Urdu (Pakistan) Scripted Monologue Speech Dataset for Retail & E-commerce Domain

The audio dataset comprises scripted monologue speech data in Retail & E-commerce domain, featuring native Urdu speakers from Pakistan. It includes speech data, detailed metadata, and accurate transcriptions.

About this Off-the-shelf Speech Dataset

Introduction

Welcome to the Urdu Scripted Monologue Speech Dataset for the Retail & E-commerce Domain. This meticulously curated dataset is designed to advance the development of Urdu language speech recognition models, particularly for the Retail & E-commerce industry.

Speech Data

This training dataset comprises over 6,000 high-quality scripted prompt recordings in Urdu. These recordings cover various topics and scenarios relevant to the Retail & E-commerce domain, designed to build robust and accurate customer service speech technology.

•Participant Diversity:

•

Speakers: 60 native Urdu speakers from different regions of Pakistan.

•

Regions: Ensures a balanced representation of Urdu accents, dialects, and demographics.

•

Participant Profile: Participants range from 18 to 70 years old, representing both males and females in a 60:40 ratio, respectively.

•Recording Details:

•

Recording Nature: Audio recordings of scripted prompts/monologues.

•

Audio Duration: Average duration of 5 to 30 seconds per recording.

•

Formats: WAV format with mono channels, a bit depth of 16 bits, and sample rates of 8 kHz and 16 kHz.

•

Environment: Recordings are conducted in quiet settings without background noise and echo.

•

Topic Diversity: The dataset encompasses a wide array of topics and conversational scenarios to ensure comprehensive coverage of the Retail & E-commerce sector. Topics include:

•Customer Service Interactions

•Order and Payment Processes

•Product and Service Inquiries

•Technical Support

•General Information and Advice

•Promotional and Sales Events

•Domain Specific Statements

•

Other Elements: To enhance realism and utility, the scripted prompts incorporate various elements commonly encountered in Retail & E-commerce interactions:

•

Names: Region-specific names of males and females in various formats.

•

Addresses: Region-specific addresses in different spoken formats.

•

Dates & Times: Inclusion of date and time in various retail and e-commerce contexts, such as delivery dates or promotional periods.

•

Product Names: Specific names of products, brands, and categories relevant to the retail sector.

•

Numbers & Prices: Various numbers and prices related to product quantities, discounts, and transaction amounts.

•

Order IDs and Tracking Numbers: Inclusion of order identification and tracking information for realistic customer service scenarios.

Each scripted prompt is crafted to reflect real-life scenarios encountered in the Retail & E-commerce domain, ensuring applicability in training robust natural language processing and speech recognition models.

Transcription Data

In addition to high-quality audio recordings, the dataset includes meticulously prepared text files with verbatim transcriptions of each audio file. These transcriptions are essential for training accurate and robust speech recognition models.

•

Content: Each text file contains the exact scripted prompt corresponding to its audio file, ensuring consistency.

•

Format: Transcriptions are provided in plain text (.TXT) format, with files named to match their associated audio files for easy reference.

•

Quality: All transcriptions are verified for accuracy and consistency by native Urdu transcribers.

Metadata

The dataset provides comprehensive metadata for each audio recording and participant:

•

Participant Metadata: Unique identifier, age, gender, country, state, and dialect.

•

Other Metadata: Recording transcript, recording environment, device details, sample rate, bit depth, and file format.

This metadata is a powerful tool for understanding and characterizing the data, enabling informed decision-making in the development of Urdu language speech recognition models.

Usage and Applications

This dataset is a versatile resource for various applications within speech recognition, natural language processing, and AI-driven conversational technologies.

•

Speech Recognition Model Training: High-quality audio recordings and precise transcriptions for training and fine-tuning Urdu speech recognition models.

•

Voice Synthesis: The diverse and high-quality audio data can train generative AI models for creating synthetic voices.

•

Voice Assistants: Ideal for training voice assistants tailored to the Retail & E-commerce domain.

•

Chatbots: Transcription data can train conversational models, enabling chatbots to respond to customer queries effectively.

•

Entity Recognition: Sentences include names, dates, currencies, and other domain-specific entities for training NLP models for named entity recognition (NER) tasks.

•

Language Understanding: Improve language understanding applications like sentiment analysis and topic modeling within the Retail & E-commerce sector.