Hindi Text-to-Speech (TTS) Dataset

Overview

This comprehensive dataset contains over 23,700 high-quality Hindi audio samples paired with their corresponding text transcriptions, specifically designed for Text-to-Speech (TTS) synthesis and speech processing research. The dataset provides a robust foundation for developing and training Hindi language speech synthesis models.

Dataset Contents

• Audio Files: 23.7k+ WAV format audio recordings
• Metadata: CSV file containing text transcriptions and audio file mappings
• Total Size: 9.12 GB of audio data
• Language: Hindi (हिन्दी)

Use Cases

• Text-to-Speech Systems: Train neural TTS models for Hindi language
• Speech Synthesis Research: Develop and evaluate speech generation algorithms
• Voice Cloning: Create synthetic Hindi voices
• Prosody Modeling: Study intonation and rhythm patterns in Hindi speech
• Academic Research: Support linguistic and phonetic studies of Hindi

Data Structure

The dataset is organized with audio files in the audio/ directory and corresponding metadata in metadata.csv, making it easy to integrate into your ML pipeline.

Applications

Ideal for researchers, developers, and organizations working on: - Voice assistants in Hindi - Accessibility tools for visually impaired users - Language learning applications - Speech technology products - AI-powered audio content creation

The English TTS Monologue Speech Dataset is a professionally curated resource built to train realistic, expressive, and production-grade text-to-speech (TTS) systems. It contains studio-recorded long-form speech by trained native English voice artists, each contributing 1 to 2 hours of clean, uninterrupted monologue audio.

Unlike typical prompt-based datasets with short, isolated phrases, this collection features long-form, topic-driven monologues that mirror natural human narration. It includes content types that are directly useful for real-world applications, like audiobook-style storytelling, educational lectures, health advisories, product explainers, digital how-tos, formal announcements, and more.

All recordings are captured in professional studios using high-end equipment and under the guidance of experienced voice directors.

Recording & Audio Quality

Only clean, production-grade audio makes it into the final dataset.

Voice Artist Selection

All voice artists are native English speakers with professional training or prior experience in narration. We ensure a diverse pool in terms of age, gender, and region to bring a balanced and rich vocal dataset.

Script Quality & Coverage

Scripts are not generic or repetitive. Scripts are professionally authored by domain experts to reflect real-world use cases. They avoid redundancy and include modern vocabulary, emotional range, and phonetically rich sentence structures.

Transcripts & Alignment

While the script is used during the recording, we also provide post-recording updates to ensure the transcript reflects the final spoken audio. Minor edits are made to adjust for skipped or rephrased words.

A Curated Crowdsourced Dataset of Luganda and Swahili Speech for Text-to-Speech Synthesis

Description

This dataset contains curated and preprocessed speech recordings in Luganda and Kiswahili for use in text-to-speech (TTS) research (Katumba et al., 2025). The audio and transcripts were sourced from Mozilla Common Voice (Luganda v12.0 and Kiswahili v15.0) and curated for voice consistency and quality. This dataset is designed for training and evaluating end-to-end TTS systems in low-resource African languages. The data is organized into two folders, Luganda and Kiswahili, each containing:

wavs.zip: A ZIP archive of .wav audio files from six selected female speakers per language. All audio files have been silence-trimmed, denoised using a causal DEMUCS model, and filtered using WV-MOS to retain only clips with a predicted MOS ≥ 3.5.

metadata.csv: A CSV file with two columns: filename and transcript. Each row corresponds to an audio file in the wavs.zip archive and provides the spoken sentence for that clip.

This dataset was used in: Katumba, A., Kagumire, S., Nakatumba-Nabende, J., Quinn, J., & Murindanyi, S. (2025). Building Text-to-Speech Models for Low-Resourced Languages From Crowdsourced Data. Applied AI Letters, 6:e117. https://doi.org/10.1002/ail2.117

Article https://doi.org/10.1002/ail2.117

Categories

Computer Science, Speech Processing, Natural Language Processing, Text-to-Speech, Deep Learning

Data Source: Mendeley Dataset

Trump Voice Dataset

This dataset serves as an example for training a text-to-speech (TTS) fine-tuning platform. It consists of three columns:

path (required): The file path to the audio. transcript (optional): The text transcript of the audio. speaker_id (optional): The unique identifier for the speaker.

