Introduction

Welcome to the Mexican Spanish General Conversation Speech Dataset — a rich, linguistically diverse corpus purpose-built to accelerate the development of Spanish speech technologies. This dataset is designed to train and fine-tune ASR systems, spoken language understanding models, and generative voice AI tailored to real-world Mexican Spanish communication.

Curated by FutureBeeAI, this 30 hours dataset offers unscripted, spontaneous two-speaker conversations across a wide array of real-life topics. It enables researchers, AI developers, and voice-first product teams to build robust, production-grade Spanish speech models that understand and respond to authentic Mexican accents and dialects.

Speech Data

The dataset comprises 30 hours of high-quality audio, featuring natural, free-flowing dialogue between native speakers of Mexican Spanish. These sessions range from informal daily talks to deeper, topic-specific discussions, ensuring variability and context richness for diverse use cases.

Topic Diversity

The dataset spans a wide variety of everyday and domain-relevant themes. This topic diversity ensures the resulting models are adaptable to broad speech contexts.

Transcription

Each audio file is paired with a human-verified, verbatim transcription available in JSON format.

These transcriptions are production-ready, enabling seamless integration into ASR model pipelines or conversational AI workflows.

Metadata

The dataset comes with granular metadata for both speakers and recordings:

Such metadata helps developers fine-tune model training and supports use-case-specific filtering or demographic analysis.

Usage and Applications

This dataset is a versatile resource for multiple Spanish speech and language AI applications:

Introduction

This Spanish Call Center Speech Dataset for the Real Estate industry is purpose-built to accelerate the development of speech recognition, spoken language understanding, and conversational AI systems tailored for Spanish -speaking Real Estate customers. With over 30 hours of unscripted, real-world audio, this dataset captures authentic conversations between customers and real estate agents ideal for building robust ASR models.

Curated by FutureBeeAI, this dataset equips voice AI developers, real estate tech platforms, and NLP researchers with the data needed to create high-accuracy, production-ready models for property-focused use cases.

Speech Data

The dataset features 30 hours of dual-channel call center recordings between native Spanish speakers. Captured in realistic real estate consultation and support contexts, these conversations span a wide array of property-related topics from inquiries to investment advice offering deep domain coverage for AI model development.

Topic Diversity

This speech corpus includes both inbound and outbound calls, featuring positive, neutral, and negative outcomes across a wide range of real estate scenarios.

Such domain-rich variety ensures model generalization across common real estate support conversations.

Transcription

All recordings are accompanied by precise, manually verified transcriptions in JSON format.

These transcriptions streamline ASR and NLP development for Spanish real estate voice applications.

Metadata

Detailed metadata accompanies each participant and conversation:

This enables smart filtering, dialect-focused model training, and structured dataset exploration.

Usage and Applications

This dataset is ideal for voice AI and NLP systems built for the real estate sector:

Introduction

Welcome to the Colombian Spanish General Conversation Speech Dataset — a rich, linguistically diverse corpus purpose-built to accelerate the development of Spanish speech technologies. This dataset is designed to train and fine-tune ASR systems, spoken language understanding models, and generative voice AI tailored to real-world Colombian Spanish communication.

Curated by FutureBeeAI, this 30 hours dataset offers unscripted, spontaneous two-speaker conversations across a wide array of real-life topics. It enables researchers, AI developers, and voice-first product teams to build robust, production-grade Spanish speech models that understand and respond to authentic Colombian accents and dialects.

Speech Data

The dataset comprises 30 hours of high-quality audio, featuring natural, free-flowing dialogue between native speakers of Colombian Spanish. These sessions range from informal daily talks to deeper, topic-specific discussions, ensuring variability and context richness for diverse use cases.

Topic Diversity

The dataset spans a wide variety of everyday and domain-relevant themes. This topic diversity ensures the resulting models are adaptable to broad speech contexts.

Transcription

Each audio file is paired with a human-verified, verbatim transcription available in JSON format.

These transcriptions are production-ready, enabling seamless integration into ASR model pipelines or conversational AI workflows.

Metadata

The dataset comes with granular metadata for both speakers and recordings:

Such metadata helps developers fine-tune model training and supports use-case-specific filtering or demographic analysis.

