#SaudiArabia #SpeechAI - Saudi Arabia’s national AI champion HUMAIN has released HUMAIN Voice, an Arabic-first speech platform combining speech recognition, text-to-speech and voice agents. the AI cloud platform supports Arabic language dialects such as Najdi, Hijazi, Khaleeji, English and Arabic-English code-switching. HUMAIN has also announced that it plans to release an open-weight, 80-million-parameter speech model under the Apache 2.0 licence.
SO WHAT? - Speech AI has mostly been built around English, with Arabic added later. However, there is a new generation of Arabic-first speech AI being built by developers across the region. The ability to handle Arabic dialects is absolutely critical for speech AI, since spoken Arabic differs greatly from the formal writtent Modern Standard Arabic (MSA). HUMAIN has tacked the issue of Arabic dialects as a priority and has also made the technology available as a small open model that runs on ordinary CPUs. This both makes the HUMAIN model cost effective to deploy and also is likely to enable feedback from a wider pool of developers and users, whcih could inform future versions of HUMAIN Voice.
KEY POINTS:
HUMAIN has launched HUMAIN Voice, an Arabic-first speech platform combining speech recognition, text-to-speech and voice agents.
HUMAIN Voice brings together automatic speech recognition (ASR), text-to-speech (TTS) and voice agents on one platform, and is built for customer service, enterprise operations and other applications that depend on natural conversation.
The platform supports Saudi dialects including Najdi and Hijazi, plus Khaleeji and other Arabic dialects, English and natural Arabic-English code-switching. HUMAIN says these are core design features, not edge cases.
HUMAIN Voice agents are full-duplex, speech-to-speech systems that listen, understand and reply in the caller’s own dialect with low latency.
HUMAIN-Voice-Lite-80M is a transducer-based ASR model with 80 million parameters, covering Arabic, English and Arabic-English code-switching. It is released as open weights under the Apache 2.0 licence.
The lightweight model is optimised for both CPU and GPU environments, so it can serve uses where cost, latency, infrastructure availability or on-device and edge deployment matter.
HUMAIN will publish benchmark results from the Open Universal Arabic ASR Leaderboard with the model, plus a blog showing how it handles Arabic, English and code-switching across sample speech scenarios.
HUMAIN says the open release is meant to let developers, researchers and enterprises build on and adapt the technology, and to speed up Arabic speech AI development.
Voice extends HUMAIN’s full-stack portfolio from AI infrastructure and models into the interfaces people use to interact with AI. The company says the HUMAIN Voice API is available to developers and already in use.
[Written and edited with the assistance of AI]
Source: HUMAIN
LINK
HUMAIN Voice cloud service (HUMAIN)
Read more HUMAIN news:
HUMAIN to launch 3 trillion parameter model (Middle East AI News)
HUMAIN readies Limitless startup accelerator for launch (Middle East AI News)
Infra, HUMAIN sign SAR 2.5 billion AI financing deals (Middle East AI News)
HUMAIN takes over Aramco-backed cloud firm CNTXT (Middle East AI News)
Aramco completes 49% acquisition of HUMAIN (Middle East AI News)


