Parecís Speech to Text: Preserving a Language with AI

Parecís (also known as Paresi or Haliti) is an Arawakan language spoken by the Pareci people in the Brazilian state of Mato Grosso. With fewer than 2,000 speakers, it is classified as endangered. Language preservation efforts rely heavily on recording and transcribing spoken narratives, songs, and daily conversations. Until recently, speech-to-text technology for Parecís was virtually nonexistent.

Why Accurate Speech-to-Text Matters for Parecís

Accurate transcription of Parecís audio enables several critical activities:

• Documenting oral traditions: Elders hold vast knowledge of myths, medicinal plants, and history. Written records ensure this knowledge is not lost.
• Creating educational materials: Bilingual textbooks and digital resources help children learn both Parecís and Portuguese.
• Generating subtitles: Videos in Parecís can reach a wider audience, including the deaf community.
• Linguistic research: Phonetic and morphological analysis requires clean, time-aligned transcriptions.

Transcription Challenges Unique to Parecís

Parecís presents several hurdles for automatic speech recognition:

Tonal and Phonetic Complexity

Parecís uses pitch accent to distinguish words. For example, the word for “house” and “to sleep” differ only in tone. Most ASR models are not designed to handle such tonal contrasts. Speechyou addresses this by incorporating pitch features into its acoustic model.

Limited Data

With only a few thousand speakers, publicly available audio corpora are tiny. Speechyou uses transfer learning from related Arawakan languages and synthetic data generation to augment the training set.

Code-Switching

In daily life, Parecís speakers frequently borrow Portuguese terms, especially for modern concepts. The model must be bilingual to avoid mislabeling Portuguese words as errors. Speechyou’s approach uses a shared vocabulary and language ID to produce accurate mixed-language transcriptions.

Use Cases in Practice

• Oral History Archives: The Pareci Indigenous Association has used Speechyou to transcribe hours of elder interviews, creating a searchable digital archive.
• School Lessons: Teachers upload audio of storytelling sessions and receive text versions to create reading materials for students.
• Subtitled Documentaries: A short film about the Pareci harvest festival was subtitled in Parecís and Portuguese using Speechyou, making it accessible to both communities.
• Community Meetings: Minutes of council meetings are now transcribed in Parecís, ensuring that decisions are recorded in the native language.

How Speechyou Helps

Speechyou offers a dedicated Parecís speech-to-text model accessible via a simple web interface. Users can upload audio or video files and receive transcripts in minutes. The output supports SRT and VTT subtitle formats, allowing easy integration with video players. The Solo plan includes unlimited transcription, making it affordable for community projects.

For Parecís, a language with few digital resources, Speechyou provides a bridge between oral tradition and written record. By enabling accurate, automated transcription, it empowers speakers to preserve their heritage and share it with the world.