If the transcript is not provided, it will be automatically generated using the Whisper-large v3 model.

The Text-To-Speech Market is Segmented by Component (Software and Services), Deployment Mode (Cloud-Based, On-Premise, and Edge Embedded), Voice Type (Neural/AI-based, Standard Concatenative, and Hybrid), Application (Consumer Media and Entertainment, E-Learning and Education, Customer Service, and More), Language (English, Spanish, Hindi, Chinese, and More), and Geography. The Market Forecasts are Provided in Terms of Value (USD).

This is the public release of the Samsung Open Mean Opinion Scores (SOMOS) dataset for the evaluation of neural text-to-speech (TTS) synthesis, which consists of audio files generated with a public domain voice from trained TTS models based on bibliography, and numbers assigned to each audio as quality (naturalness) evaluations by several crowdsourced listeners.DescriptionThe SOMOS dataset contains 20,000 synthetic utterances (wavs), 100 natural utterances and 374,955 naturalness evaluations (human-assigned scores in the range 1-5). The synthetic utterances are single-speaker, generated by training several Tacotron-like acoustic models and an LPCNet vocoder on the LJ Speech voice public dataset. 2,000 text sentences were synthesized, selected from Blizzard Challenge texts of years 2007-2016, the LJ Speech corpus as well as Wikipedia and general domain data from the Internet.Naturalness evaluations were collected via crowdsourcing a listening test on Amazon Mechanical Turk in the US, GB and CA locales. The records of listening test participants (workers) are fully anonymized. Statistics on the reliability of the scores assigned by the workers are also included, generated through processing the scores and validation controls per submission page.

To listen to audio samples of the dataset, please see our Github page.

The dataset release comes with a carefully designed train-validation-test split (70%-15%-15%) with unseen systems, listeners and texts, which can be used for experimentation on MOS prediction.

This version also contains the necessary resources to obtain the transcripts corresponding to all dataset audios.

Terms of use

The dataset may be used for research purposes only, for non-commercial purposes only, and may be distributed with the same terms.

Every time you produce research that has used this dataset, please cite the dataset appropriately.

Cite as:

@inproceedings{maniati22_interspeech, author={Georgia Maniati and Alexandra Vioni and Nikolaos Ellinas and Karolos Nikitaras and Konstantinos Klapsas and June Sig Sung and Gunu Jho and Aimilios Chalamandaris and Pirros Tsiakoulis}, title={{SOMOS: The Samsung Open MOS Dataset for the Evaluation of Neural Text-to-Speech Synthesis}}, year=2022, booktitle={Proc. Interspeech 2022}, pages={2388--2392}, doi={10.21437/Interspeech.2022-10922} }

References of resources & models used

Voice & synthesized texts:K. Ito and L. Johnson, “The LJ Speech Dataset,” https://keithito.com/LJ-Speech-Dataset/, 2017.

Vocoder:J.-M. Valin and J. Skoglund, “LPCNet: Improving neural speech synthesis through linear prediction,” in Proc. ICASSP, 2019.R. Vipperla, S. Park, K. Choo, S. Ishtiaq, K. Min, S. Bhattacharya, A. Mehrotra, A. G. C. P. Ramos, and N. D. Lane, “Bunched lpcnet: Vocoder for low-cost neural text-to-speech systems,” in Proc. Interspeech, 2020.

