# SapinSapin AI — Open Speech and Language Foundations for 10+ Philippine Languages

> SapinSapin AI builds open-source speech recognition, text-to-speech and language models for Filipino, Cebuano, Ilocano, Hiligaynon, Bikol, Pangasinan, Kapampangan, Tausug and Waray — and runs its own multilingual community agent on NYO.

**Source page:** https://philippineaireport.com/case-studies/sapinsapin
**Canonical partner source:** https://www.sapinsapin.ai

Category: Language AI / Open Source · 2026 · 48 public models · 500+ hours of speech data

## Problem

Philippine languages are underrepresented in AI training data. Commercial speech and language APIs cover Filipino at best and skip Cebuano, Ilocano, Hiligaynon and the rest, so Filipino developers and researchers have no open, documented foundation to build on in the languages most Filipinos actually speak.

## Solution

[SapinSapin AI](https://www.sapinsapin.ai) publishes datasets, models and tools freely, under open licences (MIT and UP-DSP research licences are used).

Four major datasets:

- Philippine Language Dataset — 334,268 utterances, 448+ hours of prompted speech across 10 languages
- Filipino Speech Corpus — 313,322 transcribed segments, MIT-licensed
- BantayWika — 6.09M tokens
- halohalo — 45.5K cleaned web rows

48 public models spanning speech recognition (Whisper fine-tunes), text-to-speech (SpeechT5 fine-tunes), voice conversion and text generation, with a live demo space on [Hugging Face](https://huggingface.co/sapinsapin).

> "Every voice belongs in the future. Open foundations for Philippine AI."

## Sappy: the community agent

The project's community manager, Sappy, answers visitors' questions about SapinSapin in Tagalog, Cebuano, Ilocano and Hiligaynon. It runs as a Cloudflare Worker: Ask → look it up in the project's published datasets and model cards → answer in the visitor's language, using a multilingual open-weight model through NYO's OpenAI-compatible endpoint with prepaid credits and purpose-specific guardrails. Project lead and AI systems engineer: [Marc Ocampo](https://marcocampo.com).

- How Sappy runs on NYO: https://llm.nyolab.ai/case-studies/sapinsapin
- NYO case study: https://philippineaireport.com/case-studies/nyolab

## Impact

- 500+ documented hours of Philippine-language speech and 25M+ pretraining tokens, openly licensed
- 48 public models for ASR, TTS, voice conversion and text generation
- Nine Philippine languages plus English, including languages no commercial API serves
- A volunteer project answers its own community in four Philippine languages without a support team

## Philippine Relevance

The report's language findings meet practice here: see [Filipino large language models](https://philippineaireport.com/llms) and [Philippine datasets](https://philippineaireport.com/datasets). Open speech data is the missing layer under every Filipino-language voice assistant, call-centre tool and accessibility product.

## Links

- Explore SapinSapin AI: https://www.sapinsapin.ai
- Models and datasets on Hugging Face: https://huggingface.co/sapinsapin
- Sappy on NYO: https://llm.nyolab.ai/case-studies/sapinsapin