🚧 Tagalog Health Domain TTS 25 Hours Single Speaker (Male) 🚧

Tagalog TTS Health Domain Dataset (Male speaker)

This dataset contains 25 hours of studio-quality recordings of deliberate, read speech from texts in the health domain.

Audio Information

• File type: WAV
• Bit depth: 16 bit
• Sample rate: 48000 Hz
• Channels: 1
• Background noise floor: -75.3 dBFS (10th percentile; unfiltered pooled mean would read -39.4 dBFS, inflated by breaths/mouth noise)
• Signal-to-noise ratio: mean 58.8 dB, median 58.7 dB (range 55.1 to 65.4; each clip's SNR computed against the shared dataset noise floor above)

Text Selection Criteria

The texts are from 118 health articles from Wikipedia (108) and other public domain sources such as the FDA.

Texts were selected according to the following criteria:

• Common diseases in the Philippines: Texts cover the top common diseases in the Philippines, such as asthma, dengue, tuberculosis, HIV/AIDS, and heart failure. Since these texts were sourced from Wikipedia, they were edited and updated for correct grammar and spelling. They also include additional context such as the history of the disease, symptoms, transmission, treatment, and prevention.
• Broad range of medical concerns: An effort was made to include a wide range of medical topics beyond the top diseases, including the common cold, flu, pregnancy, hypertension, the lymphatic system, and different types of cancer, among others.
• Taboo topics: Topics considered taboo in the Philippines — such as mental health (anxiety, depression, bipolar disorder) and reproductive health (male and female condoms, hormone replacement therapy) — were deliberately included. These were chosen to ensure the TTS dataset captures relevant Tagalog terminology and the Filipino-English accent, which might otherwise be underrepresented.
• Medline texts: Medline texts are sourced from posters and are more instructional in nature (e.g., "you should consult a doctor if…"). They also help dispel common taboo misconceptions (e.g., "HIV cannot be transmitted through a hug").
• Institutional sources: Additional texts were drawn from nursing, general health, and Red Cross materials.

Known Gaps Alternative medicine: The dataset does not include texts on alternative medicine, albularyos (traditional folk healers), or Traditional Chinese Medicine — practices that are culturally significant and commonly consulted in the Philippines, in part due to gaps in the formal healthcare system. Vaccine names: Vaccine names were deliberately avoided to prevent copyright issues.

Dataset structure

• mapping.tsv: TSV file with columns audio_filename, key, sentence, source_url
• clips/: WAV files