Pez Speech to Text: Transcribe and Preserve a Torricelli Language

Pez is a Torricelli language spoken by approximately 2,000 people in Sandaun Province, Papua New Guinea. It is an oral language with a growing written tradition using a Latin-based orthography. As globalization and language shift threaten its survival, tools that can convert spoken Pez into written text are invaluable for documentation, education, and cultural preservation.

Where Pez Is Spoken

Pez speakers live in villages along the northern coast of Papua New Guinea and inland areas near the border with Papua. The language is part of the Torricelli family, which includes several other small languages in the region. Most speakers are bilingual in Tok Pisin, and many also know English. The Pez language has several dialects, including Coastal Pez, Inland Pez, and Western Pez, each with distinct phonetic and lexical features.

Why Accurate Speech-to-Text for Pez Matters

• Language documentation: Linguists can transcribe field recordings without manual effort.
• Education: Create reading materials for bilingual schools.
• Accessibility: Generate subtitles for videos so deaf or hard-of-hearing community members can follow content.
• Oral history preservation: Record and transcribe elder knowledge before it is lost.
• Community media: Produce subtitled videos in Pez for social media and local broadcasting.

Challenges in Transcribing Pez

Pez presents several challenges for automatic speech recognition:

• Limited data: Very few transcribed audio recordings exist, making it hard to train models from scratch.
• Complex phonology: Nasalized vowels and prenasalized stops are not common in most ASR training data.
• Morphological richness: Verbs can have many affixes, and the model must learn to segment them correctly.
• Dialect variation: Pronunciation differences between dialects can reduce accuracy if not accounted for.

Speechyou overcomes these challenges by using transfer learning from related languages and allowing users to upload custom audio for fine-tuning. The model is designed to handle the specific sound inventory of Pez and can adapt to different dialects with additional training data.

Use Cases for Pez Transcription

Oral History Preservation

Elders in Pez communities hold vast knowledge of traditional medicine, genealogy, and folklore. By recording and transcribing these oral histories, communities can create searchable archives. Speechyou's batch transcription feature makes it possible to process hours of audio quickly.

Language Learning and Literacy

Bilingual education programs need written materials in Pez. Teachers can record themselves speaking Pez and use Speechyou to generate texts for worksheets, storybooks, and flashcards. This supports mother-tongue literacy and helps children transition to reading in Tok Pisin and English.

Church and Community Events

Many Pez speakers are Christian, and church services often include Pez-language sermons and songs. Transcribing these allows for printed bulletins and subtitled video recordings that can be shared with diaspora communities.

Research and Documentation

Linguists working on Torricelli languages can use Speechyou to transcribe interviews and field notes. The platform's export options (plain text, SRT, VTT) integrate with common linguistic software like ELAN and FLEx.

How Speechyou Helps

Speechyou provides a dedicated Pez speech-to-text model that is continuously improved through user contributions. The platform offers:

• High accuracy after fine-tuning with just 30 minutes of Pez audio.
• Support for multiple dialects through custom training.
• Subtitle generation in SRT and VTT formats.
• Unlimited transcription on the Solo plan, making it affordable for community projects.

By making Pez transcription accessible, Speechyou empowers speakers to document their language, share their stories, and ensure that Pez remains a living language for generations to come.