Acoustic models:N. Ellinas, G. Vamvoukakis, K. Markopoulos, A. Chalamandaris, G. Maniati, P. Kakoulidis, S. Raptis, J. S. Sung, H. Park, and P. Tsiakoulis, “High quality streaming speech synthesis with low, sentence-length-independent latency,” in Proc. Interspeech, 2020.Y. Wang, R. Skerry-Ryan, D. Stanton, Y. Wu, R. J. Weiss, N. Jaitly, Z. Yang, Y. Xiao, Z. Chen, S. Bengio et al., “Tacotron: Towards End-to-End Speech Synthesis,” in Proc. Interspeech, 2017.J. Shen, R. Pang, R. J. Weiss, M. Schuster, N. Jaitly, Z. Yang, Z. Chen, Y. Zhang, Y. Wang, R. Skerrv-Ryan et al., “Natural TTS Synthesis by Conditioning Wavenet on MEL Spectrogram Predictions,” in Proc. ICASSP, 2018.J. Shen, Y. Jia, M. Chrzanowski, Y. Zhang, I. Elias, H. Zen, and Y. Wu, “Non-Attentive Tacotron: Robust and Controllable Neural TTS Synthesis Including Unsupervised Duration Modeling,” arXiv preprint arXiv:2010.04301, 2020.M. Honnibal and M. Johnson, “An Improved Non-monotonic Transition System for Dependency Parsing,” in Proc. EMNLP, 2015.M. Dominguez, P. L. Rohrer, and J. Soler-Company, “PyToBI: A Toolkit for ToBI Labeling Under Python,” in Proc. Interspeech, 2019.Y. Zou, S. Liu, X. Yin, H. Lin, C. Wang, H. Zhang, and Z. Ma, “Fine-grained prosody modeling in neural speech synthesis using ToBI representation,” in Proc. Interspeech, 2021.K. Klapsas, N. Ellinas, J. S. Sung, H. Park, and S. Raptis, “WordLevel Style Control for Expressive, Non-attentive Speech Synthesis,” in Proc. SPECOM, 2021.T. Raitio, R. Rasipuram, and D. Castellani, “Controllable neural text-to-speech synthesis using intuitive prosodic features,” in Proc. Interspeech, 2020.

Synthesized texts from the Blizzard Challenges 2007, 2008, 2009, 2010, 2011, 2012, 2013, 2016:M. Fraser and S. King, "The Blizzard Challenge 2007," in Proc. SSW6, 2007.V. Karaiskos, S. King, R. A. Clark, and C. Mayo, "The Blizzard Challenge 2008," in Proc. Blizzard Challenge Workshop, 2008.A. W. Black, S. King, and K. Tokuda, "The Blizzard Challenge 2009," in Proc. Blizzard Challenge, 2009.S. King and V. Karaiskos, "The Blizzard Challenge 2010," 2010.S. King and V. Karaiskos, "The Blizzard Challenge 2011," 2011.S. King and V. Karaiskos, "The Blizzard Challenge 2012," 2012.S. King and V. Karaiskos, "The Blizzard Challenge 2013," 2013.S. King and V. Karaiskos, "The Blizzard Challenge 2016," 2016.

Contact

Alexandra Vioni - a.vioni@samsung.com

If you have any questions or comments about the dataset, please feel free to write to us.

We are interested in knowing if you find our dataset useful! If you use our dataset, please email us and tell us about your research.

Dataset Card for People's Speech

  Dataset Summary

The People's Speech Dataset is among the world's largest English speech recognition corpus today that is licensed for academic and commercial usage under CC-BY-SA and CC-BY 4.0. It includes 30,000+ hours of transcribed speech in English languages with a diverse set of speakers. This open dataset is large enough to train speech-to-text systems and crucially is available with a permissive license.

  Supported Tasks… See the full description on the dataset page: https://huggingface.co/datasets/MLCommons/peoples_speech.