Usage and Applications

This dataset is a versatile resource for multiple Spanish speech and language AI applications:

The first public large-scale multilingual Twitter dataset related to the FIFA World Cup 2022, comprising over 28 million posts in 69 unique spoken languages, including Arabic, English, Spanish, French, and many others. This dataset aims to facilitate research in future sentiment analysis, cross-linguistic studies, event-based analytics, meme and hate speech detection, fake news detection, and social manipulation detection.

The file 🚨Qatar22WC.csv🚨 contains tweet-level and user-level metadata for our collected tweets. 🚀Codebook for FIFA World Cup 2022 Twitter Dataset🚀 | Column Name | Description| |-------------------------------- |----------------------------------------------------------------------------------------| | day, month, year | The date where the tweet posted | | hou, min, sec | Hour, minute, and second of tweet timestamp | | age_of_the_user_account | User Account age in days | | tweet_count | Total number of tweets posted by the user | | location | User-defined location field | | follower_count | Number of followers the user has | | following_count | Number of accounts the user is following | | follower_to_Following | Follower-following ratio | | favouite_count | Number of likes the user did| | verified | Boolean indicating if the user is verified (1 = Verified, 0 = Not Verified) | | Avg_tweet_count | Average tweets per day for the user activity| | list_count | Number of lists the user is a member | | Tweet_Id | Tweet ID | | is_reply_tweet | ID of the tweet being replied to (if applicable) | | is_quote | boolean representing if the tweet is a quote | | retid | Retweet ID if it's a retweet; NaN otherwise | | lang | Language of the tweet | | hashtags | The keyword or hashtag used to collect the tweet | | is_image, | Boolean indicating if the tweet associated with image| | is_video | Boolean indicating if the tweet associated with video | |-------------------------------|----------------------------------------------------------------------------------------|

Examples of use case queries are described in the file 🚨fifa_wc_qatar22_examples_of_use_case_queries.ipynb🚨 and accessible via: https://github.com/khairied/Qata_FIFA_World_Cup_22

🚀 Please Cite This as: Daouadi, K. E., Boualleg, Y., Guehairia, O. & Taleb-Ahmed, A. (2025). Tracking the Global Pulse: The first public Twitter dataset from FIFA World Cup, Journal of Computational Social Science.

Introduction

The Spanish General Domain Chat Dataset is a high-quality, text-based dataset designed to train and evaluate conversational AI, NLP models, and smart assistants in real-world Spanish usage. Collected through FutureBeeAI’s trusted crowd community, this dataset reflects natural, native-level Spanish conversations covering a broad spectrum of everyday topics.

Conversational Text Data

This dataset includes over 15000 chat transcripts, each featuring free-flowing dialogue between two native Spanish speakers. The conversations are spontaneous, context-rich, and mimic informal, real-life texting behavior.

Diversity and Domain Coverage

Conversations span a wide variety of general-domain topics to ensure comprehensive model exposure:

This diversity ensures the dataset is useful across multiple NLP and language understanding applications.

Linguistic Authenticity

Chats reflect informal, native-level Spanish usage with:

Metadata

Every chat instance is accompanied by structured metadata, which includes:

This metadata supports model filtering, demographic-specific evaluation, and more controlled fine-tuning workflows.

Data Quality Assurance

All chat records pass through a rigorous QA process to maintain consistency and accuracy:

This ensures a clean, reliable dataset ready for high-performance AI model training.

Applications

This dataset is ideal for training and evaluating a wide range of text-based AI systems:

Cyberbullying has become an increasingly prevalent issue in the digital age, with the rise of social media and online communication. It can take many forms, including verbal attacks, harassment, and discrimination, and it can have serious consequences for victims, including depression, anxiety, and even suicide. While much research has been done on cyberbullying in languages such as English, Spanish, and Chinese, there has been little focus on languages spoken by smaller populations, such as Kurdish. Kurdish is a language spoken by millions of people in the Middle East, including Turkey, Iran, Iraq, and Syria. It is an Indo-European language with several dialects, and it is considered an official language in Iraq and an official regional language in Iran. Despite its widespread use, there has been very little research on cyberbullying in Kurdish, and there are currently no datasets available that specifically focus on this issue. To address this gap, we have created the first ever cyberbullying dataset for the Kurdish language. This dataset contains three classes: neutral, racism, and sexism. The neutral class includes messages that do not contain any form of cyberbullying, while the racism and sexism classes include messages that contain discriminatory language based on race or gender, respectively. The dataset was created using a combination of manual and automated techniques. We collected a large number of messages from Twitter API, that were written in Kurdish. We then manually labeled these messages based on whether they contained cyberbullying or not, and further categorized them into the three classes. The resulting dataset contains over 30,000 messages, with roughly equal distribution among the three classes. It is a valuable resource for researchers and practitioners who are interested in studying cyberbullying in the Kurdish language and developing strategies to combat it. The dataset can be used for a variety of purposes, including training machine learning models to detect cyberbullying in Kurdish, analyzing the language used in cyberbullying messages to identify patterns and trends, and developing interventions to prevent and address cyberbullying in Kurdish-speaking communities.

