Shor Speech to Text: Preserving a Siberian Turkic Language with AI

The Shor Language and Its People

Shor (Шор тили) is a Turkic language spoken by the Shor people, an indigenous group living in the mountainous regions of the Kemerovo Oblast in Russia, particularly along the Mras, Kondoma, and Tom rivers. The language is part of the Siberian Turkic subgroup and is closely related to Khakas and Altai. With fewer than 10,000 fluent speakers, Shor is classified as an endangered language by UNESCO. The Shor people have a rich oral tradition, including epic tales, songs, and shamanic rituals, much of which has been passed down through generations without being written down.

Why Accurate Speech-to-Text Matters for Shor

In the digital age, the survival of a language often depends on its presence in technology. Shor speakers face a digital divide: there are few tools that support their language. Accurate speech-to-text technology can help bridge this gap by enabling:

• Transcription of oral history recordings for archives and research.
• Subtitle generation for Shor-language videos, documentaries, and educational content.
• Written communication for Shor speakers who are more comfortable speaking than typing.
• Language learning by providing accurate text versions of spoken materials.

Without a dedicated speech-to-text system, these tasks are slow and expensive, often requiring human translators. Speechyou offers a reliable, AI-powered solution that understands the unique features of Shor.

Specific Challenges in Transcribing Shor

Shor presents several challenges for automatic speech recognition:

• Low resource availability: There is very little transcribed audio data for training ASR models. Many large tech companies ignore Shor because it is not commercially viable. Speechyou has invested in building a custom model for Shor despite this scarcity.
• Vowel harmony: Shor has a strict vowel harmony system where suffixes change their vowels based on the root. For example, the plural suffix can be -лар, -лер, -нар, -нер, etc. The ASR must correctly identify the root vowel and generate the appropriate suffix.
• Consonant clusters: The language allows complex clusters such as /rt/, /ks/, and /st/, which can be misheard by standard models. Speechyou's acoustic model is trained on Shor phonetics to handle these.
• Dialectal variation: The two main dialects, Mras and Kondoma, differ in pronunciation and vocabulary. A model trained on one dialect may perform poorly on the other. Speechyou offers dialect-specific profiles to address this.

Use Cases in Depth

Preserving Oral Traditions

Shor elders have a wealth of knowledge about the language's history, folklore, and traditional practices. Recording their voices and converting them to text ensures that this knowledge is not lost. Researchers can transcribe hours of interviews quickly, and communities can publish these texts in books or online.

Subtitle Creation for Media

Shor-language content is growing on platforms like YouTube. Creators can now generate SRT and VTT subtitles automatically, making their videos accessible to a broader audience, including younger Shor speakers who may be learning the language. Subtitles also help in language revitalization by providing reading material alongside spoken content.

Academic Research

Linguists studying Shor's syntax, phonology, and sociolinguistics can use Speechyou to transcribe field recordings. The ability to correct transcriptions and export them in plain text or subtitle formats streamlines the research workflow.

Accessibility and Inclusion

Deaf or hard-of-hearing members of the Shor community can benefit from captions on Shor-language media. Speechyou's real-time transcription feature can also be used at community events or meetings to provide live captions.

How Speechyou Helps

Speechyou is built to support low-resource languages like Shor. Our system uses a combination of acoustic models trained on Shor audio and language models that incorporate the Cyrillic script and vowel harmony rules. Users can:

• Upload audio or video files in any format.
• Choose between Mras and Kondoma dialect profiles.
• Generate transcripts in standard Shor Cyrillic.
• Export subtitles as SRT or VTT files.
• Edit and correct transcriptions using a simple interface.

Our commitment to over 100 languages includes Shor, ensuring that even endangered languages have a voice in the digital world. Whether you are a linguist, a community activist, or a content creator, Speechyou provides the tools you need to preserve and promote the Shor language.