Inflect Mini v3

Inflect Mini v3

Turns English text into speech on your own computer with six preset voices, on CPU or CUDA. For developers and creators who want a small local text-to-speech model.

Made by owensong 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: Inflect Mini v3 is a small English text-to-speech model from owensong with 34.31M core parameters. It has six preset voices and speaks at a native 24 kHz, with an optional compact 48 kHz enhancement. After setup, the speech is generated on your own machine.

Who it suits: It suits people who want English speech from a compact model that runs locally, on a CPU or a CUDA GPU.

Industry | Who uses it | How they use it | Example
Software development | Developers adding spoken output to a tool | Run the speech command with a preset voice | A developer writes a WAV file for each message
Content creation | Video and audio makers | Generate narration with a voice such as Mark or Cassidy | A creator makes a voice-over draft locally
Education | Teachers and course builders | Use James or Hope for explanations and instructions | A course builder makes audio for lesson steps

Anyone who needs a language other than English can skip it, because the record lists English only.

Where to use it: It runs from a Python environment on your own computer.

• Windows, macOS or Linux: download the package with Python and the Hugging Face tools
• A CPU: the default route in the quickstart
• A CUDA GPU: a separate dependency recipe plus a device flag

Use case | Who | What they get back
Spoken output | Developers | A WAV file for each message a tool needs to say
Narration drafts | Video and story makers | A local voice-over to review
Lesson audio | Educators and course builders | Instructions and explanations read aloud

How to set it up: You need a few things first.

• The huggingface-hub package, pinned to the version in the commands below
• A CPU, or a CUDA GPU with its own requirements file
• A folder for environment files, profiles and generated audio, kept outside the download

1. Set up a clean Python 3.12 environment somewhere other than the download folder.
2. Install huggingface-hub.
3. Download the whole repository into inflect-mini-v3, not only the weight file.
4. Change into that folder and run verify_release.py.
5. Install the CPU requirements.
6. Run the speech command with a voice and your text.

python -m pip install huggingface-hub==0.36.2
hf download owensong/Inflect-Mini-v3 --local-dir inflect-mini-v3
cd inflect-mini-v3
python verify_release.py
python -m pip install -r core/requirements-cpu.txt
python run.py speech --voice cassidy --text "A small idea can become something worth hearing." --subtype PCM_16 --output ../mini-cassidy.wav

How to use it: You run the speech command with a voice name, your text and an output file. Name a new output file, save it beside or above the package rather than inside it, and make sure its folder exists. On a CUDA GPU, install from core/requirements-cuda.txt instead and add the --device cuda flag.

Title: Preset voices and the card's sample for each
Voice | Sample type
cassidy | Conversation
mark | Narration
hope | Instructions
james | Explanation
sophia | Everyday speech
jasper | No sample shown

Tool | What it does
run.py speech | Turns your text into speech with the voice you choose
verify_release.py | Checks the downloaded package before you install anything else
core/requirements-cpu.txt | Lists the dependencies for running on a CPU
core/requirements-cuda.txt | Lists the dependencies for running on a CUDA GPU

Example: You run the speech command with the voice mark and your own text, "The meeting moves to Friday at noon." You name the output file notice.wav. A native 24 kHz PCM16 file with that name is written, and your CPU generates the audio.

For newcomers and experts: This is a small program that reads English text aloud on your own computer, in one of six voices. Your first small step is to follow the quickstart and try Cassidy and Mark, the two voices the card suggests trying first.

The model has 34.31M core parameters and speaks at a native 24 kHz, with an optional compact 48 kHz enhancement. The card's voice samples are plain Mini output with no enhancement. A verify_release.py script checks the package before use. Setup may download dependencies, and speech generation stays local after that.

Pick a preset voice and write your text
run.py speech generates audio on CPU or CUDA
Native 24 kHz speech is saved as a WAV file
An optional 48 kHz enhancement can follow

Good to know:

• The card also lists a Voice Tuner for reference-based adaptation. Use it only with voices you have the right to use.
• Each sample on the card reads a different passage, so voices are not compared on the same text.
• CPU and CUDA use different dependency recipes, so install the one that matches your setup.
• License: Apache 2.0, as the Hugging Face record lists it.
• Cost to run: none named, no API key, account or paid service is mentioned.
• Data it reaches: setup may download dependencies, and it writes the audio files you name. Generation runs locally after setup.
• Not for: languages other than English, or cloning a voice you have no right to use.

On Hugging Face, 13 November 2026

• Task: text to speech
• Downloads in the last 30 days: 86
• Likes: 21
• License: apache-2.0
• Languages: en
• Tags: inflect, text-to-speech, english, multi-speaker, local-inference

Links

• Model: https://huggingface.co/owensong/Inflect-Mini-v3
• Files: https://huggingface.co/owensong/Inflect-Mini-v3/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download owensong/Inflect-Mini-v3

Then follow the usage in its model card. Nothing in this kit runs by itself.