Introduction

Welcome to the Spanish General Conversation Speech Dataset — a rich, linguistically diverse corpus purpose-built to accelerate the development of Spanish speech technologies. This dataset is designed to train and fine-tune ASR systems, spoken language understanding models, and generative voice AI tailored to real-world Spanish communication.

Curated by FutureBeeAI, this 30 hours dataset offers unscripted, spontaneous two-speaker conversations across a wide array of real-life topics. It enables researchers, AI developers, and voice-first product teams to build robust, production-grade Spanish speech models that understand and respond to authentic Spanish accents and dialects.

Speech Data

The dataset comprises 30 hours of high-quality audio, featuring natural, free-flowing dialogue between native speakers of Spanish. These sessions range from informal daily talks to deeper, topic-specific discussions, ensuring variability and context richness for diverse use cases.

Topic Diversity

The dataset spans a wide variety of everyday and domain-relevant themes. This topic diversity ensures the resulting models are adaptable to broad speech contexts.

Transcription

Each audio file is paired with a human-verified, verbatim transcription available in JSON format.

These transcriptions are production-ready, enabling seamless integration into ASR model pipelines or conversational AI workflows.

Metadata

The dataset comes with granular metadata for both speakers and recordings:

Such metadata helps developers fine-tune model training and supports use-case-specific filtering or demographic analysis.

Usage and Applications

This dataset is a versatile resource for multiple Spanish speech and language AI applications:

Introduction

The Spanish Healthcare Chat Dataset is a rich collection of over 10,000 text-based conversations between customers and call center agents, focused on real-world healthcare interactions. Designed to reflect authentic language use and domain-specific dialogue patterns, this dataset supports the development of conversational AI, chatbots, and NLP models tailored for healthcare applications in Spanish-speaking regions.

Participant & Chat Overview

Topic Diversity

The dataset captures a wide spectrum of healthcare-related chat scenarios, ensuring comprehensive coverage for training robust AI systems:

This variety helps simulate realistic healthcare support workflows and patient-agent dynamics.

Language Diversity & Realism

This dataset reflects the natural flow of Spanish healthcare communication and includes:

These elements ensure the dataset is contextually relevant and linguistically rich for real-world use cases.

Conversational Flow & Structure

Conversations range from simple inquiries to complex advisory sessions, including:

Each conversation typically includes these structural components:

This structured flow mirrors actual healthcare support conversations and is ideal for training advanced dialogue systems.

Data Format & Structure

Available in JSON, CSV, and TXT formats, each conversation includes:

