SYSPIN Telugu Male TTS Dataset

SYSPIN Telugu Male TTS Dataset — Technical Datasheet

Version: S1.0
Released by: SPIRE Lab, Indian Institute of Science (IISc), Bengaluru, India
Dataset URL: https://syspin.iisc.ac.in/datasets/telugu%20male%20tts%20data
Download Portal: https://spiredatasets.ee.iisc.ac.in/syspincorpus
License: Creative Commons Attribution 4.0 International (CC BY 4.0)
Citation: Abhayjeet et al., "SYSPIN_S1.0 Corpus — A TTS Corpus of 900+ hours in nine Indian Languages", 2025

Overview

The SYSPIN Telugu Male TTS Dataset is a single-speaker, studio-recorded, read-speech corpus for Text-to-Speech (TTS) synthesis in the Telugu language, spoken by a professional male voice artist. It is one of eighteen speaker-language datasets produced under the SYSPIN (SYnthesizing SPeech in INdian languages) initiative — a large-scale, open-source effort by SPIRE Lab at IISc Bengaluru to build high-quality TTS corpora across nine Indian languages: Bengali, Bhojpuri, Chhattisgarhi, Hindi, Kannada, Magahi, Maithili, Marathi, and Telugu

Project Context

Dataset Summary

Audio Specifications

Note: Downstream users may resample to 16 kHz or 22.05 kHz as required by their synthesis frameworks (e.g., ESPnet, VITS, FastPitch). The original 48 kHz / 24-bit files are the canonical release format.

Text & Transcription Specifications

Intended Use Cases

• Neural TTS training: End-to-end models (VITS, NaturalSpeech, YourTTS), autoregressive models (Tacotron 2, FastSpeech 2), and codec-based systems (EnCodec, SoundStorm)

• Vocoder training/evaluation: HiFi-GAN, WaveGlow, BigVGAN

• Multi-speaker and multi-lingual TTS: Alongside other SYSPIN speaker datasets for cross-lingual voice transfer

• Voice cloning research: Base corpus for speaker-adaptation experiments

• TTS evaluation / MOS benchmarking: Ground-truth natural speech reference for Mean Opinion Score studies

• Acoustic model research: Prosody modeling, duration modeling, pitch analysis for Telugu

• Language/speech technology accessibility: Downstream applications in agriculture advisory services, financial literacy, e-governance, and healthcare for Telugu-speaking populations

Known Limitations & Quality Notes

• The dataset represents a single speaker and does not capture speaker diversity, dialectal variation, or prosodic style variation across the Telugu-speaking population.

• Speech rate is controlled and uniform, consistent with professional TTS recording standards; spontaneous or conversational speech characteristics are absent.

• Domain-specific vocabulary (agricultural and financial terminology) may affect model generalization to other domains without fine-tuning.

Corpus Structure

telugu_male_tts/
├── wav/                    # Audio files (WAV, 48 kHz, 24-bit, mono)
│   ├── ....wav
│   └── ...
├── ..._Transcripts.json            # transcripts

the Transcripts.json file contains a "Transcripts" key that looks as follows:

    "Transcripts": {
        "IISc_SYSPINProject_te_m_GENE_02044": {
            "Transcript": "ఇంపెడెన్స్ విశ్లేషణ ఉపయోగించి బయోమెట్రిక్ సమాచారాన్ని కొలిచే స్మార్ట్ స్కేల్స్ అడుగుల ద్వారా తేలికపాటి విద్యుత్ ప్రేరణలను పంపుతాయి. ",
            "Domain": "GENERAL"
        },
        "IISc_SYSPINProject_te_m_EDUC_02060": {
            "Transcript": "\"అనుష్కతో కలిసి మరీన్ పరేడ్ బీచ్లో బెంచ్పై కూర్చొని ఆస్వాదించిన క్షణాలు జీవితాంతం గుర్తుండి పోతాయి\" అని బిసిసిఐ టివికి ఇచ్చిన ఇంటర్వ్యూలో కోహ్లి తెలిపాడు.",
            "Domain": "EDUCATION"
        },
        ...

License & Attribution

This dataset is released under the Creative Commons Attribution 4.0 International License (CC BY 4.0). Users are free to share and adapt the material for any purpose, including commercial use, provided appropriate credit is given.

Required Citation

Abhayjeet et al., "SYSPIN_S1.0 Corpus — A TTS Corpus of 900+ hours
in nine Indian Languages", 2025.