This dataset represents a comprehensive resource for advancing Kurdish TTS systems. Converting text to speech is one of the important topics in the design and construction of multimedia systems, human-machine communication, and information and communication technology, and its purpose, along with speech recognition, is to establish communication between humans and machines in its most basic and natural form, that is, spoken language.
For our text corpus, we collected 6,565 sentences from a set of texts in various categories, including news, sport, health, question and exclamation sentences, science, general information, politics, education and literature, story, miscellaneous, and tourism, to create the train sentences. We thoroughly reviewed the texts and normalized them, then they were recorded by a male speaker. We recorded audios in a voice recording studio at 44,100Hz, and all audio files are down sampled to 22,050 Hz in our modeling process. The audio ranges from 3 to 36 seconds in length. We generate the speech corpus in this method, and the last speech has about 6,565 texts and audio pairings, which takes around 19 hours. Altogether, audio files are saved in wave format, and the texts are saved in text files in the corresponding sub-folders. Furthermore, for model training, all of the audio files are gathered in a single folder. Each line in the transcript files is formatted as WAVS | audio file’s name.wav| transcript. The audio file’s name includes the extensions, and the transcript was the speech's text. The audio recording and editing process lasted for 90 days. It involved capturing over 6,565 WAV files and over 19 h of recorded speech. The data set helps researchers improve Kurdish TTS early, thereby reducing the time consumed for this process.

Acknowledgments: We would like to express our sincere gratitude to Ayoub Mohammadzadeh for his invaluable support in recording the corpus.

The Russian TTS Monologue Speech Dataset is a professionally curated resource built to train realistic, expressive, and production-grade text-to-speech (TTS) systems. It contains studio-recorded long-form speech by trained native Russian voice artists, each contributing 1 to 2 hours of clean, uninterrupted monologue audio.

Unlike typical prompt-based datasets with short, isolated phrases, this collection features long-form, topic-driven monologues that mirror natural human narration. It includes content types that are directly useful for real-world applications, like audiobook-style storytelling, educational lectures, health advisories, product explainers, digital how-tos, formal announcements, and more.

All recordings are captured in professional studios using high-end equipment and under the guidance of experienced voice directors.

Recording & Audio Quality

Only clean, production-grade audio makes it into the final dataset.

Voice Artist Selection

All voice artists are native Russian speakers with professional training or prior experience in narration. We ensure a diverse pool in terms of age, gender, and region to bring a balanced and rich vocal dataset.

Script Quality & Coverage

Scripts are not generic or repetitive. Scripts are professionally authored by domain experts to reflect real-world use cases. They avoid redundancy and include modern vocabulary, emotional range, and phonetically rich sentence structures.

Transcripts & Alignment

While the script is used during the recording, we also provide post-recording updates to ensure the transcript reflects the final spoken audio. Minor edits are made to adjust for skipped or rephrased words.

Dataset Card for Dataset Name

This dataset card aims to be a base template for new datasets. It has been generated using this raw template.

  Dataset Details





  Dataset Description

Curated by: [More Information Needed] Funded by [optional]: [More Information Needed] Shared by [optional]: [More Information Needed] Language(s) (NLP): [More Information Needed] License: [More Information Needed]

  Dataset Sources [optional]

Repository: [More… See the full description on the dataset page: https://huggingface.co/datasets/shoyimobloqulov/text-to-speech-tts.

The dataset has high-quality data recorded in a professional studio.

Archives with a trimmed tag are having removed silence (aligned) using https://github.com/proger/uk

Features

Quality: high

Duration: 10h37m

Audio formats: OPUS/WAV

Text format: JSONL (a metadata.jsonl file)

Frequency: 16000/22050/48000 Hz

With the rapid development of deep learning techniques, the generation and counterfeiting of multimedia material are becoming increasingly straightforward to perform. At the same time, sharing fake content on the web has become so simple that malicious users can create unpleasant situations with minimal effort. Also, forged media are getting more and more complex, with manipulated videos (e.g., deepfakes where both the visual and audio contents can be counterfeited) that are taking the scene over still images. The multimedia forensic community has addressed the possible threats that this situation could imply by developing detectors that verify the authenticity of multimedia objects. However, the vast majority of these tools only analyze one modality at a time. This was not a problem as long as still images were considered the most widely edited media, but now, since manipulated videos are becoming customary, performing monomodal analyses could be reductive. Nonetheless, there is a lack in the literature regarding multimodal detectors (systems that consider both audio and video components). This is due to the difficulty of developing them but also to the scarsity of datasets containing forged multimodal data to train and test the designed algorithms.