Applications

The GlobalPhone corpus developed in collaboration with the Karlsruhe Institute of Technology (KIT) was designed to provide read speech data for the development and evaluation of large continuous speech recognition systems in the most widespread languages of the world, and to provide a uniform, multilingual speech and text database for language independent and language adaptive speech recognition as well as for language identification tasks. The entire GlobalPhone corpus enables the acquisition of acoustic-phonetic knowledge of the following 22 spoken languages: Arabic (ELRA-S0192), Bulgarian (ELRA-S0319), Chinese-Mandarin (ELRA-S0193), Chinese-Shanghai (ELRA-S0194), Croatian (ELRA-S0195), Czech (ELRA-S0196), French (ELRA-S0197), German (ELRA-S0198), Hausa (ELRA-S0347), Japanese (ELRA-S0199), Korean (ELRA-S0200), Polish (ELRA-S0320), Portuguese (Brazilian) (ELRA-S0201), Russian (ELRA-S0202), Spanish (Latin America) (ELRA-S0203), Swahili (ELRA-S0375), Swedish (ELRA-S0204), Tamil (ELRA-S0205), Thai (ELRA-S0321), Turkish (ELRA-S0206), Ukrainian (ELRA-S0377), and Vietnamese (ELRA-S0322).In each language about 100 sentences were read from each of the 100 speakers. The read texts were selected from national newspapers available via Internet to provide a large vocabulary. The read articles cover national and international political news as well as economic news. The speech is available in 16bit, 16kHz mono quality, recorded with a close-speaking microphone (Sennheiser 440-6). The transcriptions are internally validated and supplemented by special markers for spontaneous effects like stuttering, false starts, and non-verbal effects like laughing and hesitations. Speaker information like age, gender, occupation, etc. as well as information about the recording setup complement the database. The entire GlobalPhone corpus contains over 450 hours of speech spoken by more than 2100 native adult speakers.Data is shortened by means of the shorten program written by Tony Robinson. Alternatively, the data could be delivered unshorten.The Polish part of GlobalPhone was collected from altogether 102 native speakers in Poland, of which 48 speakers were female and 54 speakers were male. The majority of speakers are between 20 and 39 years old, the age distribution ranges from 18 to 65 years. Most of the speakers are non-smokers in good health conditions. Each speaker read on average about 100 utterances from newspaper articles, in total we recorded 10130 utterances. The speech was recorded using a close-talking microphone Sennheiser HM420 in a push-to-talk scenario. All data were recorded at 16kHz and 16bit resolution in PCM format. The data collection took place in small and large rooms, about half of the recordings took place under very quiet noise conditions, the other half with moderate background noise. Information on recording place and environmental noise conditions are provided in a separate speaker session file for each speaker. The text data used for reco...

This dataset was created from the CDC's National Vital Statistics Reports Volume 56, Number 6. The dataset includes all data available from this report by state level and includes births by race and Hispanic origin, births to unmarried women, rates of cesarean delivery, and twin and multiple birth rates. The data are final for 2005. No value is represented by a -1. "Descriptive tabulations of data reported on the birth certificates of the 4.1 million births that occurred in 2005 are presented. Denominators for population-based rates are postcensal estimates derived from the U.S. 2000 census".

The Global English Accent Conversational NLP Dataset is a comprehensive collection of validated English speech recordings sourced from native and non-native English speakers across key global regions. This dataset is designed for training Natural Language Processing models, conversational AI, Automatic Speech Recognition (ASR), and linguistic research, with a focus on regional accent variation.

Regions and Covered Countries with Primary Spoken Languages:

Africa: South Africa (English, Zulu, Afrikaans, Xhosa) Nigeria (English, Yoruba, Igbo, Hausa) Kenya (English, Swahili) Ghana (English, Twi, Ewe, Ga) Uganda (English, Luganda) Ethiopia (English, Amharic, Oromo)

Central & South America: Mexico (Spanish, English as a second language) Guatemala (Spanish, K'iche', English) El Salvador (Spanish, English) Costa Rica (Spanish, English in Caribbean regions) Colombia (Spanish, English in urban centers) Dominican Republic (Spanish, English in tourist zones) Brazil (Portuguese, English in urban areas) Argentina (Spanish, English among educated speakers)

Southeast Asia & South Asia: Philippines (Filipino, English) Vietnam (Vietnamese, English) Malaysia (Malay, English, Mandarin) Indonesia (Indonesian, Javanese, English) Singapore (English, Mandarin, Malay, Tamil) India (Hindi, English, Bengali, Tamil) Pakistan (Urdu, English, Punjabi)

Europe: United Kingdom (English) Ireland (English, Irish) Germany (German, English) France (French, English) Spain (Spanish, Catalan, English) Italy (Italian, English) Portugal (Portuguese, English)

Oceania: Australia (English) New Zealand (English, Māori) Fiji (English, Fijian) North America: United States (English, Spanish) Canada (English, French)

Dataset Attributes: - Conversational English with natural accent variation - Global coverage with balanced male/female speakers - Rich speaker metadata: age, gender, country, city - Average audio length of ~30 minutes per participant - All samples manually validated for accuracy - Structured format suitable for machine learning and AI applications

Best suited for: - NLP model training and evaluation - Multilingual ASR system development - Voice assistant and chatbot design - Accent recognition research - Voice synthesis and TTS modeling

This dataset ensures global linguistic diversity and delivers high-quality audio for AI developers, researchers, and enterprises working on voice-based applications.

