The dataset contains speech samples of English, German and Spanish languages. Samples are equally balanced between languages, genders and speakers.

More information at the spoken-language-dataset repository.

Background

The project was inspired by the TopCoder contest, Spoken Languages 2. The given dataset contains 10 second of speech recorded in 1 of 176 languages. The entire dataset has been based on bible readings. Poorly, in many cases there is a single speaker per language (male in most cases). Even worse the same single speaker exists in the test set. Of course this can't lead to a good generic solution.

There are two ways we can take:

• First approach is to use a big dataset where all voice or language properties (e.g. gender, age, accent) become equally possible. A good example is the Common Voice from Mozilla. Most likely this leads to the best performance. However processing such a huge dataset is expensive and adding new languages is challenging.
• Second approach is to use a small handcrafted dataset and boost it with data augmentation. The advantage is that we can add new languages quickly. Last but not least the dataset is small thus it can be processed quickly.

The second approach has been taken.

LibriVox recordings were used to prepare the dataset. Particular attention was paid to a big variety of unique speakers. Big variance forces the model to concentrate more on language properties than a specific voice. Samples are equally balanced between languages, genders and speakers in order not to favour any subgroup. Finally the dataset is divided into train and test set. Speakers present in the test set, are not present in the train set. This helps estimate a generalization error.

The core of the train set is based on 420 minutes (2520 samples) of original recordings. After applying several audio transformations (pitch, speed and noise) the train set was extended to 12180 minutes (73080 samples). The test set contains 90 minutes (540 samples) of original recordings. No data augmentation has been applied.

Original recordings contain 90 unique speakers. The number of unique speakers was increased by adjusting pitch (8 different levels) and speed (8 different levels). After applying audio transformations there are 1530 unique speakers.

Data structure

The dataset is divided into 2 directories:

• train (73080 samples)
• test (540 samples)

Each sample is an FLAC audio file with:

• sample rate: 22050
• bit depth: 16
• channels: 1
• duration: 10 seconds (sharp)

The original recordings are MP3 files but they are converted into FLAC files quickly to avoid re-encoding (and losing quality) during transformations.

The filename of the sample has following syntax:

(language)_(gender)_(recording ID).fragment(index)[.(transformation)(index)].flac

...and variables:

• language: en, de, or es
• gender: m or f
• recording ID: a hash of the URL
• fragment index: 1-30
• transformation: speed, pitch or noise
• transformation index:
  • if speed: 1-8
  • if pitch: 1-8
  • if noise: 1-12

For example:

es_m_f7d959494477e5e7e33d4666f15311c9.fragment9.speed8.flac

Sample Model

The dataset was used to train the spoken language identification model. The trained model has 97% score (i.e. F1 metric) against the test set. Additionally it generalizes well which was confirmed against real life content. The fact that samples are prefeclty stratified was one of the reasons to achieve such a high performance.

Feel free to create your own model and share results!

Many residents of New York City speak more than one language; a number of them speak and understand non-English languages more fluently than English. This dataset, derived from the Census Bureau's American Community Survey (ACS), includes information on over 1.7 million limited English proficient (LEP) residents and a subset of that population called limited English proficient citizens of voting age (CVALEP) at the Community District level. There are 59 community districts throughout NYC, with each district being represented by a Community Board.

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.

Advancements in deep learning have revolutionized numerous real-world applications, including image recognition, visual question answering, and image captioning. Among these, image captioning has emerged as a critical area of research, with substantial progress achieved in Arabic, Chinese, Uyghur, Hindi, and predominantly English. However, despite Urdu being a morphologically rich and widely spoken language, research in Urdu image captioning remains underexplored due to a lack of resources. This study creates a new Urdu Image Captioning Dataset (UCID) called UC-23-RY to fill in the gaps in Urdu image captioning. The Flickr30k dataset inspired the 159,816 Urdu captions in the dataset. Additionally, it suggests deep learning architectures designed especially for Urdu image captioning, including NASNetLarge-LSTM and ResNet-50-LSTM. The NASNetLarge-LSTM and ResNet-50-LSTM models achieved notable BLEU-1 scores of 0.86 and 0.84 respectively, as demonstrated through evaluation in this study accessing the model’s impact on caption quality. Additionally, it provides useful datasets and shows how well-suited sophisticated deep learning models are for improving automatic Urdu image captioning.

Common English Parts-of-speech

Over 8,000 words and their plural forms

About this dataset

This dataset provides ample information on over 8,000 various English words, including nouns and their plural forms. By mining this data, researchers can gain valuable insights into understanding the English language in a more efficient way

How to use the dataset

This dataset can be used to help researchers understand the English language in a new and innovative way. The data includes information on over 8,000 different English words, including nouns and their plural forms. This dataset is particularly useful for investigating the relationships between words and their plural forms

Research Ideas

• To create a program that can automatically generating plural forms of nouns.
• To study the relationships between different words and their plural forms.
• To develop a better understanding of the English language for non-native speakers

Acknowledgements

License

See the dataset description for more information.

