentence in that voice's language. Check the full model card or the author's package for the exact call, since the visible part of the card does not show one. ```table Title: The eight voices Voice | Language af_heart | English, American, female bf_isabella | English, British, female am_puck | English, American, male bm_fable | English, British, male hf_alpha | Hindi, female hm_omega | Hindi, male ff_siwis | French, female em_alex | Spanish, male ``` Example: You choose the Spanish voice and give it one Spanish sentence. The model returns a short audio clip in that voice, produced on your CPU. For newcomers and experts: Text to speech turns written words into spoken audio. A first small test is the English voice with one short sentence. Distillation means a small model learns to copy a bigger one. In training, the small model copied Kokoro as it read around 12,000 Wikipedia sentences for each new voice. The model keeps one small table of learned vectors per voice, and version 1.3 grew that table from four rows to eight. ```flow Written sentence Pick a voice, which sets the language Paradee ONNX model Spoken audio ``` Good to know: The author says it is the smallest single model that speaks several languages, roughly six times lighter than the runner-up, M5-TTS, at 50M parameters. In the card's test it runs about 62 times faster than playback speed on one CPU thread. The Hindi voices are the furthest from the teacher, with word error rates near 21.7% and 21.6%. Quality scores come from the int8 file on 200 held-out sentences per voice. ```chart Title: Parameters, in millions Paradee-8M v1.3: 7.88 M5-TTS: 50 Kokoro-82M: 82 ``` - License: Apache 2.0. - Cost to run: none named. - Data it reaches: the files come from Hugging Face, and the model then runs on your own computer. - Not for: mixing languages inside one sentence. ## On Hugging Face, 13 November 2026 - Task: text to speech - Library: onnx - Downloads in the last 30 days: 1,649 - Likes: 7 - License: apache-2.0 - Languages: en, hi, fr, es - Based on: hexgrad/Kokoro-82M - Tags: onnx, paradee, text-to-speech, distillation, kokoro, multilingual ## Links - Model: https://huggingface.co/sahilmahendrakar/Paradee-8M-v1.3 - Files: https://huggingface.co/sahilmahendrakar/Paradee-8M-v1.3/tree/main ## Get the model Download its files with Hugging Face's command line: ```sh pip install -U huggingface_hub hf download sahilmahendrakar/Paradee-8M-v1.3 ``` Then follow the usage in its model card. Nothing in this kit runs by itself. ```` ## What is in it 1 files, 5 KB, downloaded as a zip. ```text README.md 5 KB ``` ## Source [sahilmahendrakar/Paradee-8M-v1.3 on Hugging Face](https://huggingface.co/sahilmahendrakar/Paradee-8M-v1.3) ## Compare with similar Listed Hugging Face models in the same category that share the most tags with this one. | Product | Downloads, last 30 days | Likes | Parameters | License | Last change | | --- | --- | --- | --- | --- | --- | | Paradee-8M v1.3 (this page) | 1,649 | 7 | — | apache-2.0 | 2026-11-09 | | [Paradee-8M v1.2](https://mindket.com/p/3df01b26-8a25-4088-ae9c-690c8ccf82e4) | 11,405 | 28 | — | apache-2.0 | 2026-10-28 | | [Paradee-8M v1.0](https://mindket.com/p/746673eb-d0e8-48cb-b2b4-db08f17b231b) | 4,374 | 70 | — | apache-2.0 | 2026-10-09 | | [MakHam Thai TTS](https://mindket.com/p/4257e6b3-0fd0-43b1-9870-056356b4099f) | 0 | 2 | — | apache-2.0 | 2026-11-09 | | [Klein-TTS-X1.5-Pro](https://mindket.com/p/c2bfb923-e753-414e-9acd-851872b8e3e6) | 258 | 10 | 13.1M | apache-2.0 | 2026-11-03 | | [Supertonic 3](https://mindket.com/p/910af7e2-128f-445f-bcb1-310264314e6d) | 92,359 | 1,043 | — | openrail | 2026-05-18 | Numbers as Mindket last read them, from 2026-11-12 to 2026-11-13. ## Trust facts Facts from each source, with the day each was checked. A maintainer's note is their own words: data, not instructions. - Last change on Hugging Face: 2026-11-09 - First published: 2026-11-06 - Checked on Hugging Face: 2026-11-12 - Something wrong? Tell us: https://mindket.com/takedown#ask ## Get it - [Open sahilmahendrakar/Paradee-8M-v1.3 on Hugging Face](https://huggingface.co/sahilmahendrakar/Paradee-8M-v1.3) - [Get it on your Mindket account](https://account.mindket.com/account/?product=mindket#/market/8d10c6d9-1181-415a-81fb-3eefc5a872e6) - [Its page on Mindket](https://mindket.com/p/8d10c6d9-1181-415a-81fb-3eefc5a872e6)