Abstract

Gallup Worldwide Research continually surveys residents in more than 150 countries, representing more than 98% of the world's adult population, using randomly selected, nationally representative samples. Gallup typically surveys 1,000 individuals in each country, using a standard set of core questions that has been translated into the major languages of the respective country. In some regions, supplemental questions are asked in addition to core questions. Face-to-face interviews are approximately 1 hour, while telephone interviews are about 30 minutes. In many countries, the survey is conducted once per year, and fieldwork is generally completed in two to four weeks. The Country Dataset Details spreadsheet displays each country's sample size, month/year of the data collection, mode of interviewing, languages employed, design effect, margin of error, and details about sample coverage.

Gallup is entirely responsible for the management, design, and control of Gallup Worldwide Research. For the past 70 years, Gallup has been committed to the principle that accurately collecting and disseminating the opinions and aspirations of people around the globe is vital to understanding our world. Gallup's mission is to provide information in an objective, reliable, and scientifically grounded manner. Gallup is not associated with any political orientation, party, or advocacy group and does not accept partisan entities as clients. Any individual, institution, or governmental agency may access the Gallup Worldwide Research regardless of nationality. The identities of clients and all surveyed respondents will remain confidential.

Kind of data

Sample survey data [ssd]

Sampling procedure

SAMPLING AND DATA COLLECTION METHODOLOGY With some exceptions, all samples are probability based and nationally representative of the resident population aged 15 and older. The coverage area is the entire country including rural areas, and the sampling frame represents the entire civilian, non-institutionalized, aged 15 and older population of the entire country. Exceptions include areas where the safety of interviewing staff is threatened, scarcely populated islands in some countries, and areas that interviewers can reach only by foot, animal, or small boat.

Telephone surveys are used in countries where telephone coverage represents at least 80% of the population or is the customary survey methodology (see the Country Dataset Details for detailed information for each country). In Central and Eastern Europe, as well as in the developing world, including much of Latin America, the former Soviet Union countries, nearly all of Asia, the Middle East, and Africa, an area frame design is used for face-to-face interviewing.

The typical Gallup Worldwide Research survey includes at least 1,000 surveys of individuals. In some countries, oversamples are collected in major cities or areas of special interest. Additionally, in some large countries, such as China and Russia, sample sizes of at least 2,000 are collected. Although rare, in some instances the sample size is between 500 and 1,000. See the Country Dataset Details for detailed information for each country.

FACE-TO-FACE SURVEY DESIGN

FIRST STAGE In countries where face-to-face surveys are conducted, the first stage of sampling is the identification of 100 to 135 ultimate clusters (Sampling Units), consisting of clusters of households. Sampling units are stratified by population size and or geography and clustering is achieved through one or more stages of sampling. Where population information is available, sample selection is based on probabilities proportional to population size, otherwise simple random sampling is used. Samples are drawn independent of any samples drawn for surveys conducted in previous years.

There are two methods for sample stratification:

METHOD 1: The sample is stratified into 100 to 125 ultimate clusters drawn proportional to the national population, using the following strata: 1) Areas with population of at least 1 million 2) Areas 500,000-999,999 3) Areas 100,000-499,999 4) Areas 50,000-99,999 5) Areas 10,000-49,999 6) Areas with less than 10,000

The strata could include additional stratum to reflect populations that exceed 1 million as well as areas with populations less than 10,000. Worldwide Research Methodology and Codebook Copyright © 2008-2012 Gallup, Inc. All rights reserved. 8

METHOD 2:

A multi-stage design is used. The country is first stratified by large geographic units, and then by smaller units within geography. A minimum of 33 Primary Sampling Units (PSUs), which are first stage sampling units, are selected. The sample design results in 100 to 125 ultimate clusters.

SECOND STAGE

Random route procedures are used to select sampled households. Unless an outright refusal occurs, interviewers make up to three attempts to survey the sampled household. To increase the probability of contact and completion, attempts are made at different times of the day, and where possible, on different days. If an interviewer cannot obtain an interview at the initial sampled household, he or she uses a simple substitution method. Refer to Appendix C for a more in-depth description of random route procedures.

THIRD STAGE

Respondents are randomly selected within the selected households. Interviewers list all eligible household members and their ages or birthdays. The respondent is selected by means of the Kish grid (refer to Appendix C) in countries where face-to-face interviewing is used. The interview does not inform the person who answers the door of the selection criteria until after the respondent has been identified. In a few Middle East and Asian countries where cultural restrictions dictate gender matching, respondents are randomly selected using the Kish grid from among all eligible adults of the matching gender.

