Preserving Pipil (Nawat) with AI Speech-to-Text

Pipil, or Nāwat, is an Uto-Aztecan language indigenous to El Salvador. Once spoken widely across the region, it now survives in a few communities, primarily in the departments of Sonsonate and Ahuachapán. With fewer than 2,000 speakers, most of them elderly, the language is at a critical juncture. Efforts to revitalize it rely heavily on documenting the remaining fluent speakers. This is where accurate speech-to-text technology becomes essential.

Why Accurate Transcription Matters for Nawat

Transcribing Nawat audio is not a straightforward task. The language features phonemic vowel length, glottal stops, and a range of consonants that may be unfamiliar to outsiders. For instance, the words kisa (to leave) and kīsa (to come out) differ only in vowel duration, yet they have entirely different meanings. A transcription system that fails to capture this distinction loses crucial information. Moreover, Nawat has several dialects—Izalco, Cuisnahuat, Santo Domingo de Guzmán—each with its own pronunciation and vocabulary. A generic model trained on one dialect may produce poor results for another.

Challenges in Building a Nawat ASR Model

• Scarcity of data: High-resource languages have millions of hours of transcribed audio; Nawat has perhaps a few hundred. Training a conventional deep learning model from scratch is impossible.
• Speaker variability: Remaining speakers are elderly, and their speech may be slower, with more hesitation and code-switching with Spanish.
• Orthographic norms: While a standard orthography exists, many speakers and writers use ad-hoc spellings, leading to inconsistency.

Speechyou addresses these issues through a combination of transfer learning, user-contributed data, and dialect-specific fine-tuning. The platform allows users to upload audio along with optional text corrections, which are fed back to improve the model for everyone.

Use Cases for Nawat Transcription

The applications of reliable speech-to-text for Nawat extend across multiple domains:

• Cultural preservation: Transcribing oral histories, folk tales, and songs ensures that the language remains accessible even as fluent speakers decline.
• Educational resources: Teachers can convert spoken lessons into written texts, creating bilingual materials for language classes.
• Media and content: Indigenous radio stations can generate transcripts for archives, and filmmakers can add subtitles to documentaries about Pipil culture.
• Linguistic research: Researchers can automate the tedious process of transcribing field recordings, freeing time for analysis.
• Community archives: Local communities can build digital libraries of their heritage, searchable by keywords.

How Speechyou Helps

Speechyou is designed to support languages that are often overlooked by big tech companies. Our Nawat model is trained on a specially curated dataset of recorded conversations, stories, and speeches. When you upload a Nawat audio file, the system processes it and returns a text transcription that can be exported as plain text, SRT, or VTT subtitles. You can then edit the transcript directly in our web editor to correct any errors, and those corrections help refine the model.

For communities, we offer free transcription credits for nonprofit language documentation projects. The unlimited transcription plan included in the Solo subscription makes it affordable for individual researchers and small organizations.

Join the Effort to Save Nawat

Technology alone cannot save a language, but it can be a powerful ally. By providing accurate, easy-to-use speech-to-text for Nawat, Speechyou empowers speakers and advocates to document and share their mother tongue. Whether you are a linguist collecting data, a teacher preparing a lesson, or a community member hoping to preserve your grandmother's stories, Speechyou gives you the tools to turn spoken word into written legacy.

Start transcribing Nawat audio today and help keep the language alive for future generations.