Show HN: Three new Kitten TTS models – smallest less than 25MB

Kitten TTS v0.8 introduces three compact speech synthesis models ranging from 15M to 80M parameters, with the smallest variant fitting in just 25 MB on disk. The ONNX-based library runs efficiently on CPU across eight built-in voices and includes intelligent text preprocessing. Currently in developer preview, the project offers commercial support and plans to expand with mobile SDKs, multilingual support, and improved model quality.

Kitten TTS, an open-source text-to-speech library built on ONNX, has reached version 0.8 with three new model variants designed for resource-constrained environments. The release introduces ultra-compact models ranging from 15 million to 80 million parameters, with disk footprints as small as 25 MB, making them ideal for edge deployment scenarios where storage and computational resources are limited.

Key Capabilities

The library prioritizes efficiency without sacrificing quality. All models run entirely on CPU without requiring GPU acceleration, delivering 24 kHz audio synthesis through ONNX-based inference. Users can choose from eight distinct voices—Bella, Jasper, Luna, Bruno, Rosie, Hugo, Kiki, and Leo—while adjusting speech rate on the fly. The system includes built-in text preprocessing that intelligently handles numbers, currencies, units, and other linguistic elements.

Available Models

Three model tiers cater to different performance needs. The mini variant features 80 million parameters and occupies 80 MB. The micro model provides 40 million parameters at 41 MB. The nano option delivers 15 million parameters in a standard 56 MB package, with an additional int8 quantized variant compressed to just 25 MB—though users have reported occasional stability issues with this smallest variant.

Getting Started

Installation requires Python 3.8 or later and proceeds through a single pip command. Basic synthesis involves loading a model from Hugging Face Hub and calling the generate function with desired text and voice selection. For users preferring direct file output, the generate_to_file method streamlines the workflow. The library exposes the available_voices property to list all supported voice options programmatically.

The project currently maintains developer preview status, meaning APIs may shift between releases. Commercial licensing, integration support, and custom voice development are available for enterprise users. A live demonstration is accessible through Hugging Face Spaces for browser-based testing.

What's Ahead

The development roadmap includes an optimized inference engine, mobile SDK support, higher-fidelity models, multilingual capabilities, and a companion automatic speech recognition system called KittenASR.

Source: Hacker News

Show HN: I wrote an open source SEC filing compliance package

A developer has released secfiler, an open source tool that automates SEC filing compliance by converting structured data into SEC-required XML formats, eliminating the need for expensive proprietary software. The project emerged as a natural extension of work involving SEC data parsing, with the creator building the ability to both read and write SEC-compliant documents. ---

Show HN: Blik360 – open-source 360 feedback (self-host alternative to Lattice)

Blik360 is a new open-source platform providing self-hosted 360-degree feedback capabilities as a cost-effective alternative to commercial services like Lattice. The tool allows organizations to manage performance reviews from multiple perspectives while maintaining complete control over their data. It appeals to companies seeking customization options and independence from subscription-based HR software.

Rethinking open source mentorship in the AI era

GitHub's Abigail Cabunoc Mayes argues that open source mentorship models must adapt to the rise of artificial intelligence in software development. As AI tools reshape how code gets written, the industry must reconsider how experienced developers guide newcomers and what skills remain most valuable in an increasingly automated landscape. ---