TELEPHONE SURVEY DESIGN

In countries where telephone interviewing is employed, random-digit-dial (RDD) or a nationally representative list of phone numbers is used. In select countries where cell phone penetration is high, a dual sampling frame is used. Random respondent selection is achieved by using either the latest birthday or Kish grid method. At least three attempts are made to reach a person in each household, spread over different days and times of day. Appointments for callbacks that fall within the survey data collection period are made.

PANEL SURVEY DESIGN

Prior to 2009, United States data were collected using The Gallup Panel. The Gallup Panel is a probability-based, nationally representative panel, for which all members are recruited via random-digit-dial methodology and is only used in the United States. Participants who elect to join the panel are committing to the completion of two to three surveys per month, with the typical survey lasting 10 to 15 minutes. The Gallup Worldwide Research panel survey is conducted over the telephone and takes approximately 30 minutes. No incentives are given to panel participants. Worldwide Research Methodology and Codebook Copyright © 2008-2012 Gallup, Inc. All rights reserved. 9

Research instrument

QUESTION DESIGN

Many of the Worldwide Research questions are items that Gallup has used for years. When developing additional questions, Gallup employed its worldwide network of research and political scientists1 to better understand key issues with regard to question development and construction and data gathering. Hundreds of items were developed, tested, piloted, and finalized. The best questions were retained for the core questionnaire and organized into indexes. Most items have a simple dichotomous ("yes or no") response set to minimize contamination of data because of cultural differences in response styles and to facilitate cross-cultural comparisons.

The Gallup Worldwide Research measures key indicators such as Law and Order, Food and Shelter, Job Creation, Migration, Financial Wellbeing, Personal Health, Civic Engagement, and Evaluative Wellbeing and demonstrates their correlations with world development indicators such as GDP and Brain Gain. These indicators assist leaders in understanding the broad context of national interests and establishing organization-specific correlations between leading indexes and lagging economic outcomes.

Gallup organizes its core group of indicators into the Gallup World Path. The Path is an organizational conceptualization of the seven indexes and is not to be construed as a causal model. The individual indexes have many properties of a strong theoretical framework. A more in-depth description of the questions and Gallup indexes is included in the indexes section of this document. In addition to World Path indexes, Gallup Worldwide Research questions also measure opinions about national institutions, corruption, youth development, community basics, diversity, optimism, communications, religiosity, and numerous other topics. For many regions of the world, additional questions that are specific to that region or country are included in surveys. Region-specific questions have been developed for predominantly Muslim nations, former Soviet Union countries, the Balkans, sub-Saharan Africa, Latin America, China and India, South Asia, and Israel and the Palestinian Territories.

The questionnaire is translated into the major conversational languages of each country. The translation process starts with an English, French, or Spanish version, depending on the region. One of two translation methods may be used.

METHOD 1: Two independent translations are completed. An independent third party, with some knowledge of survey research methods, adjudicates the differences. A professional translator translates the final version back into the source language.

METHOD 2: A translator

This dataset explores Percentage of eighth-grade public school students and average scores in NAEP writing by race and state, USA, 2007 Notes: Not available. The state/jurisdiction did not participate. # Rounds to zero. Reporting standards not met. Sample size is insufficient to permit a reliable estimate. NOTE: Black includes African American, Hispanic includes Latino, and Pacifi c Islander includes Native Hawaiian. Race categories exclude Hispanic origin. Results are not shown for students whose race/ethnicity was unclassified Detail may not sum to totals because of rounding. SOURCE: U.S. Department of Education, Institute of Education Sciences, National Center for Education Statistics, National Assessment of Educational Progress (NAEP), 2007 Writing Assessment.

Introduction

This Colombian Spanish Call Center Speech Dataset for the Telecom industry is purpose-built to accelerate the development of speech recognition, spoken language understanding, and conversational AI systems tailored for Spanish-speaking telecom customers. Featuring over 30 hours of real-world, unscripted audio, it delivers authentic customer-agent interactions across key telecom support scenarios to help train robust ASR models.

Curated by FutureBeeAI, this dataset empowers voice AI engineers, telecom automation teams, and NLP researchers to build high-accuracy, production-ready models for telecom-specific use cases.

Speech Data

The dataset contains 30 hours of dual-channel call center recordings between native Colombian Spanish speakers. Captured in realistic customer support settings, these conversations span a wide range of telecom topics from network complaints to billing issues, offering a strong foundation for training and evaluating telecom voice AI solutions.

