Torwali Speech to Text: Preserving a Language Through AI

Torwali is a Dardic language spoken by around 120,000 people in the Swat Valley of northern Pakistan. It belongs to the Indo-Aryan language family and is written in a modified Arabic script, which includes additional letters for sounds like the retroflex ݜ (sh) and ݙ (d). Despite its small speaker population, Torwali has a rich oral tradition of folk tales, poetry, and songs. However, like many minority languages, it is endangered due to the dominance of Urdu and Pashto in education and media.

Why Accurate Speech to Text for Torwali Matters

Accurate speech-to-text technology can play a crucial role in language preservation. By transcribing spoken Torwali into written form, we create a permanent record that can be used for education, research, and cultural transmission. For example:

• Oral history projects can record elders' stories and convert them into searchable text.
• Schools can produce Torwali-language textbooks and reading materials.
• Community media can add subtitles to videos, making them accessible to a wider audience.

Without reliable transcription tools, much of this knowledge remains locked in audio files that are difficult to index or share.

Challenges in Transcribing Torwali

Torwali presents several unique challenges for automatic speech recognition:

1. Retroflex consonants: Sounds like /ʈ/ and /ɖ/ are produced with the tongue curled back, and they are not found in many major languages. Generic ASR models often confuse them with dental or alveolar stops.
2. Tonal system: Torwali uses pitch to distinguish words. For example, the word for "horse" and "laughter" may differ only by tone. Accurate transcription requires tonal awareness.
3. Limited training data: With only about 120,000 speakers, there are few publicly available transcribed datasets. Speechyou overcomes this through transfer learning and data augmentation.
4. Arabic script adaptation: The script includes characters like ݜ (for /ʂ/) and ٿ (for /tʰ/) that are not part of standard Arabic. The model must be trained on these specific glyphs.

Use Cases for Torwali Transcription

• Podcast transcription: Torwali podcasters can convert their episodes into text for blog posts, show notes, and social media.
• Subtitle generation: Video creators can add Torwali subtitles to documentaries, interviews, or entertainment content, improving accessibility.
• Research documentation: Linguists and anthropologists can transcribe field recordings quickly, enabling detailed analysis of grammar and vocabulary.
• Accessibility: Deaf or hard-of-hearing community members can read subtitles in their native language, fostering inclusion.
• Language learning: Students can read along with audio recordings to improve literacy in Torwali.

How Speechyou Helps

Speechyou is designed to handle the specific needs of low-resource languages like Torwali. Our AI model is trained on Torwali audio with accurate transcriptions, including the extended Arabic script. Key features include:

• High accuracy: Over 95% word accuracy on clear audio, even with retroflex and tonal distinctions.
• No per-minute fees: Unlimited transcription with the Solo plan, ideal for long recordings.
• Multi-format export: Download subtitles as SRT or VTT files for use in video editors.
• Multi-language support: Transcribe in Torwali and generate subtitles in 100+ other languages.

Whether you are a linguist documenting an endangered language, a teacher creating classroom materials, or a community member sharing stories online, Speechyou provides the tools you need to turn spoken Torwali into written text. Start transcribing today and help preserve the Torwali language for future generations.