---
title: "Josh Talks models: release history and timeline"
description: "Gurugram-based Indian-language media and learning company whose AI unit builds conversational speech datasets, benchmarks and speech models."
type: "release-archive"
canonical_url: "https://theaiengineering.com/releases/josh-talks"
markdown_url: "https://theaiengineering.com/releases/josh-talks.md"
author: "The AI Engineering Editors"
publisher: "The AI Engineering"
updated: "2026-03-09"
citation: "The AI Engineering Editors. \"Josh Talks models: release history and timeline.\" The AI Engineering, updated Mar 09, 2026. https://theaiengineering.com/releases/josh-talks"
license: "CC BY 4.0"
license_url: "https://creativecommons.org/licenses/by/4.0/"
data_csv: "https://theaiengineering.com/releases.csv"
data_json: "https://theaiengineering.com/releases.json"
---

# Josh Talks models: release history and timeline

Gurugram-based Indian-language media and learning company whose AI unit builds conversational speech datasets, benchmarks and speech models.

- Headquarters: Gurugram, India
- Website: https://www.joshtalks.com

1 release from 1 lab, 1 with open weights. Dates are the labs' official announcement dates.

## March 2026

### Human-1

Human-1 is a full-duplex spoken dialogue model for Hindi that listens and speaks at the same time, built by adapting Kyutai's Moshi architecture and retraining it on real Hindi conversations.

- Date: 2026-03-09
- Lab: Josh Talks
- Type: Speech
- Weights: Open weights
- Modalities: audio → audio
- Parameters: 7.7B
- License: CC-BY-4.0
- Available at launch: Open weights
- Permalink: https://theaiengineering.com/releases/2026/03#josh-talks-human-1

Highlights:

- Josh Talks trained it on 26,000 hours of spontaneous two-speaker Hindi conversations from 14,695 speakers, recorded with a separate channel per speaker
- The team swapped Moshi's English tokenizer for a 32,000-token Hindi SentencePiece model and kept the Mimi audio codec frozen
- Josh Talks presented it as a research proof point rather than a production system, and published a paper on arXiv on April 25, 2026
- Weights were released on Hugging Face under CC BY 4.0

Source: [Human-1: A Full-Duplex Conversational Speech Model for Hindi](https://ai.joshtalks.com/research/human-1)

## Cite and reuse this data

These release records are free to reuse under CC BY 4.0, including commercially. Credit The AI Engineering and link to this page's URL. Dates are the labs' official announcement dates, and every record links to its source. List fields in the CSV are separated by " | ".

The AI Engineering Editors. "Josh Talks models: release history and timeline." The AI Engineering, updated Mar 09, 2026. https://theaiengineering.com/releases/josh-talks

HTML link:

```html
<a href="https://theaiengineering.com/releases/josh-talks">Josh Talks models: release history and timeline</a> (The AI Engineering)
```

- Title: Josh Talks models: release history and timeline
- Author: The AI Engineering Editors
- Publisher: The AI Engineering
- Updated: 2026-03-09
- URL: https://theaiengineering.com/releases/josh-talks
- Markdown: https://theaiengineering.com/releases/josh-talks.md
- License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)
- Full dataset as CSV: https://theaiengineering.com/releases.csv
- Full dataset as JSON: https://theaiengineering.com/releases.json