Topic Diversity

This speech corpus includes both inbound and outbound calls with varied conversational outcomes like positive, negative, and neutral ensuring broad scenario coverage for telecom AI development.

This variety helps train telecom-specific models to manage real-world customer interactions and understand context-specific voice patterns.

Transcription

All audio files are accompanied by manually curated, time-coded verbatim transcriptions in JSON format.

These transcriptions are production-ready, allowing for faster development of ASR and conversational AI systems in the Telecom domain.

Metadata

Rich metadata is available for each participant and conversation:

The GlobalPhone corpus developed in collaboration with the Karlsruhe Institute of Technology (KIT) was designed to provide read speech data for the development and evaluation of large continuous speech recognition systems in the most widespread languages of the world, and to provide a uniform, multilingual speech and text database for language independent and language adaptive speech recognition as well as for language identification tasks. The entire GlobalPhone corpus enables the acquisition of acoustic-phonetic knowledge of the following 22 spoken languages: Arabic (ELRA-S0192), Bulgarian (ELRA-S0319), Chinese-Mandarin (ELRA-S0193), Chinese-Shanghai (ELRA-S0194), Croatian (ELRA-S0195), Czech (ELRA-S0196), French (ELRA-S0197), German (ELRA-S0198), Hausa (ELRA-S0347), Japanese (ELRA-S0199), Korean (ELRA-S0200), Polish (ELRA-S0320), Portuguese (Brazilian) (ELRA-S0201), Russian (ELRA-S0202), Spanish (Latin America) (ELRA-S0203), Swahili (ELRA-S0375), Swedish (ELRA-S0204), Tamil (ELRA-S0205), Thai (ELRA-S0321), Turkish (ELRA-S0206), Ukrainian (ELRA-S0377), and Vietnamese (ELRA-S0322).In each language about 100 sentences were read from each of the 100 speakers. The read texts were selected from national newspapers available via Internet to provide a large vocabulary. The read articles cover national and international political news as well as economic news. The speech is available in 16bit, 16kHz mono quality, recorded with a close-speaking microphone (Sennheiser 440-6). The transcriptions are internally validated and supplemented by special markers for spontaneous effects like stuttering, false starts, and non-verbal effects like laughing and hesitations. Speaker information like age, gender, occupation, etc. as well as information about the recording setup complement the database. The entire GlobalPhone corpus contains over 450 hours of speech spoken by more than 2100 native adult speakers.Data is shortened by means of the shorten program written by Tony Robinson. Alternatively, the data could be delivered unshorten.The Spanish (Latin America) corpus was produced using the La Nacion newspaper. It contains recordings of 100 speakers (44 males, 56 females) recorded in Heredia and San Jose, Costa Rica. The following age distribution has been obtained: 20 speakers are below 19, 54 speakers are between 20 and 29, 13 speakers are between 30 and 39, 5 speakers are between 40 and 49, and 8 speakers are over 50.

Introduction

This Colombian Spanish Call Center Speech Dataset for the Healthcare industry is purpose-built to accelerate the development of Spanish speech recognition, spoken language understanding, and conversational AI systems. With 30 Hours of unscripted, real-world conversations, it delivers the linguistic and contextual depth needed to build high-performance ASR models for medical and wellness-related customer service.

Created by FutureBeeAI, this dataset empowers voice AI teams, NLP researchers, and data scientists to develop domain-specific models for hospitals, clinics, insurance providers, and telemedicine platforms.

Speech Data

The dataset features 30 Hours of dual-channel call center conversations between native Colombian Spanish speakers. These recordings cover a variety of healthcare support topics, enabling the development of speech technologies that are contextually aware and linguistically rich.

Topic Diversity

The dataset spans inbound and outbound calls, capturing a broad range of healthcare-specific interactions and sentiment types (positive, neutral, negative).

These real-world interactions help build speech models that understand healthcare domain nuances and user intent.

Transcription

Every audio file is accompanied by high-quality, manually created transcriptions in JSON format.

Metadata

Each conversation and speaker includes detailed metadata to support fine-tuned training and analysis.

