A Pocket Offline Model for Simultaneous Speech Translation as CUNI Submission to IWSLT 2026

2026-06-29 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, quick

Summary

The CUNI submission to the IWSLT 2026 Simultaneous Speech Translation Shared task presents a direct speech translation model named Canary, specifically engineered for simultaneous translation. This model integrates the AlignAtt simultaneous policy and is built within the Nemo toolkit. A core component is the Canary-1B-v2 foundation model, notable for its compact size with only one billion parameters, which makes it highly suitable for deployment on small, pocket-sized devices. The initiative targets simultaneous translation for language pairs including Czech to English, and English to German and Italian, showcasing a practical and efficient solution for real-time speech translation in resource-constrained environments.

Key takeaway

For NLP engineers developing real-time speech translation for mobile or edge devices, this work demonstrates a viable path. You should consider the Nemo toolkit and compact models like Canary-1B-v2 to achieve simultaneous translation on resource-constrained hardware. This approach allows for efficient, direct speech translation, enabling new applications where offline capability and low latency are critical. Evaluate the AlignAtt policy for your simultaneous translation needs.

Key insights

A 1-billion-parameter model enables simultaneous speech translation on pocket devices using Nemo toolkit and AlignAtt policy.

Principles

• Direct speech translation is viable.
• Small models enable device deployment.
• Simultaneous policy enhances real-time.

Method

The approach implements the Canary direct speech translation model with the AlignAtt simultaneous policy, leveraging the Nemo toolkit and the compact Canary-1B-v2 foundation model.

In practice

• Deploy speech translation on mobile.
• Use Canary-1B-v2 for efficiency.
• Apply AlignAtt for real-time.

Topics

• Simultaneous Speech Translation
• Canary-1B-v2
• Nemo Toolkit
• AlignAtt Policy
• Pocket Devices
• IWSLT 2026

Best for: AI Engineer, Research Scientist, AI Scientist, NLP Engineer, Machine Learning Engineer

Related on AIssential

• Kyutai Releases Hibiki-Zero: A3B Parameter Simultaneous Speech-to-Speech Translation Model Using GRPO Reinforcement Learning Without Any Word-Level Aligned Data — Hibiki-Zero enables simultaneous speech translation without word-level alignments using GRPO reinforcement learning.
• AlignAtt4LLM: Fast AlignAtt for Decoder-Only LLMs at IWSLT 2026 Simultaneous Speech Translation Task — Adapting AlignAtt for decoder-only LLMs enables simultaneous speech translation by re-engineering attention mechanisms.
• Celebrating 20 years of Google Translate: Fun facts, tips and new features to try — Google Translate's 20-year evolution highlights AI's role in breaking language barriers and fostering global connection.
• Streaming Speech-to-Text Translation with a SpeechLLM — A novel SpeechLLM adaptively decides when to emit translation tokens, achieving low latency and high quality.
• Granite 4.0 1B Speech: Compact, Multilingual, and Built for the Edge — Granite 4.0 1B Speech offers high accuracy and multilingual support in a compact, efficient model for edge devices.
• NaturalFlow: Reducing Disruptive Pauses for Natural Speech Flow in Simultaneous Speech-to-Speech Translation — Simultaneous S2ST can achieve natural speech flow by optimizing for fluency and translation quality concurrently, avoiding disruptive pauses.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.