Gold Standard Data For ASR Research
We are on a mission to reduce WER to < 5%

Gold Standard Data For ASR Research
We are on a mission to reduce WER to < 5%

We are an audio data research lab dedicated to advancing speech technology. We engineer multilingual speech datasets, gold test sets, and evaluations that advance global speech recognition.

The Josh Talks Data Engine powers leading ML teams with wide linguistic coverage, scale, and research-grade quality.

Why choose Josh Talks?

sheidl

Grass-roots Diversity

Voices, faces, documents and street scenes from every state, age group and socioeconomic tier—captured exactly where AI products are used.

sheidl

Measurable Quality

Dual human-in-the-loop review + anomaly detection keeps mean label error < 2 % across all modalities.

sheidl

Global Multilingual Coverage

17 languages of two-channel (speaker-separated) conversations, rigorously curated across tens of thousands of topics.

sheidl

Ethical Sourcing

GDPR-ready consent flows, automated PII redaction in audio and imagery, plus community revenue-share for every contributor.

sheidl

Enterprise-grade Security

Air-gapped annotation rooms, ISO-27001 cloud tenancy and per-file audit trails built for Big-Tech compliance teams.

Data Scarcity to Abundance

Our patented means of data production and annotation allows us to generate and label 10 Million Hours of voice data every year

Channel Separated Conversational Audio Datasets

Train your AI models using data from the grassroots of India through Josh Talks’ network of Training Data Specialists

trans

Multi-lingual Datasets

check

Multilingual and code-switching Voice Datasets

check

Low Resource, Rare Language and Dialect Voice Datasets

check

Accented English Speech Dataset

check

Voice Datasets in - English, Hindi, Tamil, Marathi, Telugu, Bangla, Kannada, Malayalam, Punjabi, Oriya, Gujarati, Assamese

trans

Multi-Speaker Datasets

check

Multiple Speaker spontaneous conversations

check

Diarized speaker stems for up to 16 speakers

check

Multi-Speaker debate datasets

trans

Emotion Rich Data

check

Emotionally Aware and Annotated Conversations

check

Datasets covering 9 emotions - Neutral, Angry, Happy, Sad, Fear, Anxious, Surprised, Confused, Excited

trans

Other Datasets

check

Noisy and Adverse Environment Datasets

check

Privacy Preserving Highly Personalized Voice Datasets

check

Voice Datasets for Accessibility

check

Child Speech Datasets

Two Channel Separated Conversational Audio

Voice Datasets in - English, Hindi, Tamil, Marathi, Telugu, Bangla, Kannada, Malayalam, Punjabi, Oriya, Gujarati, Assamese, Japanese, Korean, US English, Indonesian, Arabic and other languages.

Customers across our portfolio of products