Usage and Applications

This dataset can be used across a range of healthcare and voice AI use cases:

Hokkien (a.k.a. Minnan 閩南話, Southern Min, Taiwanese) is a variety of Chinese spoken in the southern part of Fujian province (China), Taiwan and by a large number of overseas Chinese all over Southeast Asia. Hokkien dictionaries and textbooks have been published since the 16th century in a number of locations and in a variety of languages (Classical Chinese, Dutch, English, Hokkien, Japanese, Latin, Mandarin, Spanish) for purposes such as education for locals, Christian missionary work and colonial administration.

In this post I would like to share a database project I created for the "Working with Digital Data for Historians" class taught by Prof. Tara Andrews at Uni Vienna in the autumn semester of 2018/19. The project work is based on my 3-years MA studies at Xiamen University (Fujian prov., China) and is connected to an article of mine (Xiamen at the Crossroads of Sino-Foreign Interaction During the Late Qing and Republican Periods: The Issue of Hokkien Phoneticization) already peer-reviewed and accepted for publication in the journal Crossroads - Studies on the History of Exchange Relations in the East Asian World. The database is an Excel spreadsheet-based SQL database containing 121 titles of Hokkien linguistic works, connected to relevant information (author, place/date of publishing, publisher, language, transcription method, etc.).

Hereby, I submit the dump version of the database (a .sql file containing the SQL scheme and the data as well) and a zip collection containing further relevant materials (project description, original spreadsheets, SQL command scheme, .ipynb documentation of converting the Excel spreadsheets into the SQL database using Jupyter Notebook).

Since during my research work I focused on the pre-WWII period, data regarding the post-WWII is highly insufficient. Therefore, the database is up for further expansion. (My e-mail address: sebestyen.hompot@outlook.com)

This Dataset shows some basic demographic data from the US census located around the San Francisco MSA at tract level. Attributes include Average age, female and male population, white population, hispanic population, population density, and total population.

Introduction

This Mexican Spanish Call Center Speech Dataset for the Retail and E-commerce industry is purpose-built to accelerate the development of speech recognition, spoken language understanding, and conversational AI systems tailored for Spanish speakers. Featuring over 30 hours of real-world, unscripted audio, it provides authentic human-to-human customer service conversations vital for training robust ASR models.

Curated by FutureBeeAI, this dataset empowers voice AI developers, data scientists, and language model researchers to build high-accuracy, production-ready models across retail-focused use cases.

Speech Data

The dataset contains 30 hours of dual-channel call center recordings between native Mexican Spanish speakers. Captured in realistic scenarios, these conversations span diverse retail topics from product inquiries to order cancellations, providing a wide context range for model training and testing.

Topic Diversity

This speech corpus includes both inbound and outbound calls with varied conversational outcomes like positive, negative, and neutral, ensuring real-world scenario coverage.

Such variety enhances your model’s ability to generalize across retail-specific voice interactions.

Transcription

All audio files are accompanied by manually curated, time-coded verbatim transcriptions in JSON format.

These transcriptions are production-ready, making model training faster and more accurate.

Metadata

Rich metadata is available for each participant and conversation:

This granularity supports advanced analytics, dialect filtering, and fine-tuned model evaluation.

Usage and Applications

This dataset is ideal for a range of voice AI and NLP applications:

Introduction

This Mexican Spanish Call Center Speech Dataset for the Delivery and Logistics industry is purpose-built to accelerate the development of speech recognition, spoken language understanding, and conversational AI systems tailored for Spanish-speaking customers. With over 30 hours of real-world, unscripted call center audio, this dataset captures authentic delivery-related conversations essential for training high-performance ASR models.

Curated by FutureBeeAI, this dataset empowers AI teams, logistics tech providers, and NLP researchers to build accurate, production-ready models for customer support automation in delivery and logistics.

Speech Data

The dataset contains 30 hours of dual-channel call center recordings between native Mexican Spanish speakers. Captured across various delivery and logistics service scenarios, these conversations cover everything from order tracking to missed delivery resolutions offering a rich, real-world training base for AI models.

Topic Diversity

This speech corpus includes both inbound and outbound delivery-related conversations, covering varied outcomes (positive, negative, neutral) to train adaptable voice models.

This comprehensive coverage reflects real-world logistics workflows, helping voice AI systems interpret context and intent with precision.

Transcription

All recordings come with high-quality, human-generated verbatim transcriptions in JSON format.

These transcriptions support fast, reliable model development for Spanish voice AI applications in the delivery sector.

Metadata

Detailed metadata is included for each participant and conversation:

This metadata aids in training specialized models, filtering demographics, and running advanced analytics.

Usage and Applications