In this paper we focus on the generation of an audio-visual deepfake dataset. First, we present a general pipeline for synthesizing speech deepfake content from a given real or fake video, facilitating the creation of counterfeit multimodal material. The proposed method uses Text-to-Speech (TTS) and Dynamic Time Warping (DTW) techniques to achieve realistic speech tracks. Then, we use the pipeline to generate and release TIMIT-TTS, a synthetic speech dataset containing the most cutting-edge methods in the TTS field. This can be used as a standalone audio dataset, or combined with DeepfakeTIMIT and VidTIMIT video datasets to perform multimodal research. Finally, we present numerous experiments to benchmark the proposed dataset in both monomodal (i.e., audio) and multimodal (i.e., audio and video) conditions. This highlights the need for multimodal forensic detectors and more multimodal deepfake data.

For the initial version of TIMIT-TTS v1.0

Arxiv: https://arxiv.org/abs/2209.08000

TIMIT-TTS Database v1.0: https://zenodo.org/record/6560159

The Thai TTS Monologue Speech Dataset is a professionally curated resource built to train realistic, expressive, and production-grade text-to-speech (TTS) systems. It contains studio-recorded long-form speech by trained native Thai voice artists, each contributing 1 to 2 hours of clean, uninterrupted monologue audio.

Unlike typical prompt-based datasets with short, isolated phrases, this collection features long-form, topic-driven monologues that mirror natural human narration. It includes content types that are directly useful for real-world applications, like audiobook-style storytelling, educational lectures, health advisories, product explainers, digital how-tos, formal announcements, and more.

All recordings are captured in professional studios using high-end equipment and under the guidance of experienced voice directors.

Recording & Audio Quality

Only clean, production-grade audio makes it into the final dataset.

Voice Artist Selection

All voice artists are native Thai speakers with professional training or prior experience in narration. We ensure a diverse pool in terms of age, gender, and region to bring a balanced and rich vocal dataset.

Script Quality & Coverage

Scripts are not generic or repetitive. Scripts are professionally authored by domain experts to reflect real-world use cases. They avoid redundancy and include modern vocabulary, emotional range, and phonetically rich sentence structures.

Transcripts & Alignment

While the script is used during the recording, we also provide post-recording updates to ensure the transcript reflects the final spoken audio. Minor edits are made to adjust for skipped or rephrased words.

The text-to-speech market size is expected to reach USD 7.06 Billion in 2028 and register a CAGR of 14.7%. Text-to-speech industry report classifies global market by share, trend, and on the basis of deployment mode, voice type, organization size, vertical, and region

The Arabic TTS Monologue Speech Dataset is a professionally curated resource built to train realistic, expressive, and production-grade text-to-speech (TTS) systems. It contains studio-recorded long-form speech by trained native Arabic voice artists, each contributing 1 to 2 hours of clean, uninterrupted monologue audio.

Unlike typical prompt-based datasets with short, isolated phrases, this collection features long-form, topic-driven monologues that mirror natural human narration. It includes content types that are directly useful for real-world applications, like audiobook-style storytelling, educational lectures, health advisories, product explainers, digital how-tos, formal announcements, and more.

All recordings are captured in professional studios using high-end equipment and under the guidance of experienced voice directors.

Recording & Audio Quality

Only clean, production-grade audio makes it into the final dataset.

Voice Artist Selection

All voice artists are native Arabic speakers with professional training or prior experience in narration. We ensure a diverse pool in terms of age, gender, and region to bring a balanced and rich vocal dataset.