Columns

File: adjectives.csv

File: adverbs.csv

File: nouns.csv | Column name | Description | |:--------------|:-----------------------------------------------------------| | 007 | The code name of the character. (String) | | 007s | The number of times the character has been used. (Integer) |

File: plural-nouns.csv

File: verbs.csv | Column name | Description | |:--------------|:-----------------------------| | awake | (adjective) to stop sleeping | | awoke | (verb) to stop sleeping | | awoken | (verb) to stop sleeping |

File: words-multiple-present-participle.csv | Column name | Description | |:-----------------------------------|:-------------------------------------------------------------| | Word | The word being described. (String) | | Present Participle | The present participle form of the word. (String) | | Present Participle Alternative | An alternative present participle form of the word. (String) |

BLEU scores for different datasets in different languages.

Advancements in deep learning have revolutionized numerous real-world applications, including image recognition, visual question answering, and image captioning. Among these, image captioning has emerged as a critical area of research, with substantial progress achieved in Arabic, Chinese, Uyghur, Hindi, and predominantly English. However, despite Urdu being a morphologically rich and widely spoken language, research in Urdu image captioning remains underexplored due to a lack of resources. This study creates a new Urdu Image Captioning Dataset (UCID) called UC-23-RY to fill in the gaps in Urdu image captioning. The Flickr30k dataset inspired the 159,816 Urdu captions in the dataset. Additionally, it suggests deep learning architectures designed especially for Urdu image captioning, including NASNetLarge-LSTM and ResNet-50-LSTM. The NASNetLarge-LSTM and ResNet-50-LSTM models achieved notable BLEU-1 scores of 0.86 and 0.84 respectively, as demonstrated through evaluation in this study accessing the model’s impact on caption quality. Additionally, it provides useful datasets and shows how well-suited sophisticated deep learning models are for improving automatic Urdu image captioning.

Introduction

Welcome to the New Zealand English Scripted Monologue Speech Dataset for the Retail & E-commerce domain. This dataset is built to accelerate the development of English language speech technologies especially for use in retail-focused automatic speech recognition (ASR), natural language processing (NLP), voicebots, and conversational AI applications.

Speech Data

This training dataset includes 6,000+ high-quality scripted audio recordings in New Zealand English, created to reflect real-world scenarios in the Retail & E-commerce sector. These prompts are tailored to improve the accuracy and robustness of customer-facing speech technologies.

Topic Diversity

This dataset includes a comprehensive set of retail-specific topics to ensure wide linguistic coverage for AI training:

Contextual Enrichment

To increase training utility, prompts include contextual data such as:

These additions help your models learn to recognize structured and unstructured retail-related speech.

Transcription

Every audio file is paired with a verbatim transcription, ensuring consistency and alignment for model training.

Metadata

Detailed metadata is included to support filtering, analysis, and model evaluation:

Advancements in deep learning have revolutionized numerous real-world applications, including image recognition, visual question answering, and image captioning. Among these, image captioning has emerged as a critical area of research, with substantial progress achieved in Arabic, Chinese, Uyghur, Hindi, and predominantly English. However, despite Urdu being a morphologically rich and widely spoken language, research in Urdu image captioning remains underexplored due to a lack of resources. This study creates a new Urdu Image Captioning Dataset (UCID) called UC-23-RY to fill in the gaps in Urdu image captioning. The Flickr30k dataset inspired the 159,816 Urdu captions in the dataset. Additionally, it suggests deep learning architectures designed especially for Urdu image captioning, including NASNetLarge-LSTM and ResNet-50-LSTM. The NASNetLarge-LSTM and ResNet-50-LSTM models achieved notable BLEU-1 scores of 0.86 and 0.84 respectively, as demonstrated through evaluation in this study accessing the model’s impact on caption quality. Additionally, it provides useful datasets and shows how well-suited sophisticated deep learning models are for improving automatic Urdu image captioning.

Advancements in deep learning have revolutionized numerous real-world applications, including image recognition, visual question answering, and image captioning. Among these, image captioning has emerged as a critical area of research, with substantial progress achieved in Arabic, Chinese, Uyghur, Hindi, and predominantly English. However, despite Urdu being a morphologically rich and widely spoken language, research in Urdu image captioning remains underexplored due to a lack of resources. This study creates a new Urdu Image Captioning Dataset (UCID) called UC-23-RY to fill in the gaps in Urdu image captioning. The Flickr30k dataset inspired the 159,816 Urdu captions in the dataset. Additionally, it suggests deep learning architectures designed especially for Urdu image captioning, including NASNetLarge-LSTM and ResNet-50-LSTM. The NASNetLarge-LSTM and ResNet-50-LSTM models achieved notable BLEU-1 scores of 0.86 and 0.84 respectively, as demonstrated through evaluation in this study accessing the model’s impact on caption quality. Additionally, it provides useful datasets and shows how well-suited sophisticated deep learning models are for improving automatic Urdu image captioning.

Feature and textural extraction model with random weights and urdu vector for a multimodal approach.

Division of training and testing images based on a split ratio.

Time elapsed during training of ResNet-50-LSTM and NASNetLarge-LSTM model.