Script Quality & Coverage

Scripts are not generic or repetitive. Scripts are professionally authored by domain experts to reflect real-world use cases. They avoid redundancy and include modern vocabulary, emotional range, and phonetically rich sentence structures.

Transcripts & Alignment

While the script is used during the recording, we also provide post-recording updates to ensure the transcript reflects the final spoken audio. Minor edits are made to adjust for skipped or rephrased words.

German(Germany) Spontaneous Dialogue Smartphone speech dataset, collected from dialogues based on given topics. Transcribed with text content, timestamp, speaker's ID, gender and other attributes.

The Finnish TTS Monologue Speech Dataset is a professionally curated resource built to train realistic, expressive, and production-grade text-to-speech (TTS) systems. It contains studio-recorded long-form speech by trained native Finnish voice artists, each contributing 1 to 2 hours of clean, uninterrupted monologue audio.

Unlike typical prompt-based datasets with short, isolated phrases, this collection features long-form, topic-driven monologues that mirror natural human narration. It includes content types that are directly useful for real-world applications, like audiobook-style storytelling, educational lectures, health advisories, product explainers, digital how-tos, formal announcements, and more.

All recordings are captured in professional studios using high-end equipment and under the guidance of experienced voice directors.

Recording & Audio Quality

Only clean, production-grade audio makes it into the final dataset.

Voice Artist Selection

All voice artists are native Finnish speakers with professional training or prior experience in narration. We ensure a diverse pool in terms of age, gender, and region to bring a balanced and rich vocal dataset.

Script Quality & Coverage

Scripts are not generic or repetitive. Scripts are professionally authored by domain experts to reflect real-world use cases. They avoid redundancy and include modern vocabulary, emotional range, and phonetically rich sentence structures.

Transcripts & Alignment

While the script is used during the recording, we also provide post-recording updates to ensure the transcript reflects the final spoken audio. Minor edits are made to adjust for skipped or rephrased words.

Neural network based end-to-end text to speech (TTS) has significantly improved the quality of synthesized speech. Prominent methods (e.g., Tacotron 2) usually first generate mel-spectrogram from text, and then synthesize speech from the mel-spectrogram using vocoder such as WaveNet. Compared with traditional concatenative and statistical parametric approaches, neural network based end-to-end models suffer from slow inference speed, and the synthesized speech is usually not robust (i.e., some words are skipped or repeated) and lack of con-trollability (voice speed or prosody control). In this work, we propose a novel feed-forward network based on Transformer to generate mel-spectrogram in parallel for TTS.

English(the United States) Spontaneous Dialogue Smartphone speech dataset, collected from dialogues based on given topics, covering generic domain. Transcribed with text content, speaker's ID, gender and other attributes. Our dataset was collected from extensive and diversify speakers(1,416 Americans), geographicly speaking, enhancing model performance in real and complex tasks. Quality tested by various AI companies. We strictly adhere to data protection regulations and privacy standards, ensuring the maintenance of user privacy and legal rights throughout the data collection, storage, and usage processes, our datasets are all GDPR, CCPA, PIPL complied.

Format

16kHz, 16 bit, uncompressed wav, mono channel;

Content category

generic domain;

Recording condition

Low background noise;

Recording device

Android Smartphone, iPhone;

Speaker

1,416 Americans, with 45% male and 55% female;

Country

the United States(USA);

Language(Region) Code

en-US;

Language

English;

Features of annotation

Transcription text, speaker ID, gender;

Accuracy Rate

Sentence Accuracy Rate (SAR) 95%

Dataset Summary

We present a speech corpus for Classical Arabic Text-to-Speech (ClArTTS) to support the development of end-to-end TTS systems for Arabic. The speech is extracted from a LibriVox audiobook, which is then processed, segmented, and manually transcribed and annotated. The final ClArTTS corpus contains about 12 hours of speech from a single male speaker sampled at 40100 kHz.

  Dataset Structure

A typical data point comprises the name of the audio file, called… See the full description on the dataset page: https://huggingface.co/datasets/MBZUAI/ClArTTS.